path: root/doc/rfc/rfc9275.txt

Internet Engineering Task Force (IETF)                            K. Gao
Request for Comments: 9275                            Sichuan University
Category: Experimental                                            Y. Lee
ISSN: 2070-1721                                                  Samsung
                                                          S. Randriamasy
                                                         Nokia Bell Labs
                                                                 Y. Yang
                                                         Yale University
                                                                J. Zhang
                                                       Tongji University
                                                          September 2022


    An Extension for Application-Layer Traffic Optimization (ALTO):
                              Path Vector

Abstract

   This document is an extension to the base Application-Layer Traffic
   Optimization (ALTO) protocol.  It extends the ALTO cost map and ALTO
   property map services so that an application can decide to which
   endpoint(s) to connect based not only on numerical/ordinal cost
   values but also on fine-grained abstract information regarding the
   paths.  This is useful for applications whose performance is impacted
   by specific components of a network on the end-to-end paths, e.g.,
   they may infer that several paths share common links and prevent
   traffic bottlenecks by avoiding such paths.  This extension
   introduces a new abstraction called the "Abstract Network Element"
   (ANE) to represent these components and encodes a network path as a
   vector of ANEs.  Thus, it provides a more complete but still abstract
   graph representation of the underlying network(s) for informed
   traffic optimization among endpoints.

Status of This Memo

   This document is not an Internet Standards Track specification; it is
   published for examination, experimental implementation, and
   evaluation.

   This document defines an Experimental Protocol for the Internet
   community.  This document is a product of the Internet Engineering
   Task Force (IETF).  It represents the consensus of the IETF
   community.  It has received public review and has been approved for
   publication by the Internet Engineering Steering Group (IESG).  Not
   all documents approved by the IESG are candidates for any level of
   Internet Standard; see Section 2 of RFC 7841.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at
   https://www.rfc-editor.org/info/rfc9275.

Copyright Notice

   Copyright (c) 2022 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents
   (https://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document must
   include Revised BSD License text as described in Section 4.e of the
   Trust Legal Provisions and are provided without warranty as described
   in the Revised BSD License.

Table of Contents

   1.  Introduction
   2.  Requirements Language
   3.  Terminology
   4.  Requirements and Use Cases
     4.1.  Design Requirements
     4.2.  Sample Use Cases
       4.2.1.  Exposing Network Bottlenecks
       4.2.2.  Resource Exposure for CDNs and Service Edges
   5.  Path Vector Extension: Overview
     5.1.  Abstract Network Element (ANE)
       5.1.1.  ANE Entity Domain
       5.1.2.  Ephemeral and Persistent ANEs
       5.1.3.  Property Filtering
     5.2.  Path Vector Cost Type
     5.3.  Multipart Path Vector Response
       5.3.1.  Identifying the Media Type of the Object Root
       5.3.2.  References to Part Messages
   6.  Specification: Basic Data Types
     6.1.  ANE Name
     6.2.  ANE Entity Domain
       6.2.1.  Entity Domain Type
       6.2.2.  Domain-Specific Entity Identifier
       6.2.3.  Hierarchy and Inheritance
       6.2.4.  Media Type of Defining Resource
     6.3.  ANE Property Name
     6.4.  Initial ANE Property Types
       6.4.1.  Maximum Reservable Bandwidth
       6.4.2.  Persistent Entity ID
       6.4.3.  Examples
     6.5.  Path Vector Cost Type
       6.5.1.  Cost Metric: "ane-path"
       6.5.2.  Cost Mode: "array"
     6.6.  Part Resource ID and Part Content ID
   7.  Specification: Service Extensions
     7.1.  Notation
     7.2.  Multipart Filtered Cost Map for Path Vector
       7.2.1.  Media Type
       7.2.2.  HTTP Method
       7.2.3.  Accept Input Parameters
       7.2.4.  Capabilities
       7.2.5.  Uses
       7.2.6.  Response
     7.3.  Multipart Endpoint Cost Service for Path Vector
       7.3.1.  Media Type
       7.3.2.  HTTP Method
       7.3.3.  Accept Input Parameters
       7.3.4.  Capabilities
       7.3.5.  Uses
       7.3.6.  Response
   8.  Examples
     8.1.  Sample Setup
     8.2.  Information Resource Directory
     8.3.  Multipart Filtered Cost Map
     8.4.  Multipart Endpoint Cost Service Resource
     8.5.  Incremental Updates
     8.6.  Multi-Cost
   9.  Compatibility with Other ALTO Extensions
     9.1.  Compatibility with Legacy ALTO Clients/Servers
     9.2.  Compatibility with Multi-Cost Extension
     9.3.  Compatibility with Incremental Update Extension
     9.4.  Compatibility with Cost Calendar Extension
   10. General Discussion
     10.1.  Constraint Tests for General Cost Types
     10.2.  General Multi-Resource Query
   11. Security Considerations
   12. IANA Considerations
     12.1.  "ALTO Cost Metrics" Registry
     12.2.  "ALTO Cost Modes" Registry
     12.3.  "ALTO Entity Domain Types" Registry
     12.4.  "ALTO Entity Property Types" Registry
       12.4.1.  New ANE Property Type: Maximum Reservable Bandwidth
       12.4.2.  New ANE Property Type: Persistent Entity ID
   13. References
     13.1.  Normative References
     13.2.  Informative References
   Acknowledgments
   Authors' Addresses

1.  Introduction

   Network performance metrics are crucial for assessing the Quality of
   Experience (QoE) of applications.  The Application-Layer Traffic
   Optimization (ALTO) protocol allows Internet Service Providers (ISPs)
   to provide guidance, such as topological distances between different
   end hosts, to overlay applications.  Thus, the overlay applications
   can potentially improve the perceived QoE by better orchestrating
   their traffic to utilize the resources in the underlying network
   infrastructure.

   The existing ALTO cost map (Section 11.2.3 of [RFC7285]) and Endpoint
   Cost Service (Section 11.5 of [RFC7285]) provide only cost
   information for an end-to-end path defined by its <source,
   destination> endpoints: the base protocol [RFC7285] allows the
   services to expose the topological distances of end-to-end paths,
   while various extensions have been proposed to extend the capability
   of these services, e.g., to express other performance metrics
   [ALTO-PERF-METRICS], to query multiple costs simultaneously
   [RFC8189], and to obtain time-varying values [RFC8896].

   While numerical/ordinal cost values for end-to-end paths provided by
   the existing extensions are sufficient to optimize the QoE of many
   overlay applications, the QoE of some overlay applications also
   depends on the properties of particular components on the paths.  For
   example, job completion time, which is an important QoE metric for a
   large-scale data analytics application, is impacted by shared
   bottleneck links inside the carrier network, as link capacity may
   impact the rate of data input/output to the job.  We refer to such
   components of a network as Abstract Network Elements (ANEs).

   Predicting such information can be very complex without the help of
   ISPs; for example, [BOXOPT] has shown that finding the optimal
   bandwidth reservation for multiple flows can be NP-hard without
   further information than whether a reservation succeeds.  With proper
   guidance from the ISP, an overlay application may be able to schedule
   its traffic for better QoE.  In the meantime, it may be helpful as
   well for ISPs if applications could avoid using bottlenecks or
   challenging the network with poorly scheduled traffic.

   Despite the claimed benefits, ISPs are not likely to expose raw
   details on their network paths: first because ISPs have requirements
   to hide their network topologies, second because these details may
   increase volume and computation overhead, and last because
   applications do not necessarily need all the network path details and
   are likely not able to understand them.

   Therefore, it is beneficial for both ISPs and applications if an ALTO
   server provides ALTO clients with an "abstract network state" that
   provides the necessary information to applications, while hiding
   network complexity and confidential information.  An "abstract
   network state" is a selected set of abstract representations of ANEs
   traversed by the paths between <source, destination> pairs combined
   with properties of these ANEs that are relevant to the overlay
   applications' QoE.  Both an application via its ALTO client and the
   ISP via the ALTO server can achieve better confidentiality and
   resource utilization by appropriately abstracting relevant ANEs.
   Server scalability can also be improved by combining ANEs and their
   properties in a single response.

   This document extends the ALTO base protocol [RFC7285] to allow an
   ALTO server to convey "abstract network state" for paths defined by
   their <source, destination> pairs.  To this end, it introduces a new
   cost type called a "Path Vector", following the cost metric
   registration specified in [RFC7285] and the updated cost mode
   registration specified in [RFC9274].  A Path Vector is an array of
   identifiers that identifies an ANE, which can be associated with
   various properties.  The associations between ANEs and their
   properties are encoded in an ALTO information resource called the
   "entity property map", which is specified in [RFC9240].

   For better confidentiality, this document aims to minimize
   information exposure of an ALTO server when providing Path Vector
   services.  In particular, this document enables the capability, and
   also recommends that 1) ANEs be constructed on demand and 2) an ANE
   only be associated with properties that are requested by an ALTO
   client.  A Path Vector response involves two ALTO maps: the cost map,
   which contains the Path Vector results; and the up-to-date entity
   property map, which contains the properties requested for these ANEs.
   To enforce consistency and improve server scalability, this document
   uses the "multipart/related" content type as defined in [RFC2387] to
   return the two maps in a single response.

   As a single ISP may not have knowledge of the full Internet paths
   between arbitrary endpoints, this document is mainly applicable when

   *  there is a single ISP between the requested source and destination
      Provider-defined Identifiers (PIDs) or endpoints -- for example,
      ISP-hosted Content Delivery Network (CDN) / edge, tenant
      interconnection in a single public cloud platform, etc., or

   *  the Path Vectors are generated from end-to-end measurement data.

2.  Requirements Language

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
   "OPTIONAL" in this document are to be interpreted as described in
   BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all
   capitals, as shown here.

3.  Terminology

   This document extends the ALTO base protocol [RFC7285] and the entity
   property map extension [RFC9240].  In addition to the terms defined
   in those documents, this document also uses the following terms:

   Abstract Network Element (ANE):  An abstract representation for a
      component in a network that handles data packets and whose
      properties can potentially have an impact on the end-to-end
      performance of traffic.  An ANE can be a physical device such as a
      router, a link, or an interface; or an aggregation of devices such
      as a subnetwork or a data center.

      The definition of an ANE is similar to that for a network element
      as defined in [RFC2216] in the sense that they both provide an
      abstract representation of specific components of a network.
      However, they have different criteria on how these particular
      components are selected.  Specifically, a network element requires
      the components to be capable of exercising QoS control, while an
      ANE only requires the components to have an impact on end-to-end
      performance.

   ANE name:  A string that uniquely identifies an ANE in a specific
      scope.  An ANE can be constructed either statically in advance or
      on demand based on the requested information.  Thus, different
      ANEs may only be valid within a particular scope, either ephemeral
      or persistent.  Within each scope, an ANE is uniquely identified
      by an ANE name, as defined in Section 6.1.  Note that an ALTO
      client must not assume ANEs in different scopes but with the same
      ANE name refer to the same component(s) of the network.

   Path Vector (or ANE Path Vector):  Refers to a JSON array of ANE
      names.  It is a generalization of a BGP path vector.  While a
      standard BGP path vector (Section 5.1.2 of [RFC4271]) specifies a
      sequence of Autonomous Systems (ASes) for a destination IP prefix,
      the Path Vector defined in this extension specifies a sequence of
      ANEs for either 1) a source PID and a destination PID, as in the
      CostMapData object (Section 11.2.3.6 of [RFC7285]) or 2) a source
      endpoint and a destination endpoint, as in the EndpointCostMapData
      object (Section 11.5.1.6 of [RFC7285]).

   Path Vector resource:  An ALTO information resource (Section 8.1 of
      [RFC7285]) that supports the extension defined in this document.

   Path Vector cost type:  A special cost type, which is specified in
      Section 6.5.  When this cost type is present in an Information
      Resource Directory (IRD) entry, it indicates that the information
      resource is a Path Vector resource.  When this cost type is
      present in a filtered cost map request or an Endpoint Cost Service
      request, it indicates that each cost value must be interpreted as
      a Path Vector.

   Path Vector request:  The POST message sent to an ALTO Path Vector
      resource.

   Path Vector response:  Refers to the multipart/related message
      returned by a Path Vector resource.

4.  Requirements and Use Cases

4.1.  Design Requirements

   This section gives an illustrative example of how an overlay
   application can benefit from the extension defined in this document.

   Assume that an application has control over a set of flows, which may
   go through shared links/nodes and share bottlenecks.  The application
   seeks to schedule the traffic among multiple flows to get better
   performance.  The constraints of feasible rate allocations of those
   flows will benefit the scheduling.  However, cost maps as defined in
   [RFC7285] cannot reveal such information.

   Specifically, consider the example network shown in Figure 1.  The
   network has seven switches ("sw1" to "sw7") forming a dumbbell
   topology.  Switches "sw1", "sw2", "sw3", and "sw4" are access
   switches, and "sw5-sw7" form the backbone.  End hosts "eh1" to "eh4"
   are connected to access switches "sw1" to "sw4", respectively.
   Assume that the bandwidth of link "eh1 -> sw1" and link "sw1 -> sw5"
   is 150 Mbps and the bandwidth of the other links is 100 Mbps.

                                 +-----+
                                 |     |
                               --+ sw6 +--
                              /  |     |  \
        PID1 +-----+         /   +-----+   \          +-----+  PID2
        eh1__|     |_       /               \     ____|     |__eh2
   192.0.2.2 | sw1 | \   +--|--+         +--|--+ /    | sw2 | 192.0.2.3
             +-----+  \  |     |         |     |/     +-----+
                       \_| sw5 +---------+ sw7 |
        PID3 +-----+   / |     |         |     |\     +-----+  PID4
        eh3__|     |__/  +-----+         +-----+ \____|     |__eh4
   192.0.2.4 | sw3 |                                  | sw4 | 192.0.2.5
             +-----+                                  +-----+

   bw(eh1--sw1) = bw(sw1--sw5) = 150 Mbps
   bw(eh2--sw2) = bw(eh3--sw3) = bw(eh4--sw4) = 100 Mbps
   bw(sw1--sw5) = bw(sw3--sw5) = bw(sw2--sw7) = bw(sw4--sw7) = 100 Mbps
   bw(sw5--sw6) = bw(sw5--sw7) = bw(sw6--sw7) = 100 Mbps

                       Figure 1: Raw Network Topology

   The base ALTO topology abstraction of the network is shown in
   Figure 2.  Assume that the cost map returns a hypothetical cost type
   representing the available bandwidth between a source and a
   destination.

                             +----------------------+
                    {eh1}    |                      |     {eh2}
                    PID1     |                      |     PID2
                      +------+                      +------+
                             |                      |
                             |                      |
                    {eh3}    |                      |     {eh4}
                    PID3     |                      |     PID4
                      +------+                      +------+
                             |                      |
                             +----------------------+

                    Figure 2: Base Topology Abstraction

   Now, assume that the application wants to maximize the total rate of
   the traffic among a set of <source, destination> pairs -- say, "eh1
   -> eh2" and "eh1 -> eh4".  Let "x" denote the transmission rate of
   "eh1 -> eh2" and "y" denote the rate of "eh1 -> eh4".  The objective
   function is

       max(x + y).

   With the ALTO cost map, the costs between PID1 and PID2 and between
   PID1 and PID4 will both be 100 Mbps.  The client can get a capacity
   region of

       x <= 100 Mbps
       y <= 100 Mbps.

   With this information, the client may mistakenly think it can achieve
   a maximum total rate of 200 Mbps.  However, this rate is infeasible,
   as there are only two potential cases:

   Case 1:  "eh1 -> eh2" and "eh1 -> eh4" take different path segments
      from "sw5" to "sw7".  For example, if "eh1 -> eh2" uses path "eh1
      -> sw1 -> sw5 -> sw6 -> sw7 -> sw2 -> eh2" and "eh1 -> eh4" uses
      path "eh1 -> sw1 -> sw5 -> sw7 -> sw4 -> eh4", then the shared
      bottleneck links are "eh1 -> sw1" and "sw1 -> sw5".  In this case,
      the capacity region is:

          x     <= 100 Mbps
          y     <= 100 Mbps
          x + y <= 150 Mbps

      and the real optimal total rate is 150 Mbps.

   Case 2:  "eh1 -> eh2" and "eh1 -> eh4" take the same path segment
      from "sw5" to "sw7".  For example, if "eh1 -> eh2" uses path "eh1
      -> sw1 -> sw5 -> sw7 -> sw2 -> eh2" and "eh1 -> eh4" also uses
      path "eh1 -> sw1 -> sw5 -> sw7 -> sw4 -> eh4", then the shared
      bottleneck link is "sw5 -> sw7".  In this case, the capacity
      region is:

          x     <= 100 Mbps
          y     <= 100 Mbps
          x + y <= 100 Mbps

      and the real optimal total rate is 100 Mbps.

   Clearly, with more accurate and fine-grained information, the
   application can better predict its traffic and may orchestrate its
   resources accordingly.  However, to provide such information, the
   network needs to expose abstract information beyond the simple cost
   map abstraction.  In particular:

   *  The ALTO server must expose abstract information about the network
      paths that are traversed by the traffic between a source and a
      destination beyond a simple numerical value, which allows the
      overlay application to distinguish between Cases 1 and 2 and to
      compute the optimal total rate accordingly.

   *  The ALTO server must allow the client to distinguish the common
      ANE shared by "eh1 -> eh2" and "eh1 -> eh4", e.g., "eh1--sw1" and
      "sw1--sw5" in Case 1.

   *  The ALTO server must expose abstract information on the properties
      of the ANEs used by "eh1 -> eh2" and "eh1 -> eh4".  For example,
      an ALTO server can either expose the available bandwidth between
      "eh1--sw1", "sw1--sw5", "sw5--sw7", "sw5--sw6", "sw6--sw7",
      "sw7--sw2", "sw7--sw4", "sw2--eh2", "sw4--eh4" in Case 1 or expose
      three abstract elements "A", "B", and "C", which represent the
      linear constraints that define the same capacity region in Case 1.

   In general, we can conclude that to support the use case for multiple
   flow scheduling, the ALTO framework must be extended to satisfy the
   following additional requirements (ARs):

   AR1:  An ALTO server must provide the ANEs that are important for
      assessing the QoE of the overlay application on the path of a
      <source, destination> pair.

   AR2:  An ALTO server must provide information to identify how ANEs
      are shared on the paths of different <source, destination> pairs.

   AR3:  An ALTO server must provide information on the properties that
      are important for assessing the QoE of the application for ANEs.

   The extension defined in this document specifies a solution to expose
   such abstract information.

4.2.  Sample Use Cases

   While the problem related to multiple flow scheduling is used to help
   identify the additional requirements, the extension defined in this
   document can be applied to a wide range of applications.  This
   section highlights some of the reported use cases.

4.2.1.  Exposing Network Bottlenecks

   One important use case for the Path Vector extension is to expose
   network bottlenecks.  Applications that need to perform large-scale
   data transfers can benefit from being aware of the resource
   constraints exposed by this extension even if they have different
   objectives.  One such example is the Worldwide LHC Computing Grid
   (WLCG) (where "LHC" means "Large Hadron Collider"), which is the
   largest example of a distributed computation collaboration in the
   research and education world.

   Figure 3 illustrates an example of using an ALTO Path Vector as an
   interface between the job optimizer for a data analytics system and
   the network manager.  In particular, we assume that the objective of
   the job optimizer is to minimize the job completion time.

   In such a setting, the network-aware job optimizer (e.g., [CLARINET])
   takes a query and generates multiple query execution plans (QEPs).
   It can encode the QEPs as Path Vector requests that are sent to an
   ALTO server.  The ALTO server obtains the routing information for the
   flows in a QEP and finds links, routers, or middleboxes (e.g., a
   stateful firewall) that can potentially become bottlenecks for the
   QEP (e.g., see [NOVA] and [G2] for mechanisms to identify bottleneck
   links under different settings).  The resource constraint information
   is encoded in a Path Vector response and returned to the ALTO client.

   With the network resource constraints, the job optimizer may choose
   the QEP with the optimal job completion time to be executed.  It must
   be noted that the ALTO framework itself does not offer the capability
   to control the traffic.  However, certain network managers may offer
   ways to enforce resource guarantees, such as on-demand tunnels (e.g.,
   [SWAN]), demand vectors (e.g., [HUG], [UNICORN]), etc.  The traffic
   control interfaces and mechanisms are out of scope for this document.

                                        Data schema      Queries
                                             |             |
                                             \             /
          +-------------+                   +-----------------+
          | ALTO Client | <===============> |  Job Optimizer  |
          +-------------+                   +-----------------+
   PV          |   ^ PV                                    |
   Request     |   | Response                              |
               |   |                  On-demand resource   |
   (Potential  |   | (Network         allocation, demand   |
   Data        |   | Resource         vectors, etc.        |
   Transfers)  |   | Constraints)     (Non-ALTO interfaces)|
               v   |                                       v
          +-------------+                   +-----------------+
          | ALTO Server | <===============> | Network Manager |
          +-------------+                   +-----------------+
                                              /      |      \
                                              |      |      |
                                             WAN    DC1    DC2

               Figure 3: Example Use Case for Data Analytics

   Another example is illustrated in Figure 4.  Consider a network
   consisting of multiple sites and a non-blocking core network, i.e.,
   the links in the core network have sufficient bandwidth that they
   will not become a bottleneck for the data transfers.

                  Ongoing transfers    New transfer requests
                                \----\        |
                                     |        |
                                     v        v
      +-------------+               +---------------+
      | ALTO Client | <===========> | Data Transfer |
      +-------------+               |   Scheduler   |
        ^ |      ^ | PV Request     +---------------+
        | |      | \--------------\
        | |      \--------------\ |
        | v       PV Response   | v
      +-------------+          +-------------+
      | ALTO Server |          | ALTO Server |
      +-------------+          +-------------+
            ||                       ||
        +---------+              +---------+
        | Network |              | Network |
        | Manager |              | Manager |
        +---------+              +---------+
         .                           .
        .             _~_  __         . . .
       .             (   )(  )             .___
     ~v~v~       /--(         )------------(   )
    (     )-----/    (       )            (     )
     ~w~w~            ~^~^~^~              ~v~v~
    Site 1        Non-blocking Core        Site 2

       Figure 4: Example Use Case for Cross-Site Bottleneck Discovery

   With the Path Vector extension, a site can reveal the bottlenecks
   inside its own network with necessary information (such as link
   capacities) to the ALTO client, instead of providing the full
   topology and routing information, or no bottleneck information at
   all.  The bottleneck information can be used to analyze the impact of
   adding/removing data transfer flows, e.g., using the framework
   defined in [G2].  For example, assume that hosts "a", "b", and "c"
   are in Site 1 and hosts "d", "e", and "f" are in Site 2, and there
   are three flows in two sites: "a -> b", "c -> d", and "e -> f"
   (Figure 5).

   Site 1:

   [c]
    .
    ........................................> [d]
     +---+ 10 Gbps +---+ 10 Gbps +----+ 50 Gbps
     | A |---------| B |---------| GW |--------- Core
     +---+         +---+         +----+
    ...................
    .                 .
    .                 v
   [a]               [b]

   Site 2:

   [d] <........................................ [c]
     +---+ 5 Gbps +---+ 10 Gbps +----+ 20 Gbps
     | X |--------| Y |---------| GW |--------- Core
     +---+        +---+         +----+
                ....................
                .                  .
                .                  v
               [e]                [f]

                Figure 5: Example: Three Flows in Two Sites

   For these flows, Site 1 returns:

   a: { b: [ane1] },
   c: { d: [ane1, ane2, ane3] }

   ane1: bw = 10 Gbps (link: A->B)
   ane2: bw = 10 Gbps (link: B->GW)
   ane3: bw = 50 Gbps (link: GW->Core)

   and Site 2 returns:

   c: { d: [anei, aneii, aneiii] }
   e: { f: [aneiv] }

   anei: bw = 5 Gbps (link Y->X)
   aneii: bw = 10 Gbps (link GW->Y)
   aneiii: bw = 20 Gbps (link Core->GW)
   aneiv: bw = 10 Gbps (link Y->GW)

   With this information, the data transfer scheduler can use algorithms
   such as the theory on bottleneck structure [G2] to predict the
   potential throughput of the flows.

4.2.2.  Resource Exposure for CDNs and Service Edges

   At the time of this writing, a growing trend in today's applications
   is to bring storage and computation closer to the end users for
   better QoE, such as CDNs, augmented reality / virtual reality, and
   cloud gaming, as reported in various documents (e.g., [SEREDGE] and
   [MOWIE]).  ISPs may deploy multiple layers of CDN caches or, more
   generally, service edges, with different latencies and available
   resources, including the number of CPU cores, memory, and storage.

   For example, Figure 6 illustrates a typical edge-cloud scenario where
   memory is measured in gigabytes (GB) and storage is measured in
   terabytes (TB).  The "on-premise" edge nodes are closest to the end
   hosts and have the lowest latency, and the site-radio edge node and
   access central office (CO) have higher latencies but more available
   resources.

         +-------------+              +----------------------+
         | ALTO Client | <==========> | Application Provider |
         +-------------+              +----------------------+
   PV         |   ^ PV                      |
   Request    |   | Response                | Resource allocation,
              |   |                         | service establishment,
   (End hosts |   | (Edge nodes             | etc.
   and cloud  |   | and metrics)            |
   servers)   |   |                         |
              v   |                         v
         +-------------+             +---------------------+
         | ALTO Server | <=========> | Cloud-Edge Provider |
         +-------------+             +---------------------+
          ____________________________________/\___________
         /                                                 \
         |           (((o                                  |
                        |
                       /_\             _~_            __   __
     a               (/\_/\)          (   )          (  )~(  )_
      \      /------(      )---------(     )----\\---(          )
      _|_   /        (______)         (___)          (          )
      |_| -/         Site-radio     Access CO       (__________)
     /---\          Edge Node 1         |             Cloud DC
   On premise                           |
                              /---------/
              (((o           /
                 |          /
    Site-radio  /_\        /
   Edge Node 2(/\_/\)-----/
             /(_____)\
      ___   /         \   ---
   b--|_| -/           \--|_|--c
     /---\               /---\
   On premise          On premise

            Figure 6: Example Use Case for Service Edge Exposure

   With the extension defined in this document, an ALTO server can
   selectively reveal the CDNs and service edges that reside along the
   paths between different end hosts and/or the cloud servers, together
   with their properties (e.g., storage capabilities or Graphics
   Processing Unit (GPU) capabilities) and available Service Level
   Agreement (SLA) plans.  See Figure 7 for an example where the query
   is made for sources [a, b] and destinations [b, c, DC].  Here, each
   ANE represents a service edge, and the properties include access
   latency, available resources, etc.  Note that the properties here are
   only used for illustration purposes and are not part of this
   extension.

   a: { b: [ane1, ane2, ane3, ane4, ane5],
        c: [ane1, ane2, ane3, ane4, ane6],
        DC: [ane1, ane2, ane3] }
   b: { c: [ane5, ane4, ane6], DC: [ane5, ane4, ane3] }

   ane1: latency = 5 ms  cpu = 2  memory = 8 GB  storage = 10 TB
   (On premise, a)

   ane2: latency = 20 ms  cpu = 4  memory = 8 GB  storage = 10 TB
   (Site-radio Edge Node 1)

   ane3: latency = 100 ms  cpu = 8  memory = 128 GB  storage = 100 TB
   (Access CO)

   ane4: latency = 20 ms  cpu = 4  memory = 8 GB  storage = 10 TB
   (Site-radio Edge Node 2)

   ane5: latency = 5 ms  cpu = 2  memory = 8 GB  storage = 10 TB
   (On premise, b)

   ane6: latency = 5 ms  cpu = 2  memory = 8 GB  storage = 10 TB
   (On premise, c)

                Figure 7: Example Service Edge Query Results

   With the service edge information, an ALTO client may better conduct
   CDN request routing or offload functionalities from the user
   equipment to the service edge, with considerations in place for
   customized quality of experience.

5.  Path Vector Extension: Overview

   This section provides a non-normative overview of the Path Vector
   extension defined in this document.  It is assumed that readers are
   familiar with both the base protocol [RFC7285] and the entity
   property map extension [RFC9240].

   To satisfy the additional requirements listed in Section 4.1, this
   extension:

   1.  introduces the concept of an ANE as the abstraction of components
       in a network whose properties may have an impact on end-to-end
       performance of the traffic handled by those components,

   2.  extends the cost map and Endpoint Cost Service to convey the ANEs
       traversed by the path of a <source, destination> pair as Path
       Vectors, and

   3.  uses the entity property map to convey the association between
       the ANEs and their properties.

   Thus, an ALTO client can learn about the ANEs that are important for
   assessing the QoE of different <source, destination> pairs by
   investigating the corresponding Path Vector value (AR1) and can also
   (1) identify common ANEs if an ANE appears in the Path Vectors of
   multiple <source, destination> pairs (AR2) and (2) retrieve the
   properties of the ANEs by searching the entity property map (AR3).

5.1.  Abstract Network Element (ANE)

   This extension introduces the ANE as an indirect and network-agnostic
   way to specify a component or an aggregation of components of a
   network whose properties have an impact on end-to-end performance for
   application traffic between endpoints.

   ANEs allow ALTO servers to focus on common properties of different
   types of network components.  For example, the throughput of a flow
   can be constrained by different components in a network: the capacity
   of a physical link, the maximum throughput of a firewall, the
   reserved bandwidth of an MPLS tunnel, etc.  In the example below,
   assume that the throughput of the firewall is 100 Mbps and the
   capacity for link (A, B) is also 100 Mbps; they result in the same
   constraint on the total throughput of f1 and f2.  Thus, they are
   identical when treated as an ANE.

      f1 |      ^                  f1
         |      |                 ----------------->
       +----------+                +---+     +---+
       | Firewall |                | A |-----| B |
       +----------+                +---+     +---+
         |      |                 ----------------->
         v      | f2               f2

   When an ANE is defined by an ALTO server, it is assigned an
   identifier by the ALTO server, i.e., a string of type ANEName as
   specified in Section 6.1, and a set of associated properties.

5.1.1.  ANE Entity Domain

   In this extension, the associations between ANEs and their properties
   are conveyed in an entity property map.  Thus, ANEs must constitute
   an "entity domain" (Section 5.1 of [RFC9240]), and each ANE property
   must be an entity property (Section 5.2 of [RFC9240]).

   Specifically, this document defines a new entity domain called "ane"
   as specified in Section 6.2; Section 6.4 defines two initial property
   types for the ANE entity domain.

5.1.2.  Ephemeral and Persistent ANEs

   By design, ANEs are ephemeral and not to be used in further requests
   to other ALTO resources.  More precisely, the corresponding ANE names
   are no longer valid beyond the scope of a Path Vector response or the
   incremental update stream for a Path Vector request.  Compared with
   globally unique ANE names, ephemeral ANEs have several benefits,
   including better privacy for the ISP's internal structure and more
   flexible ANE computation.

   For example, an ALTO server may define an ANE for each aggregated
   bottleneck link between the sources and destinations specified in the
   request.  For requests with different sources and destinations, the
   bottlenecks may be different but can safely reuse the same ANE names.
   The client can still adjust its traffic based on the information, but
   it is difficult to infer the underlying topology with multiple
   queries.

   However, sometimes an ISP may intend to selectively reveal some
   "persistent" network components that, as opposed to being ephemeral,
   have a longer life cycle.  For example, an ALTO server may define an
   ANE for each service edge cluster.  Once a client chooses to use a
   service edge, e.g., by deploying some user-defined functions, it may
   want to stick to the service edge to avoid the complexity of state
   transition or synchronization, and continuously query the properties
   of the edge cluster.

   This document provides a mechanism to expose such network components
   as persistent ANEs.  A persistent ANE has a persistent ID that is
   registered in a property map, together with its properties.  See
   Sections 6.2.4 and 6.4.2 for more detailed instructions on how to
   identify ephemeral ANEs and persistent ANEs.

5.1.3.  Property Filtering

   Resource-constrained ALTO clients (see Section 4.1.2 of [RFC7285])
   may benefit from the filtering of Path Vector query results at the
   ALTO server, as an ALTO client may only require a subset of the
   available properties.

   Specifically, the available properties for a given resource are
   announced in the Information Resource Directory (IRD) as a new
   filtering capability called "ane-property-names".  The properties
   selected by a client as being of interest are specified in the
   subsequent Path Vector queries using the "ane-property-names" filter.
   The response only includes the selected properties for the ANEs.

   The "ane-property-names" capability for the cost map and the Endpoint
   Cost Service is specified in Sections 7.2.4 and 7.3.4, respectively.
   The "ane-property-names" filter for the cost map and the Endpoint
   Cost Service is specified in Sections 7.2.3 and 7.3.3 accordingly.

5.2.  Path Vector Cost Type

   For an ALTO client to correctly interpret the Path Vector, this
   extension specifies a new cost type called the "Path Vector cost
   type".

   The Path Vector cost type must convey both the interpretation and
   semantics in the "cost-mode" and "cost-metric" parameters,
   respectively.  Unfortunately, a single "cost-mode" value cannot fully
   specify the interpretation of a Path Vector, which is a compound data
   type.  For example, in programming languages such as C++, if there
   existed a JSON array type named JSONArray, a Path Vector would have
   the type of JSONArray<ANEName>.

   Instead of extending the "type system" of ALTO, this document takes a
   simple and backward-compatible approach.  Specifically, the "cost-
   mode" of the Path Vector cost type is "array", which indicates that
   the value is a JSON array.  Then, an ALTO client must check the value
   of the "cost-metric" parameter.  If the value is "ane-path", it means
   that the JSON array should be further interpreted as a path of
   ANENames.

   The Path Vector cost type is specified in Section 6.5.

5.3.  Multipart Path Vector Response

   For a basic ALTO information resource, a response contains only one
   type of ALTO resource, e.g., network map, cost map, or property
   map.  Thus, only one round of communication is required: an ALTO
   client sends a request to an ALTO server, and the ALTO server returns
   a response, as shown in Figure 8.

            ALTO client                              ALTO server
                 |-------------- Request ---------------->|
                 |<------------- Response ----------------|

               Figure 8: A Typical ALTO Request and Response

   The extension defined in this document, on the other hand, involves
   two types of information resources: Path Vectors conveyed in an
   InfoResourceCostMap data component (defined in Section 11.2.3.6 of
   [RFC7285]) or an InfoResourceEndpointCostMap data component (defined
   in Section 11.5.1.6 of [RFC7285]), and ANE properties conveyed in an
   InfoResourceProperties data component (defined in Section 7.6 of
   [RFC9240]).

   Instead of two consecutive message exchanges, the extension defined
   in this document enforces one round of communication.  Specifically,
   the ALTO client must include the source and destination pairs and the
   requested ANE properties in a single request, and the ALTO server
   must return a single response containing both the Path Vectors and
   properties associated with the ANEs in the Path Vectors, as shown in
   Figure 9.  Since the two parts are bundled together in one response
   message, their orders are interchangeable.  See Sections 7.2.6 and
   7.3.6 for details.

            ALTO client                              ALTO server
                 |------------- PV Request -------------->|
                 |<----- PV Response (Cost Map Part) -----|
                 |<--- PV Response (Property Map Part) ---|

          Figure 9: The Path Vector Extension Request and Response

   This design is based on the following considerations:

   1.  ANEs may be constructed on demand and, potentially, based on the
       requested properties (see Section 5.1 for more details).  If
       sources and destinations are not in the same request as the
       properties, an ALTO server either cannot construct ANEs on demand
       or must wait until both requests are received.

   2.  As ANEs may be constructed on demand, mappings of each ANE to its
       underlying network devices and resources can be specific to the
       request.  In order to respond to the property map request
       correctly, an ALTO server must store the mapping of each Path
       Vector request until the client fully retrieves the property
       information.  This "stateful" behavior may substantially harm
       server scalability and potentially lead to denial-of-service
       attacks.

   One approach for realizing the one-round communication is to define a
   new media type to contain both objects, but this violates modular
   design.  This document follows the standard-conforming usage of the
   "multipart/related" media type as defined in [RFC2387] to elegantly
   combine the objects.  Path Vectors are encoded in an
   InfoResourceCostMap data component or InfoResourceEndpointCostMap
   data component, and the property map is encoded in an
   InfoResourceProperties data component.  They are encapsulated as
   parts of a multipart message.  This modular composition allows ALTO
   servers and clients to reuse the data models of the existing
   information resources.  Specifically, this document addresses the
   following practical issues using "multipart/related".

5.3.1.  Identifying the Media Type of the Object Root

   ALTO uses a media type to indicate the type of an entry in the IRD
   (e.g., "application/alto-costmap+json" for the cost map and
   "application/alto-endpointcost+json" for the Endpoint Cost Service).
   Simply using "multipart/related" as the media type, however, makes it
   impossible for an ALTO client to identify the type of service
   provided by related entries.

   To address this issue, this document uses the "type" parameter to
   indicate the object root of a multipart/related message.  For a cost
   map resource, the "media-type" field in the IRD entry is "multipart/
   related" with the parameter "type=application/alto-costmap+json"; for
   an Endpoint Cost Service, the parameter is "type=application/alto-
   endpointcost+json".

5.3.2.  References to Part Messages

   As the response of a Path Vector resource is a multipart message with
   two different parts, it is important that each part can be uniquely
   identified.  Following the design provided in [RFC8895], this
   extension requires that an ALTO server assign a unique identifier to
   each part of the multipart response message.  This identifier,
   referred to as a Part Resource ID (see Section 6.6 for details), is
   present in the part message's "Content-ID" header field.  By
   concatenating the Part Resource ID to the identifier of the Path
   Vector request, an ALTO server/client can uniquely identify the Path
   Vector part or the property map part.

6.  Specification: Basic Data Types

6.1.  ANE Name

   An ANE name is encoded as a JSON string with the same format as that
   of the type PIDName (Section 10.1 of [RFC7285]).

   The type ANEName is used in this document to indicate a string of
   this format.

6.2.  ANE Entity Domain

   The ANE entity domain associates property values with the ANEs in a
   property map.  Accordingly, the ANE entity domain always depends on a
   property map.

   It must be noted that the term "domain" here does not refer to a
   network domain.  Rather, it is inherited from the entity domain as
   defined in Section 3.2 of [RFC9240]; the entity domain represents the
   set of valid entities defined by an ALTO information resource (called
   the "defining information resource").

6.2.1.  Entity Domain Type

   The entity domain type is "ane".

6.2.2.  Domain-Specific Entity Identifier

   The entity identifiers are the ANE names in the associated property
   map.

6.2.3.  Hierarchy and Inheritance

   There is no hierarchy or inheritance for properties associated with
   ANEs.

6.2.4.  Media Type of Defining Resource

   The defining resource for entity domain type "ane" MUST be a property
   map, i.e., the media type of defining resources is:

   application/alto-propmap+json

   Specifically, for ephemeral ANEs that appear in a Path Vector
   response, their entity domain names MUST be exactly ".ane", and the
   defining resource of these ANEs is the property map part of the
   multipart response.  Meanwhile, for any persistent ANE whose defining
   resource is a property map resource, its entity domain name MUST have
   the format of "PROPMAP.ane", where PROPMAP is the resource ID of the
   defining resource.  Persistent entities are "persistent" because
   standalone queries can be made by an ALTO client to their defining
   resource(s) when the connection to the Path Vector service is closed.

   For example, the defining resource of an ephemeral ANE whose entity
   identifier is ".ane:NET1" is the property map part that contains this
   identifier.  The defining resource of a persistent ANE whose entity
   identifier is "dc-props.ane:DC1" is the property map with the
   resource ID "dc-props".

6.3.  ANE Property Name

   An ANE property name is encoded as a JSON string with the same format
   as that of an entity property name (Section 5.2.2 of [RFC9240]).

6.4.  Initial ANE Property Types

   Two initial ANE property types are specified: "max-reservable-
   bandwidth" and "persistent-entity-id".

   Note that these property types do not depend on any information
   resources.  As such, the "EntityPropertyName" parameter MUST only
   have the EntityPropertyType part.

6.4.1.  Maximum Reservable Bandwidth

   The maximum reservable bandwidth property ("max-reservable-
   bandwidth") stands for the maximum bandwidth that can be reserved for
   all the traffic that traverses an ANE.  The value MUST be encoded as
   a non-negative numerical cost value as defined in Section 6.1.2.1 of
   [RFC7285], and the unit is bits per second (bps).  If this property
   is requested by the ALTO client but is not present for an ANE in the
   server response, it MUST be interpreted as meaning that the property
   is not defined for the ANE.

   This property can be offered in a setting where the ALTO server is
   part of a network system that provides on-demand resource allocation
   and the ALTO client is part of a user application.  One existing
   example is [NOVA]: the ALTO server is part of a Software-Defined
   Networking (SDN) controller and exposes a list of traversed network
   elements and associated link bandwidth to the client.  The encoding
   in [NOVA] differs from the Path Vector response defined in this
   document in that the Path Vector part and property map part are
   placed in the same JSON object.

   In such a framework, the ALTO server exposes resource availability
   information (e.g., reservable bandwidth) to the ALTO client.  How the
   client makes resource requests based on the information, and how the
   resource allocation is achieved, respectively, depend on interfaces
   between the management system and the users or a higher-layer
   protocol (e.g., SDN network intents [INTENT-BASED-NETWORKING] or MPLS
   tunnels), which are out of scope for this document.

6.4.2.  Persistent Entity ID

   This document enables the discovery of a persistent ANE by exposing
   its entity identifier as the persistent entity ID property of an
   ephemeral ANE in the path vector response.  The value of this
   property is encoded with the EntityID format defined in Section 5.1.3
   of [RFC9240].

   In this format, the entity ID combines:

   *  a defining information resource for the ANE on which a
      "persistent-entity-id" is queried, which is the property map
      resource defining the ANE as a persistent entity, together with
      the properties.

   *  the persistent name of the ANE in that property map.

   With this format, the client has all the needed information for
   further standalone query properties on the persistent ANE.

6.4.3.  Examples

   To illustrate the use of "max-reservable-bandwidth", consider the
   following network with five nodes.  Assume that the client wants to
   query the maximum reservable bandwidth from H1 to H2.  An ALTO server
   may split the network into two ANEs: "ane1", which represents the
   subnetwork with routers A, B, and C; and "ane2", which represents the
   subnetwork with routers B, D, and E.  The maximum reservable
   bandwidth for "ane1" is 15 Mbps (using path A->C->B), and the maximum
   reservable bandwidth for "ane2" is 20 Mbps (using path B->D->E).

                        20 Mbps  20 Mbps
             10 Mbps +---+   +---+    +---+
                /----| B |---| D |----| E |---- H2
          +---+/     +---+   +---+    +---+
   H1 ----| A | 15 Mbps|
          +---+\     +---+
                \----| C |
             15 Mbps +---+

   To illustrate the use of "persistent-entity-id", consider the
   scenario in Figure 6.  As the life cycles of service edges are
   typically long, the service edges may contain information that is not
   specific to the query.  Such information can be stored in an
   individual entity property map and can later be accessed by an ALTO
   client.

   For example, "ane1" in Figure 7 represents the on-premise service
   edge closest to host "a".  Assume that the properties of the service
   edges are provided in an entity property map called "se-props" and
   the ID of the on-premise service edge is "9a0b55f7-7442-4d56-8a2c-
   b4cc6a8e3aa1"; the "persistent-entity-id" setting for "ane1" will be
   "se-props.ane:9a0b55f7-7442-4d56-8a2c-b4cc6a8e3aa1".  With this
   persistent entity ID, an ALTO client may send queries to the "se-
   props" resource with the entity ID ".ane:9a0b55f7-7442-4d56-8a2c-
   b4cc6a8e3aa1".

6.5.  Path Vector Cost Type

   This document defines a new cost type, which is referred to as the
   Path Vector cost type.  An ALTO server MUST offer this cost type if
   it supports the extension defined in this document.

6.5.1.  Cost Metric: "ane-path"

   The cost metric "ane-path" indicates that the value of such a cost
   type conveys an array of ANE names, where each ANE name uniquely
   represents an ANE traversed by traffic from a source to a
   destination.

   An ALTO client MUST interpret the Path Vector as if the traffic
   between a source and a destination logically traverses the ANEs in
   the same order as they appear in the Path Vector.

   When the Path Vector procedures defined in this document are in use,
   an ALTO server using the "ane-path" cost metric and the "array" cost
   mode (see Section 6.5.2) MUST return as the cost value a JSON array
   of data type ANEName, and the client MUST also check that each
   element contained in the array is an ANEName (Section 6.1).
   Otherwise, the client MUST discard the response and SHOULD follow the
   guidance in Section 8.3.4.3 of [RFC7285] to handle the error.

6.5.2.  Cost Mode: "array"

   The cost mode "array" indicates that every cost value in the response
   body of a (filtered) cost map or an Endpoint Cost Service MUST be
   interpreted as a JSON array object.  While this cost mode can be
   applied to all cost metrics, additional specifications will be needed
   to clarify the semantics of the "array" cost mode when combined with
   cost metrics other than "ane-path".

6.6.  Part Resource ID and Part Content ID

   A Part Resource ID is encoded as a JSON string with the same format
   as that of the type ResourceID (Section 10.2 of [RFC7285]).

   Even though the "client-id" assigned to a Path Vector request and the
   Part Resource ID MAY contain up to 64 characters by their own
   definition, their concatenation (see Section 5.3.2) MUST also conform
   to the same length constraint.  The same requirement applies to the
   resource ID of the Path Vector resource, too.  Thus, it is
   RECOMMENDED to limit the length of the resource ID and client ID
   related to a Path Vector resource to 31 characters.

   A Part Content ID conforms to the format of "msg-id" as specified in
   [RFC2387] and [RFC5322].  Specifically, it has the following format:

   "<" PART-RESOURCE-ID "@" DOMAIN-NAME ">"

   PART-RESOURCE-ID:  PART-RESOURCE-ID has the same format as the Part
      Resource ID.  It is used to identify whether a part message is a
      Path Vector or a property map.

   DOMAIN-NAME:  DOMAIN-NAME has the same format as "dot-atom-text" as
      specified in Section 3.2.3 of [RFC5322].  It must be the domain
      name of the ALTO server.

7.  Specification: Service Extensions

7.1.  Notation

   This document uses the same syntax and notation as those introduced
   in Section 8.2 of [RFC7285] to specify the extensions to existing
   ALTO resources and services.

7.2.  Multipart Filtered Cost Map for Path Vector

   This document introduces a new ALTO resource called the "multipart
   filtered cost map resource", which allows an ALTO server to provide
   other ALTO resources associated with the cost map resource in the
   same response.

7.2.1.  Media Type

   The media type of the multipart filtered cost map resource is
   "multipart/related", and the required "type" parameter MUST have a
   value of "application/alto-costmap+json".

7.2.2.  HTTP Method

   The multipart filtered cost map is requested using the HTTP POST
   method.

7.2.3.  Accept Input Parameters

   The input parameters of the multipart filtered cost map are supplied
   in the body of an HTTP POST request.  This document extends the input
   parameters to a filtered cost map, which is defined as a JSON object
   of type ReqFilteredCostMap in Section 4.1.2 of [RFC8189], with a data
   format indicated by the media type "application/alto-
   costmapfilter+json", which is a JSON object of type
   PVReqFilteredCostMap:

   object {
     [EntityPropertyName ane-property-names<0..*>;]
   } PVReqFilteredCostMap : ReqFilteredCostMap;

   with field:

   ane-property-names:  This field provides a list of selected ANE
      properties to be included in the response.  Each property in this
      list MUST match one of the supported ANE properties indicated in
      the resource's "ane-property-names" capability (Section 7.2.4).
      If the field is not present, it MUST be interpreted as an empty
      list.

   Example: Consider the network in Figure 1.  If an ALTO client wants
   to query the "max-reservable-bandwidth" setting between PID1 and
   PID2, it can submit the following request.

      POST /costmap/pv HTTP/1.1
      Host: alto.example.com
      Accept: multipart/related;type=application/alto-costmap+json,
              application/alto-error+json
      Content-Length: 212
      Content-Type: application/alto-costmapfilter+json

      {
        "cost-type": {
          "cost-mode": "array",
          "cost-metric": "ane-path"
        },
        "pids": {
          "srcs": [ "PID1" ],
          "dsts": [ "PID2" ]
        },
        "ane-property-names": [ "max-reservable-bandwidth" ]
      }

7.2.4.  Capabilities

   The multipart filtered cost map resource extends the capabilities
   defined in Section 4.1.1 of [RFC8189].  The capabilities are defined
   by a JSON object of type PVFilteredCostMapCapabilities:

   object {
     [EntityPropertyName ane-property-names<0..*>;]
   } PVFilteredCostMapCapabilities : FilteredCostMapCapabilities;

   with field:

   ane-property-names:  This field provides a list of ANE properties
      that can be returned.  If the field is not present, it MUST be
      interpreted as an empty list, indicating that the ALTO server
      cannot provide any ANE properties.

   This extension also introduces additional restrictions for the
   following fields:

   cost-type-names:  The "cost-type-names" field MUST include the Path
      Vector cost type, unless explicitly documented by a future
      extension.  This also implies that the Path Vector cost type MUST
      be defined in the "cost-types" of the IRD's "meta" field.

   cost-constraints:  If the "cost-type-names" field includes the Path
      Vector cost type, the "cost-constraints" field MUST be either
      "false" or not present, unless specifically instructed by a future
      document.

   testable-cost-type-names (Section 4.1.1 of [RFC8189]):  If the "cost-
      type-names" field includes the Path Vector cost type and the
      "testable-cost-type-names" field is present, the Path Vector cost
      type MUST NOT be included in the "testable-cost-type-names" field
      unless specifically instructed by a future document.

7.2.5.  Uses

   This member MUST include the resource ID of the network map based on
   which the PIDs are defined.  If this resource supports "persistent-
   entity-id", it MUST also include the defining resources of persistent
   ANEs that may appear in the response.

7.2.6.  Response

   The response MUST indicate an error, using ALTO Protocol error
   handling as defined in Section 8.5 of [RFC7285], if the request is
   invalid.

   The "Content-Type" header field of the response MUST be "multipart/
   related" as defined by [RFC2387], with the following parameters:

   type:  The "type" parameter is mandatory and MUST be "application/
      alto-costmap+json".  Note that [RFC2387] permits parameters both
      with and without double quotes.

   start:  The "start" parameter is as defined in [RFC2387] and is
      optional.  If present, it MUST have the same value as the
      "Content-ID" header field of the Path Vector part.

   boundary:  The "boundary" parameter is as defined in Section 5.1.1 of
      [RFC2046] and is mandatory.

   The body of the response MUST consist of two parts:

   *  The Path Vector part MUST include "Content-ID" and "Content-Type"
      in its header.  The "Content-Type" MUST be "application/alto-
      costmap+json".  The value of "Content-ID" MUST have the same
      format as the Part Content ID as specified in Section 6.6.

      The body of the Path Vector part MUST be a JSON object with the
      same format as that defined in Section 11.2.3.6 of [RFC7285] when
      the "cost-type" field is present in the input parameters and MUST
      be a JSON object with the same format as that defined in
      Section 4.1.3 of [RFC8189] if the "multi-cost-types" field is
      present.  The JSON object MUST include the "vtag" field in the
      "meta" field, which provides the version tag of the returned
      CostMapData object.  The resource ID of the version tag MUST
      follow the format of

      resource-id '.' part-resource-id

      where "resource-id" is the resource ID of the Path Vector resource
      and "part-resource-id" has the same value as the PART-RESOURCE-ID
      in the "Content-ID" of the Path Vector part.  The "meta" field
      MUST also include the "dependent-vtags" field, whose value is a
      single-element array to indicate the version tag of the network
      map used, where the network map is specified in the "uses"
      attribute of the multipart filtered cost map resource in the IRD.

   *  The entity property map part MUST also include "Content-ID" and
      "Content-Type" in its header.  The "Content-Type" MUST be
      "application/alto-propmap+json".  The value of "Content-ID" MUST
      have the same format as the Part Content ID as specified in
      Section 6.6.

      The body of the entity property map part is a JSON object with the
      same format as that defined in Section 7.6 of [RFC9240].  The JSON
      object MUST include the "dependent-vtags" field in the "meta"
      field.  The value of the "dependent-vtags" field MUST be an array
      of VersionTag objects as defined by Section 10.3 of [RFC7285].
      The "vtag" of the Path Vector part MUST be included in the
      "dependent-vtags" field.  If "persistent-entity-id" is requested,
      the version tags of the dependent resources that may expose the
      entities in the response MUST also be included.

      The PropertyMapData object has one member for each ANEName that
      appears in the Path Vector part, which is an entity identifier
      belonging to the self-defined entity domain as defined in
      Section 5.1.2.3 of [RFC9240].  The EntityProps object for each ANE
      has one member for each property that is both 1) associated with
      the ANE and 2) specified in the "ane-property-names" field in the
      request.  If the Path Vector cost type is not included in the
      "cost-type" field or the "multi-cost-type" field, the "property-
      map" field MUST be present and the value MUST be an empty object
      ({}).

   A complete and valid response MUST include both the Path Vector part
   and the property map part in the multipart message.  If any part is
   *not* present, the client MUST discard the received information and
   send another request if necessary.

   The Path Vector part, whose media type is the same as the "type"
   parameter of the multipart response message, is the root body part as
   defined in [RFC2387].  Thus, it is the element that the application
   processes first.  Even though the "start" parameter allows it to be
   placed anywhere in the part sequence, it is RECOMMENDED that the
   parts arrive in the same order as they are processed, i.e., the Path
   Vector part is always placed as the first part, followed by the
   property map part.  When doing so, an ALTO server MAY choose not to
   set the "start" parameter, which implies that the first part is the
   object root.

   Example: Consider the network in Figure 1.  The response to the
   example request in Section 7.2.3 is as follows, where "ANE1"
   represents the aggregation of all the switches in the network.

   HTTP/1.1 200 OK
   Content-Length: 911
   Content-Type: multipart/related; boundary=example-1;
                 type=application/alto-costmap+json

   --example-1
   Content-ID: <costmap@alto.example.com>
   Content-Type: application/alto-costmap+json

   {
     "meta": {
       "vtag": {
         "resource-id": "filtered-cost-map-pv.costmap",
         "tag": "fb20b76204814e9db37a51151faaaef2"
       },
       "dependent-vtags": [
         {
           "resource-id": "my-default-networkmap",
           "tag": "75ed013b3cb58f896e839582504f6228"
         }
       ],
       "cost-type": { "cost-mode": "array", "cost-metric": "ane-path" }
     },
     "cost-map": {
       "PID1": { "PID2": [ "ANE1" ] }
     }
   }
   --example-1
   Content-ID: <propmap@alto.example.com>
   Content-Type: application/alto-propmap+json

   {
     "meta": {
       "dependent-vtags": [
         {
           "resource-id": "filtered-cost-map-pv.costmap",
           "tag": "fb20b76204814e9db37a51151faaaef2"
         }
       ]
     },
     "property-map": {
       ".ane:ANE1": { "max-reservable-bandwidth": 100000000 }
     }
   }
   --example-1

7.3.  Multipart Endpoint Cost Service for Path Vector

   This document introduces a new ALTO resource called the "multipart
   Endpoint Cost Service", which allows an ALTO server to provide other
   ALTO resources associated with the Endpoint Cost Service resource in
   the same response.

7.3.1.  Media Type

   The media type of the multipart Endpoint Cost Service resource is
   "multipart/related", and the required "type" parameter MUST have a
   value of "application/alto-endpointcost+json".

7.3.2.  HTTP Method

   The multipart Endpoint Cost Service resource is requested using the
   HTTP POST method.

7.3.3.  Accept Input Parameters

   The input parameters of the multipart Endpoint Cost Service resource
   are supplied in the body of an HTTP POST request.  This document
   extends the input parameters to an Endpoint Cost Service, which is
   defined as a JSON object of type ReqEndpointCostMap in Section 4.2.2
   of [RFC8189], with a data format indicated by the media type
   "application/alto-endpointcostparams+json", which is a JSON object of
   type PVReqEndpointCostMap:

   object {
     [EntityPropertyName ane-property-names<0..*>;]
   } PVReqEndpointCostMap : ReqEndpointCostMap;

   with field:

   ane-property-names:  This document defines the "ane-property-names"
      field in PVReqEndpointCostMap as being the same as in
      PVReqFilteredCostMap.  See Section 7.2.3.

   Example: Consider the network in Figure 1.  If an ALTO client wants
   to query the "max-reservable-bandwidth" setting between "eh1" and
   "eh2", it can submit the following request.

   POST /ecs/pv HTTP/1.1
   Host: alto.example.com
   Accept: multipart/related;type=application/alto-endpointcost+json,
           application/alto-error+json
   Content-Length: 238
   Content-Type: application/alto-endpointcostparams+json

   {
     "cost-type": {
       "cost-mode": "array",
       "cost-metric": "ane-path"
     },
     "endpoints": {
       "srcs": [ "ipv4:192.0.2.2" ],
       "dsts": [ "ipv4:192.0.2.18" ]
     },
     "ane-property-names": [ "max-reservable-bandwidth" ]
   }

7.3.4.  Capabilities

   The capabilities of the multipart Endpoint Cost Service resource are
   defined by a JSON object of type PVEndpointCostCapabilities, which is
   defined as being the same as PVFilteredCostMapCapabilities.  See
   Section 7.2.4.

7.3.5.  Uses

   If this resource supports "persistent-entity-id", it MUST also
   include the defining resources of persistent ANEs that may appear in
   the response.

7.3.6.  Response

   The response MUST indicate an error, using ALTO Protocol error
   handling as defined in Section 8.5 of [RFC7285], if the request is
   invalid.

   The "Content-Type" header field of the response MUST be "multipart/
   related" as defined by [RFC2387], with the following parameters:

   type:  The "type" parameter MUST be "application/alto-
      endpointcost+json" and is mandatory.

   start:  The "start" parameter is as defined in Section 7.2.6.

   boundary:  The "boundary" parameter is as defined in Section 5.1.1 of
      [RFC2046] and is mandatory.

   The body of the response MUST consist of two parts:

   *  The Path Vector part MUST include "Content-ID" and "Content-Type"
      in its header.  The "Content-Type" MUST be "application/alto-
      endpointcost+json".  The value of "Content-ID" MUST have the same
      format as the Part Content ID as specified in Section 6.6.

      The body of the Path Vector part MUST be a JSON object with the
      same format as that defined in Section 11.5.1.6 of [RFC7285] when
      the "cost-type" field is present in the input parameters and MUST
      be a JSON object with the same format as that defined in
      Section 4.2.3 of [RFC8189] if the "multi-cost-types" field is
      present.  The JSON object MUST include the "vtag" field in the
      "meta" field, which provides the version tag of the returned
      EndpointCostMapData object.  The resource ID of the version tag
      MUST follow the format of

      resource-id '.' part-resource-id

      where "resource-id" is the resource ID of the Path Vector resource
      and "part-resource-id" has the same value as the PART-RESOURCE-ID
      in the "Content-ID" of the Path Vector part.

   *  The entity property map part MUST also include "Content-ID" and
      "Content-Type" in its header.  The "Content-Type" MUST be
      "application/alto-propmap+json".  The value of "Content-ID" MUST
      have the same format as the Part Content ID as specified in
      Section 6.6.

      The body of the entity property map part MUST be a JSON object
      with the same format as that defined in Section 7.6 of [RFC9240].
      The JSON object MUST include the "dependent-vtags" field in the
      "meta" field.  The value of the "dependent-vtags" field MUST be an
      array of VersionTag objects as defined by Section 10.3 of
      [RFC7285].  The "vtag" of the Path Vector part MUST be included in
      the "dependent-vtags" field.  If "persistent-entity-id" is
      requested, the version tags of the dependent resources that may
      expose the entities in the response MUST also be included.

      The PropertyMapData object has one member for each ANEName that
      appears in the Path Vector part, which is an entity identifier
      belonging to the self-defined entity domain as defined in
      Section 5.1.2.3 of [RFC9240].  The EntityProps object for each ANE
      has one member for each property that is both 1) associated with
      the ANE and 2) specified in the "ane-property-names" field in the
      request.  If the Path Vector cost type is not included in the
      "cost-type" field or the "multi-cost-type" field, the "property-
      map" field MUST be present and the value MUST be an empty object
      ({}).

   A complete and valid response MUST include both the Path Vector part
   and the property map part in the multipart message.  If any part is
   *not* present, the client MUST discard the received information and
   send another request if necessary.

   The Path Vector part, whose media type is the same as the "type"
   parameter of the multipart response message, is the root body part as
   defined in [RFC2387].  Thus, it is the element that the application
   processes first.  Even though the "start" parameter allows it to be
   placed anywhere in the part sequence, it is RECOMMENDED that the
   parts arrive in the same order as they are processed, i.e., the Path
   Vector part is always placed as the first part, followed by the
   property map part.  When doing so, an ALTO server MAY choose not to
   set the "start" parameter, which implies that the first part is the
   object root.

   Example: Consider the network in Figure 1.  The response to the
   example request in Section 7.3.3 is as follows.

   HTTP/1.1 200 OK
   Content-Length: 899
   Content-Type: multipart/related; boundary=example-1;
                 type=application/alto-endpointcost+json

   --example-1
   Content-ID: <ecs@alto.example.com>
   Content-Type: application/alto-endpointcost+json

   {
     "meta": {
       "vtag": {
         "resource-id": "ecs-pv.ecs",
         "tag": "ec137bb78118468c853d5b622ac003f1"
       },
       "dependent-vtags": [
         {
           "resource-id": "my-default-networkmap",
           "tag": "677fe5f4066848d282ece213a84f9429"
         }
       ],
       "cost-type": { "cost-mode": "array", "cost-metric": "ane-path" }
     },
     "cost-map": {
       "ipv4:192.0.2.2": { "ipv4:192.0.2.18": [ "ANE1" ] }
     }
   }
   --example-1
   Content-ID: <propmap@alto.example.com>
   Content-Type: application/alto-propmap+json

   {
     "meta": {
       "dependent-vtags": [
         {
           "resource-id": "ecs-pv.ecs",
           "tag": "ec137bb78118468c853d5b622ac003f1"
         }
       ]
     },
     "property-map": {
       ".ane:ANE1": { "max-reservable-bandwidth": 100000000 }
     }
   }
   --example-1

8.  Examples

   This section lists some examples of Path Vector queries and the
   corresponding responses.  Some long lines are truncated for better
   readability.

8.1.  Sample Setup

   Figure 10 illustrates the network properties and thus the message
   contents.  There are three subnetworks (NET1, NET2, and NET3) and two
   interconnection links (L1 and L2).  It is assumed that each
   subnetwork has sufficiently large bandwidth to be reserved.

                                         ----- L1
                                        /
            PID1   +----------+ 10 Gbps +----------+    PID3
     192.0.2.0/28+-+ +------+ +---------+          +--+192.0.2.32/28
                   | | MEC1 | |         |          |   2001:db8::3:0/16
                   | +------+ |   +-----+          |
            PID2   |          |   |     +----------+
    192.0.2.16/28+-+          |   |         NET3
                   |          |   | 15 Gbps
                   |          |   |        \
                   +----------+   |         -------- L2
                       NET1       |
                                +----------+
                                | +------+ |   PID4
                                | | MEC2 | +--+192.0.2.48/28
                                | +------+ |   2001:db8::4:0/16
                                +----------+
                                    NET2

                   Figure 10: Examples of ANE Properties

8.2.  Information Resource Directory

   To give a comprehensive example of the extension defined in this
   document, we consider the network in Figure 10.  Assume that the ALTO
   server provides the following information resources:

   "my-default-networkmap":  A network map resource that contains the
      PIDs in the network.

   "filtered-cost-map-pv":  A multipart filtered cost map resource for
      the Path Vector.  Exposes the "max-reservable-bandwidth" property
      for the PIDs in "my-default-networkmap".

   "ane-props":  A filtered entity property resource that exposes the
      information for persistent ANEs in the network.

   "endpoint-cost-pv":  A multipart Endpoint Cost Service for the Path
      Vector.  Exposes the "max-reservable-bandwidth" and "persistent-
      entity-id" properties.

   "update-pv":  An update stream service that provides the incremental
      update service for the "endpoint-cost-pv" service.

   "multicost-pv":  A multipart Endpoint Cost Service with both the
      Multi-Cost extension and Path Vector extension enabled.

   Below is the IRD of the example ALTO server.  To enable the extension
   defined in this document, the Path Vector cost type (Section 6.5),
   represented by "path-vector" below, is defined in the "cost-types" of
   the "meta" field and is included in the "cost-type-names" of
   resources "filtered-cost-map-pv" and "endpoint-cost-pv".

   {
     "meta": {
       "cost-types": {
         "path-vector": {
           "cost-mode": "array",
           "cost-metric": "ane-path"
         },
         "num-rc": {
           "cost-mode": "numerical",
           "cost-metric": "routingcost"
         }
       }
     },
     "resources": {
       "my-default-networkmap": {
         "uri": "https://alto.example.com/networkmap",
         "media-type": "application/alto-networkmap+json"
       },
       "filtered-cost-map-pv": {
         "uri": "https://alto.example.com/costmap/pv",
         "media-type": "multipart/related;
                        type=application/alto-costmap+json",
         "accepts": "application/alto-costmapfilter+json",
         "capabilities": {
           "cost-type-names": [ "path-vector" ],
           "ane-property-names": [ "max-reservable-bandwidth" ]
         },
         "uses": [ "my-default-networkmap" ]
       },
       "ane-props": {
         "uri": "https://alto.example.com/ane-props",
         "media-type": "application/alto-propmap+json",
         "accepts": "application/alto-propmapparams+json",
         "capabilities": {
           "mappings": {
             ".ane": [ "cpu" ]
           }
         }
       },
       "endpoint-cost-pv": {
         "uri": "https://alto.exmaple.com/endpointcost/pv",
         "media-type": "multipart/related;
                        type=application/alto-endpointcost+json",
         "accepts": "application/alto-endpointcostparams+json",
         "capabilities": {
           "cost-type-names": [ "path-vector" ],
           "ane-property-names": [
             "max-reservable-bandwidth", "persistent-entity-id"
           ]
         },
         "uses": [ "ane-props" ]
       },
       "update-pv": {
         "uri": "https://alto.example.com/updates/pv",
         "media-type": "text/event-stream",
         "uses": [ "endpoint-cost-pv" ],
         "accepts": "application/alto-updatestreamparams+json",
         "capabilities": {
           "support-stream-control": true
         }
       },
       "multicost-pv": {
         "uri": "https://alto.exmaple.com/endpointcost/mcpv",
         "media-type": "multipart/related;
                        type=application/alto-endpointcost+json",
         "accepts": "application/alto-endpointcostparams+json",
         "capabilities": {
           "cost-type-names": [ "path-vector", "num-rc" ],
           "max-cost-types": 2,
           "testable-cost-type-names": [ "num-rc" ],
           "ane-property-names": [
             "max-reservable-bandwidth", "persistent-entity-id"
           ]
         },
         "uses": [ "ane-props" ]
       }
     }
   }

8.3.  Multipart Filtered Cost Map

   The following examples demonstrate the request to the "filtered-cost-
   map-pv" resource and the corresponding response.

   The request uses the "path-vector" cost type in the "cost-type"
   field.  The "ane-property-names" field is missing, indicating that
   the client only requests the Path Vector and not the ANE properties.

   The response consists of two parts:

   *  The first part returns the array of data type ANEName for each
      source and destination pair.  There are two ANEs, where "L1"
      represents interconnection link L1 and "L2" represents
      interconnection link L2.

   *  The second part returns the property map.  Note that the
      properties of the ANE entries are equal to the literal string "{}"
      (see Section 8.3 of [RFC9240]).

   POST /costmap/pv HTTP/1.1
   Host: alto.example.com
   Accept: multipart/related;type=application/alto-costmap+json,
           application/alto-error+json
   Content-Length: 163
   Content-Type: application/alto-costmapfilter+json

   {
     "cost-type": {
       "cost-mode": "array",
       "cost-metric": "ane-path"
     },
     "pids": {
       "srcs": [ "PID1" ],
       "dsts": [ "PID3", "PID4" ]
     }
   }

   HTTP/1.1 200 OK
   Content-Length: 952
   Content-Type: multipart/related; boundary=example-1;
                 type=application/alto-costmap+json

   --example-1
   Content-ID: <costmap@alto.example.com>
   Content-Type: application/alto-costmap+json

   {
     "meta": {
       "vtag": {
         "resource-id": "filtered-cost-map-pv.costmap",
         "tag": "d827f484cb66ce6df6b5077cb8562b0a"
       },
       "dependent-vtags": [
         {
           "resource-id": "my-default-networkmap",
           "tag": "c04bc5da49534274a6daeee8ea1dec62"
         }
       ],
       "cost-type": {
         "cost-mode": "array",
         "cost-metric": "ane-path"
       }
     },
     "cost-map": {
       "PID1": {
         "PID3": [ "L1" ],
         "PID4": [ "L1", "L2" ]
       }
     }
   }
   --example-1
   Content-ID: <propmap@alto.example.com>
   Content-Type: application/alto-propmap+json

   {
     "meta": {
       "dependent-vtags": [
         {
           "resource-id": "filtered-cost-map-pv.costmap",
           "tag": "d827f484cb66ce6df6b5077cb8562b0a"
         }
       ]
     },
     "property-map": {
       ".ane:L1": {},
       ".ane:L2": {}
     }
   }
   --example-1

8.4.  Multipart Endpoint Cost Service Resource

   The following examples demonstrate the request to the "endpoint-cost-
   pv" resource and the corresponding response.

   The request uses the "path-vector" cost type in the "cost-type" field
   and queries the maximum reservable bandwidth ANE property and the
   persistent entity ID property for two IPv4 source and destination
   pairs (192.0.2.34 -> 192.0.2.2 and 192.0.2.34 -> 192.0.2.50) and one
   IPv6 source and destination pair (2001:db8::3:1 -> 2001:db8::4:1).

   The response consists of two parts:

   *  The first part returns the array of data type ANEName for each
      valid source and destination pair.  As one can see in Figure 10,
      flow 192.0.2.34 -> 192.0.2.2 traverses NET3, L1, and NET1; and
      flows 192.0.2.34 -> 192.0.2.50 and 2001:db8::3:1 -> 2001:db8::4:1
      traverse NET2, L2, and NET3.

   *  The second part returns the requested properties of ANEs.  Assume
      that NET1, NET2, and NET3 have sufficient bandwidth and their
      "max-reservable-bandwidth" values are set to a sufficiently large
      number (50 Gbps in this case).  On the other hand, assume that
      there are no prior reservations on L1 and L2 and their "max-
      reservable-bandwidth" values are the corresponding link capacity
      (10 Gbps for L1 and 15 Gbps for L2).

   Both NET1 and NET2 have a mobile edge deployed, i.e., MEC1 in NET1
   and MEC2 in NET2.  Assume that the ANEName values for MEC1 and MEC2
   are "MEC1" and "MEC2" and their properties can be retrieved from the
   property map "ane-props".  Thus, the "persistent-entity-id" property
   values for NET1 and NET2 are "ane-props.ane:MEC1" and "ane-
   props.ane:MEC2", respectively.

   POST /endpointcost/pv HTTP/1.1
   Host: alto.example.com
   Accept: multipart/related;
           type=application/alto-endpointcost+json,
           application/alto-error+json
   Content-Length: 383
   Content-Type: application/alto-endpointcostparams+json

   {
     "cost-type": {
       "cost-mode": "array",
       "cost-metric": "ane-path"
     },
     "endpoints": {
       "srcs": [
         "ipv4:192.0.2.34",
         "ipv6:2001:db8::3:1"
       ],
       "dsts": [
         "ipv4:192.0.2.2",
         "ipv4:192.0.2.50",
         "ipv6:2001:db8::4:1"
       ]
     },
     "ane-property-names": [
       "max-reservable-bandwidth",
       "persistent-entity-id"
     ]
   }

   HTTP/1.1 200 OK
   Content-Length: 1508
   Content-Type: multipart/related; boundary=example-2;
                 type=application/alto-endpointcost+json

   --example-2
   Content-ID: <ecs@alto.example.com>
   Content-Type: application/alto-endpointcost+json

   {
     "meta": {
       "vtags": {
         "resource-id": "endpoint-cost-pv.ecs",
         "tag": "bb6bb72eafe8f9bdc4f335c7ed3b10822a391cef"
       },
       "cost-type": {
         "cost-mode": "array",
         "cost-metric": "ane-path"
       }
     },
     "endpoint-cost-map": {
       "ipv4:192.0.2.34": {
         "ipv4:192.0.2.2":   [ "NET3", "L1", "NET1" ],
         "ipv4:192.0.2.50":   [ "NET3", "L2", "NET2" ]
       },
       "ipv6:2001:db8::3:1": {
         "ipv6:2001:db8::4:1": [ "NET3", "L2", "NET2" ]
       }
     }
   }
   --example-2
   Content-ID: <propmap@alto.example.com>
   Content-Type: application/alto-propmap+json

   {
     "meta": {
       "dependent-vtags": [
         {
           "resource-id": "endpoint-cost-pv.ecs",
           "tag": "bb6bb72eafe8f9bdc4f335c7ed3b10822a391cef"
         },
         {
           "resource-id": "ane-props",
           "tag": "bf3c8c1819d2421c9a95a9d02af557a3"
         }
       ]
     },
     "property-map": {
       ".ane:NET1": {
         "max-reservable-bandwidth": 50000000000,
         "persistent-entity-id": "ane-props.ane:MEC1"
       },
       ".ane:NET2": {
         "max-reservable-bandwidth": 50000000000,
         "persistent-entity-id": "ane-props.ane:MEC2"
       },
       ".ane:NET3": {
         "max-reservable-bandwidth": 50000000000
       },
       ".ane:L1": {
         "max-reservable-bandwidth": 10000000000
       },
       ".ane:L2": {
         "max-reservable-bandwidth": 15000000000
       }
     }
   }
   --example-2

   In certain scenarios where the traversal order is not crucial, an
   ALTO server implementation may choose not to strictly follow the
   physical traversal order and may even obfuscate the order
   intentionally to preserve its own privacy or conform to its own
   policies.  For example, an ALTO server may choose to aggregate NET1
   and L1 as a new ANE with ANE name "AGGR1" and aggregate NET2 and L2
   as a new ANE with ANE name "AGGR2".  The "max-reservable-bandwidth"
   property of "AGGR1" takes the value of L1, which is smaller than that
   of NET1, and the "persistent-entity-id" property of "AGGR1" takes the
   value of NET1.  The properties of "AGGR2" are computed in a similar
   way; the obfuscated response is as shown below.  Note that the
   obfuscation of Path Vector responses is implementation specific and
   is out of scope for this document.  Developers may refer to
   Section 11 for further references.

   HTTP/1.1 200 OK
   Content-Length: 1333
   Content-Type: multipart/related; boundary=example-2;
                 type=application/alto-endpointcost+json

   --example-2
   Content-ID: <ecs@alto.example.com>
   Content-Type: application/alto-endpointcost+json

   {
     "meta": {
       "vtags": {
         "resource-id": "endpoint-cost-pv.ecs",
         "tag": "bb975862fbe3422abf4dae386b132c1d"
       },
       "cost-type": {
         "cost-mode": "array",
         "cost-metric": "ane-path"
       }
     },
     "endpoint-cost-map": {
       "ipv4:192.0.2.34": {
         "ipv4:192.0.2.2":   [ "NET3", "AGGR1" ],
         "ipv4:192.0.2.50":   [ "NET3", "AGGR2" ]
       },
       "ipv6:2001:db8::3:1": {
         "ipv6:2001:db8::4:1": [ "NET3", "AGGR2" ]
       }
     }
   }
   --example-2
   Content-ID: <propmap@alto.example.com>
   Content-Type: application/alto-propmap+json

   {
     "meta": {
       "dependent-vtags": [
         {
           "resource-id": "endpoint-cost-pv.ecs",
           "tag": "bb975862fbe3422abf4dae386b132c1d"
         },
         {
           "resource-id": "ane-props",
           "tag": "bf3c8c1819d2421c9a95a9d02af557a3"
         }
       ]
     },
     "property-map": {
       ".ane:AGGR1": {
         "max-reservable-bandwidth": 10000000000,
         "persistent-entity-id": "ane-props.ane:MEC1"
       },
       ".ane:AGGR2": {
         "max-reservable-bandwidth": 15000000000,
         "persistent-entity-id": "ane-props.ane:MEC2"
       },
       ".ane:NET3": {
         "max-reservable-bandwidth": 50000000000
       }
     }
   }
   --example-2

8.5.  Incremental Updates

   In this example, an ALTO client subscribes to the incremental update
   for the multipart Endpoint Cost Service resource "endpoint-cost-pv".

   POST /updates/pv HTTP/1.1
   Host: alto.example.com
   Accept: text/event-stream
   Content-Type: application/alto-updatestreamparams+json
   Content-Length: 120

   {
     "add": {
       "ecspvsub1": {
         "resource-id": "endpoint-cost-pv",
         "input": <ecs-input>
       }
     }
   }

   Based on the server-side process defined in [RFC8895], the ALTO
   server will send the "control-uri" first, using a Server-Sent Event
   (SSE) followed by the full response of the multipart message.

   HTTP/1.1 200 OK
   Connection: keep-alive
   Content-Type: text/event-stream

   event: application/alto-updatestreamcontrol+json
   data: {"control-uri": "https://alto.example.com/updates/streams/123"}

   event: multipart/related;boundary=example-3;
          type=application/alto-endpointcost+json,ecspvsub1
   data: --example-3
   data: Content-ID: <ecsmap@alto.example.com>
   data: Content-Type: application/alto-endpointcost+json
   data:
   data: <endpoint-cost-map-entry>
   data: --example-3
   data: Content-ID: <propmap@alto.example.com>
   data: Content-Type: application/alto-propmap+json
   data:
   data: <property-map-entry>
   data: --example-3--

   When the contents change, the ALTO server will publish the updates
   for each node in this tree separately, based on Section 6.7.3 of
   [RFC8895].

   event: application/merge-patch+json,
      ecspvsub1.ecsmap@alto.example.com
   data: <Merge patch for endpoint-cost-map-update>

   event: application/merge-patch+json,
      ecspvsub1.propmap@alto.example.com
   data: <Merge patch for property-map-update>

8.6.  Multi-Cost

   The following examples demonstrate the request to the "multicost-pv"
   resource and the corresponding response.

   The request asks for two cost types: the first is the Path Vector
   cost type, and the second is a numerical routing cost.  It also
   queries the maximum reservable bandwidth ANE property and the
   persistent entity ID property for two IPv4 source and destination
   pairs (192.0.2.34 -> 192.0.2.2 and 192.0.2.34 -> 192.0.2.50) and one
   IPv6 source and destination pair (2001:db8::3:1 -> 2001:db8::4:1).

   The response consists of two parts:

   *  The first part returns a JSONArray that contains two JSONValue
      entries for each requested source and destination pair: the first
      JSONValue is a JSONArray of ANENames, which is the value of the
      Path Vector cost type; and the second JSONValue is a JSONNumber,
      which is the value of the routing cost.

   *  The second part contains a property map that maps the ANEs to
      their requested properties.

   POST /endpointcost/mcpv HTTP/1.1
   Host: alto.example.com
   Accept: multipart/related;
           type=application/alto-endpointcost+json,
           application/alto-error+json
   Content-Length: 454
   Content-Type: application/alto-endpointcostparams+json

   {
     "multi-cost-types": [
       { "cost-mode": "array", "cost-metric": "ane-path" },
       { "cost-mode": "numerical", "cost-metric": "routingcost" }
     ],
     "endpoints": {
       "srcs": [
         "ipv4:192.0.2.34",
         "ipv6:2001:db8::3:1"
       ],
       "dsts": [
         "ipv4:192.0.2.2",
         "ipv4:192.0.2.50",
         "ipv6:2001:db8::4:1"
       ]
     },
     "ane-property-names": [
       "max-reservable-bandwidth",
       "persistent-entity-id"
     ]
   }

   HTTP/1.1 200 OK
   Content-Length: 1419
   Content-Type: multipart/related; boundary=example-4;
                 type=application/alto-endpointcost+json

   --example-4
   Content-ID: <ecs@alto.example.com>
   Content-Type: application/alto-endpointcost+json

   {
     "meta": {
       "vtags": {
         "resource-id": "endpoint-cost-pv.ecs",
         "tag": "84a4f9c14f9341f0983e3e5f43a371c8"
       },
       "multi-cost-types": [
         { "cost-mode": "array", "cost-metric": "ane-path" },
         { "cost-mode": "numerical", "cost-metric": "routingcost" }
       ]
     },
     "endpoint-cost-map": {
       "ipv4:192.0.2.34": {
         "ipv4:192.0.2.2":   [[ "NET3", "AGGR1" ], 3],
         "ipv4:192.0.2.50":   [[ "NET3", "AGGR2" ], 2]
       },
       "ipv6:2001:db8::3:1": {
         "ipv6:2001:db8::4:1": [[ "NET3", "AGGR2" ], 2]
       }
     }
   }
   --example-4
   Content-ID: <propmap@alto.example.com>
   Content-Type: application/alto-propmap+json

   {
     "meta": {
       "dependent-vtags": [
         {
           "resource-id": "endpoint-cost-pv.ecs",
           "tag": "84a4f9c14f9341f0983e3e5f43a371c8"
         },
         {
           "resource-id": "ane-props",
           "tag": "be157afa031443a187b60bb80a86b233"
         }
       ]
     },
     "property-map": {
       ".ane:AGGR1": {
         "max-reservable-bandwidth": 10000000000,
         "persistent-entity-id": "ane-props.ane:MEC1"
       },
       ".ane:AGGR2": {
         "max-reservable-bandwidth": 15000000000,
         "persistent-entity-id": "ane-props.ane:MEC2"
       },
       ".ane:NET3": {
         "max-reservable-bandwidth": 50000000000
       }
     }
   }
   --example-4

9.  Compatibility with Other ALTO Extensions

9.1.  Compatibility with Legacy ALTO Clients/Servers

   The multipart filtered cost map resource and the multipart Endpoint
   Cost Service resource have no backward-compatibility issues with
   legacy ALTO clients and servers.  Although these two types of
   resources reuse the media types defined in the base ALTO Protocol for
   the "Accept" input parameters, they have different media types for
   responses.  If the ALTO server provides these two types of resources
   but the ALTO client does not support them, the ALTO client will
   ignore the resources without incurring any incompatibility problems.

9.2.  Compatibility with Multi-Cost Extension

   The extension defined in this document is compatible with the multi-
   cost extension [RFC8189].  Such a resource has a media type of either
   "multipart/related; type=application/alto-costmap+json" or
   "multipart/related; type=application/alto-endpointcost+json".  Its
   "cost-constraints" field must be either "false" or not present, and
   the Path Vector cost type must be present in the "cost-type-names"
   capability field but must not be present in the "testable-cost-type-
   names" field, as specified in Sections 7.2.4 and 7.3.4.

9.3.  Compatibility with Incremental Update Extension

   This extension is compatible with the incremental update extension
   [RFC8895].  ALTO clients and servers MUST follow the specifications
   given in Sections 5.2 and 6.7.3 of [RFC8895] to support incremental
   updates for a Path Vector resource.

9.4.  Compatibility with Cost Calendar Extension

   The extension specified in this document is compatible with the Cost
   Calendar extension [RFC8896].  When used together with the Cost
   Calendar extension, the cost value between a source and a destination
   is an array of Path Vectors, where the k-th Path Vector refers to the
   abstract network paths traversed in the k-th time interval by traffic
   from the source to the destination.

   When used with time-varying properties, e.g., maximum reservable
   bandwidth, a property of a single ANE may also have different values
   in different time intervals.  In this case, if such an ANE has
   different property values in two time intervals, it MUST be treated
   as two different ANEs, i.e., with different entity identifiers.
   However, if it has the same property values in two time intervals, it
   MAY use the same identifier.

   This rule allows the Path Vector extension to represent both changes
   of ANEs and changes of the ANEs' properties in a uniform way.  The
   Path Vector part is calendared in a compatible way, and the property
   map part is not affected by the Cost Calendar extension.

   The two extensions combined together can provide the historical
   network correlation information for a set of source and destination
   pairs.  A network broker or client may use this information to derive
   other resource requirements such as Time-Block-Maximum Bandwidth,
   Bandwidth-Sliding-Window, and Time-Bandwidth-Product (TBP) (see
   [SENSE] for details).

10.  General Discussion

10.1.  Constraint Tests for General Cost Types

   The constraint test is a simple approach for querying the data.  It
   allows users to filter query results by specifying some boolean
   tests.  This approach is already used in the ALTO Protocol.  ALTO
   clients are permitted to specify either the "constraints" test
   [RFC7285] [RFC8189] or the "or-constraints" test [RFC8189] to better
   filter the results.

   However, the current syntax can only be used to test scalar cost
   types and cannot easily express constraints on complex cost types,
   e.g., the Path Vector cost type defined in this document.

   In practice, developing a bespoke language for general-purpose
   boolean tests can be a complex undertaking, and it is conceivable
   that such implementations already exist (the authors have not done an
   exhaustive search to determine whether such implementations exist).
   One avenue for developing such a language may be to explore extending
   current query languages like XQuery [XQuery] or JSONiq [JSONiq] and
   integrating these with ALTO.

   Filtering the Path Vector results or developing a more sophisticated
   filtering mechanism is beyond the scope of this document.

10.2.  General Multi-Resource Query

   Querying multiple ALTO information resources continuously is a
   general requirement.  Enabling such a capability, however, must
   address general issues like efficiency and consistency.  The
   incremental update extension [RFC8895] supports submitting multiple
   queries in a single request and allows flexible control over the
   queries.  However, it does not cover the case introduced in this
   document where multiple resources are needed for a single request.

   The extension specified in this document gives an example of using a
   multipart message to encode the responses from two specific ALTO
   information resources: a filtered cost map or an Endpoint Cost
   Service, and a property map.  By packing multiple resources in a
   single response, the implication is that servers may proactively push
   related information resources to clients.

   Thus, it is worth looking into extending the SSE mechanism as used in
   the incremental update extension [RFC8895]; or upgrading to HTTP/2
   [RFC9113] and HTTP/3 [RFC9114], which provides the ability to
   multiplex queries and to allow servers to proactively send related
   information resources.

   Defining a general multi-resource query mechanism is out of scope for
   this document.

11.  Security Considerations

   This document is an extension of the base ALTO Protocol, so the
   security considerations provided for the base ALTO Protocol [RFC7285]
   fully apply when this extension is provided by an ALTO server.

   The Path Vector extension requires additional scrutiny of three
   security considerations discussed in the base protocol:
   confidentiality of ALTO information (Section 15.3 of [RFC7285]),
   potential undesirable guidance from authenticated ALTO information
   (Section 15.2 of [RFC7285]), and availability of ALTO services
   (Section 15.5 of [RFC7285]).

   For confidentiality of ALTO information, a network operator should be
   aware that this extension may introduce a new risk: the Path Vector
   information, when used together with sensitive ANE properties such as
   capacities of bottleneck links, may make network attacks easier.  For
   example, as the Path Vector information may reveal more fine-grained
   internal network structures than the base protocol, an attacker may
   identify the bottleneck link or links and start a distributed denial-
   of-service (DDoS) attack involving minimal flows, triggering in-
   network congestion.  Given the potential risk of leaking sensitive
   information, the Path Vector extension is mainly applicable in
   scenarios where 1) the ANE structures and ANE properties do not
   impose security risks on the ALTO service provider (e.g., they do not
   carry sensitive information) or 2) the ALTO server and client have
   established a reliable trust relationship (e.g., they operate in the
   same administrative domain or are managed by business partners with
   legal contracts).

   Three risk types are identified in Section 15.3.1 of [RFC7285]:

   (1)  excess disclosure of the ALTO service provider's data to an
        unauthorized ALTO client,

   (2)  disclosure of the ALTO service provider's data (e.g., network
        topology information or endpoint addresses) to an unauthorized
        third party, and

   (3)  excess retrieval of the ALTO service provider's data by
        collaborating ALTO clients.

   To mitigate these risks, an ALTO server MUST follow the guidelines in
   Section 15.3.2 of [RFC7285].  Furthermore, an ALTO server MUST follow
   the following additional protections strategies for risk types (1)
   and (3).

   For risk type (1), an ALTO server MUST use the authentication methods
   specified in Section 15.3.2 of [RFC7285] to authenticate the identity
   of an ALTO client and apply access control techniques to restrict the
   retrieval of sensitive Path Vector information by unprivileged ALTO
   clients.  For settings where the ALTO server and client are not in
   the same trust domain, the ALTO server should reach agreements with
   the ALTO client regarding protection of confidentiality before
   granting access to Path Vector services with sensitive information.
   Such agreements may include legal contracts or Digital Rights
   Management (DRM) techniques.  Otherwise, the ALTO server MUST NOT
   offer Path Vector services that carry sensitive information to the
   clients, unless the potential risks are fully assessed and mitigated.

   For risk type (3), an ALTO service provider must be aware that
   persistent ANEs may be used as "landmarks" in collaborative
   inferences.  Thus, they should only be used when exposing public
   service access points (e.g., API gateways, CDN Interconnections) and/
   or when the granularity is coarse grained (e.g., when an ANE
   represents an AS, a data center, or a WAN).  Otherwise, an ALTO
   server MUST use dynamic mappings from ephemeral ANE names to
   underlying physical entities.  Specifically, for the same physical
   entity, an ALTO server SHOULD assign a different ephemeral ANE name
   when the entity appears in the responses to different clients or even
   for different requests from the same client.  A RECOMMENDED
   assignment strategy is to generate ANE names from random numbers.

   Further, to protect the network topology from graph reconstruction
   (e.g., through isomorphic graph identification [BONDY]), the ALTO
   server SHOULD consider protection mechanisms to reduce information
   exposure or obfuscate the real information.  When doing so, the ALTO
   server must be aware that information reduction/obfuscation may lead
   to a potential risk of undesirable guidance from authenticated ALTO
   information (Section 15.2 of [RFC7285]).

   Thus, implementations of ALTO servers involving reduction or
   obfuscation of the Path Vector information SHOULD consider reduction/
   obfuscation mechanisms that can preserve the integrity of ALTO
   information -- for example, by using minimal feasible region
   compression algorithms [NOVA] or obfuscation protocols [RESA]
   [MERCATOR].  However, these obfuscation methods are experimental, and
   their practical applicability to the generic capability provided by
   this extension has not been fully assessed.  The ALTO server MUST
   carefully verify that the deployment scenario satisfies the security
   assumptions of these methods before applying them to protect Path
   Vector services with sensitive network information.

   For availability of ALTO services, an ALTO server should be cognizant
   that using a Path Vector extension might introduce a new risk:
   frequent requests for Path Vectors might consume intolerable amounts
   of server-side computation and storage.  This behavior can break the
   ALTO server.  For example, if an ALTO server implementation
   dynamically computes the Path Vectors for each request, the service
   that provides the Path Vectors may become an entry point for denial-
   of-service attacks on the availability of an ALTO server.

   To mitigate this risk, an ALTO server may consider using such
   optimizations as precomputation-and-projection mechanisms [MERCATOR]
   to reduce the overhead for processing each query.  An ALTO server may
   also protect itself from malicious clients by monitoring client
   behavior and stopping service to clients that exhibit suspicious
   behavior (e.g., sending requests at a high frequency).

   The ALTO service providers must be aware that providing incremental
   updates of "max-reservable-bandwidth" may provide information about
   other consumers of the network.  For example, a change in value may
   indicate that one or more reservations have been made or changed.  To
   mitigate this risk, an ALTO server can batch the updates and/or add a
   random delay before publishing the updates.

12.  IANA Considerations

12.1.  "ALTO Cost Metrics" Registry

   This document registers a new entry in the "ALTO Cost Metrics"
   registry, per Section 14.2 of [RFC7285].  The new entry is as shown
   below in Table 1.

              +============+====================+===========+
              | Identifier | Intended Semantics | Reference |
              +============+====================+===========+
              | ane-path   | See Section 6.5.1  | RFC 9275  |
              +------------+--------------------+-----------+

                   Table 1: "ALTO Cost Metrics" Registry

12.2.  "ALTO Cost Modes" Registry

   This document registers a new entry in the "ALTO Cost Modes"
   registry, per Section 5 of [RFC9274].  The new entry is as shown
   below in Table 2.

    +============+=========================+=============+===========+
    | Identifier | Description             | Intended    | Reference |
    |            |                         | Semantics   |           |
    +============+=========================+=============+===========+
    | array      | Indicates that the cost | See Section | RFC 9275  |
    |            | value is a JSON array   | 6.5.2       |           |
    +------------+-------------------------+-------------+-----------+

                   Table 2: "ALTO Cost Modes" Registry

12.3.  "ALTO Entity Domain Types" Registry

   This document registers a new entry in the "ALTO Entity Domain Types"
   registry, per Section 12.3 of [RFC9240].  The new entry is as shown
   below in Table 3.

   +============+============+=============+===================+=======+
   | Identifier |Entity      |Hierarchy and| Media Type of     |Mapping|
   |            |Identifier  |Inheritance  | Defining Resource |to ALTO|
   |            |Encoding    |             |                   |Address|
   |            |            |             |                   |Type   |
   +============+============+=============+===================+=======+
   | ane        |See Section |None         | application/alto- |false  |
   |            |6.2.2       |             | propmap+json      |       |
   +------------+------------+-------------+-------------------+-------+

                Table 3: "ALTO Entity Domain Types" Registry

   Identifier:  See Section 6.2.1.

   Entity Identifier Encoding:  See Section 6.2.2.

   Hierarchy:  None

   Inheritance:  None

   Media Type of Defining Resource:  See Section 6.2.4.

   Mapping to ALTO Address Type:  This entity type does not map to an
      ALTO address type.

   Security Considerations:  In some usage scenarios, ANE addresses
      carried in ALTO Protocol messages may reveal information about an
      ALTO client or an ALTO service provider.  If a naming schema is
      used to generate ANE names, either used privately or standardized
      by a future extension, how (or if) the naming schema relates to
      private information and network proximity must be explained to
      ALTO implementers and service providers.

12.4.  "ALTO Entity Property Types" Registry

   Two initial entries -- "max-reservable-bandwidth" and "persistent-
   entity-id" -- are registered for the ALTO domain "ane" in the "ALTO
   Entity Property Types" registry, per Section 12.4 of [RFC9240].  The
   two new entries are shown below in Table 4, and their details can be
   found in Sections 12.4.1 and 12.4.2 of this document.

   +==========================+====================+===================+
   | Identifier               | Intended           | Media Type of     |
   |                          | Semantics          | Defining Resource |
   +==========================+====================+===================+
   | max-reservable-bandwidth | See Section        | application/alto- |
   |                          | 6.4.1              | propmap+json      |
   +--------------------------+--------------------+-------------------+
   | persistent-entity-id     | See Section        | application/alto- |
   |                          | 6.4.2              | propmap+json      |
   +--------------------------+--------------------+-------------------+

     Table 4: Initial Entries for the "ane" Domain in the "ALTO Entity
                          Property Types" Registry

12.4.1.  New ANE Property Type: Maximum Reservable Bandwidth

   Identifier:  "max-reservable-bandwidth"

   Intended Semantics:  See Section 6.4.1.

   Media Type of Defining Resource:  application/alto-propmap+json

   Security Considerations:  To make better choices regarding bandwidth
      reservation, this property is essential for applications such as
      large-scale data transfers or an overlay network interconnection.
      It may reveal the bandwidth usage of the underlying network and
      can potentially be leveraged to reduce the cost of conducting
      denial-of-service attacks.  Thus, the ALTO server MUST consider
      such protection mechanisms as providing the information to
      authorized clients only and applying information reduction and
      obfuscation as discussed in Section 11.

12.4.2.  New ANE Property Type: Persistent Entity ID

   Identifier:  "persistent-entity-id"

   Intended Semantics:  See Section 6.4.2.

   Media Type of Defining Resource:  application/alto-propmap+json

   Security Considerations:  This property is useful when an ALTO server
      wants to selectively expose certain service points whose detailed
      properties can be further queried by applications.  As mentioned
      in Section 12.3.2 of [RFC9240], the entity IDs may reveal
      sensitive information about the underlying network.  An ALTO
      server should follow the security considerations provided in
      Section 11 of [RFC9240].

13.  References

13.1.  Normative References

   [RFC2046]  Freed, N. and N. Borenstein, "Multipurpose Internet Mail
              Extensions (MIME) Part Two: Media Types", RFC 2046,
              DOI 10.17487/RFC2046, November 1996,
              <https://www.rfc-editor.org/info/rfc2046>.

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119,
              DOI 10.17487/RFC2119, March 1997,
              <https://www.rfc-editor.org/info/rfc2119>.

   [RFC2387]  Levinson, E., "The MIME Multipart/Related Content-type",
              RFC 2387, DOI 10.17487/RFC2387, August 1998,
              <https://www.rfc-editor.org/info/rfc2387>.

   [RFC5322]  Resnick, P., Ed., "Internet Message Format", RFC 5322,
              DOI 10.17487/RFC5322, October 2008,
              <https://www.rfc-editor.org/info/rfc5322>.

   [RFC7285]  Alimi, R., Ed., Penno, R., Ed., Yang, Y., Ed., Kiesel, S.,
              Previdi, S., Roome, W., Shalunov, S., and R. Woundy,
              "Application-Layer Traffic Optimization (ALTO) Protocol",
              RFC 7285, DOI 10.17487/RFC7285, September 2014,
              <https://www.rfc-editor.org/info/rfc7285>.

   [RFC8174]  Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
              2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
              May 2017, <https://www.rfc-editor.org/info/rfc8174>.

   [RFC8189]  Randriamasy, S., Roome, W., and N. Schwan, "Multi-Cost
              Application-Layer Traffic Optimization (ALTO)", RFC 8189,
              DOI 10.17487/RFC8189, October 2017,
              <https://www.rfc-editor.org/info/rfc8189>.

   [RFC8895]  Roome, W. and Y. Yang, "Application-Layer Traffic
              Optimization (ALTO) Incremental Updates Using Server-Sent
              Events (SSE)", RFC 8895, DOI 10.17487/RFC8895, November
              2020, <https://www.rfc-editor.org/info/rfc8895>.

   [RFC8896]  Randriamasy, S., Yang, R., Wu, Q., Deng, L., and N.
              Schwan, "Application-Layer Traffic Optimization (ALTO)
              Cost Calendar", RFC 8896, DOI 10.17487/RFC8896, November
              2020, <https://www.rfc-editor.org/info/rfc8896>.

   [RFC9240]  Roome, W., Randriamasy, S., Yang, Y., Zhang, J., and K.
              Gao, "An Extension for Application-Layer Traffic
              Optimization (ALTO): Entity Property Maps", RFC 9240,
              DOI 10.17487/RFC9240, July 2022,
              <https://www.rfc-editor.org/info/rfc9240>.

   [RFC9274]  Boucadair, M. and Q. Wu, "A Cost Mode Registry for the
              Application-Layer Traffic Optimization (ALTO) Protocol",
              RFC 9274, DOI 10.17487/RFC9274, July 2022,
              <https://www.rfc-editor.org/info/rfc9274>.

13.2.  Informative References

   [ALTO-PERF-METRICS]
              Wu, Q., Yang, Y., Lee, Y., Dhody, D., Randriamasy, S., and
              L. Contreras, "ALTO Performance Cost Metrics", Work in
              Progress, Internet-Draft, draft-ietf-alto-performance-
              metrics-28, 21 March 2022,
              <https://datatracker.ietf.org/doc/html/draft-ietf-alto-
              performance-metrics-28>.

   [BONDY]    Bondy, J.A. and R.L. Hemminger, "Graph reconstruction--a
              survey", Journal of Graph Theory, Volume 1, Issue 3, pp.
              227-268, DOI 10.1002/jgt.3190010306, 1977,
              <https://onlinelibrary.wiley.com/doi/10.1002/
              jgt.3190010306>.

   [BOXOPT]   Xiang, Q., Yu, H., Aspnes, J., Le, F., Kong, L., and Y.R.
              Yang, "Optimizing in the Dark: Learning an Optimal
              Solution through a Simple Request Interface", Proceedings
              of the AAAI Conference on Artificial Intelligence 33,
              1674-1681, DOI 10.1609/aaai.v33i01.33011674, July 2019,
              <https://ojs.aaai.org//index.php/AAAI/article/view/3984>.

   [CLARINET] Viswanathan, R., Ananthanarayanan, G., and A. Akella,
              "CLARINET: WAN-aware optimization for analytics queries",
              Proceedings of the 12th USENIX conference on Operating
              Systems Design and Implementation (OSDI'16), Savannah, GA,
              pp. 435-450, November 2016,
              <https://dl.acm.org/doi/abs/10.5555/3026877.3026911>.

   [G2]       Ros-Giralt, J., Bohara, A., Yellamraju, S., Langston,
              M.H., Lethin, R., Jiang, Y., Tassiulas, L., Li, J., Tan,
              Y., and M. Veeraraghavan, "On the Bottleneck Structure of
              Congestion-Controlled Networks", Proceedings of the ACM on
              Measurement and Analysis of Computing Systems, Volume 3,
              Issue 3, pp. 1-31, DOI 10.1145/3366707, December 2019,
              <https://dl.acm.org/doi/10.1145/3366707>.

   [HUG]      Chowdhury, M., Liu, Z., Ghodsi, A., and I. Stoica, "HUG:
              multi-resource fairness for correlated and elastic
              demands", Proceedings of the 13th USENIX Conference on
              Networked Systems Design and Implementation (NSDI'16),
              Santa Clara, CA, pp. 407-424, March 2016,
              <https://dl.acm.org/doi/10.5555/2930611.2930638>.

   [INTENT-BASED-NETWORKING]
              Clemm, A., Ciavaglia, L., Granville, L. Z., and J.
              Tantsura, "Intent-Based Networking - Concepts and
              Definitions", Work in Progress, Internet-Draft, draft-
              irtf-nmrg-ibn-concepts-definitions-09, 24 March 2022,
              <https://datatracker.ietf.org/doc/html/draft-irtf-nmrg-
              ibn-concepts-definitions-09>.

   [JSONiq]   JSONiq, "The JSON Query Language", 2022,
              <https://www.jsoniq.org/>.

   [MERCATOR] Xiang, Q., Zhang, J., Wang, X., Liu, Y., Guok, C., Le, F.,
              MacAuley, J., Newman, H., and Y.R. Yang, "Toward Fine-
              Grained, Privacy-Preserving, Efficient Multi-Domain
              Network Resource Discovery", IEEE/ACM, IEEE Journal on
              Selected Areas in Communications, Volume 37, Issue 8, pp.
              1924-1940, DOI 10.1109/JSAC.2019.2927073, August 2019,
              <https://ieeexplore.ieee.org/document/8756056>.

   [MOWIE]    Zhang, Y., Li, G., Xiong, C., Lei, Y., Huang, W., Han, Y.,
              Walid, A., Yang, Y.R., and Z. Zhang, "MoWIE: Toward
              Systematic, Adaptive Network Information Exposure as an
              Enabling Technique for Cloud-Based Applications over 5G
              and Beyond", Proceedings of the Workshop on Network
              Application Integration/CoDesign (NAI '20), ACM, Virtual
              Event USA, pp. 20-27, DOI 10.1145/3405672.3409489, August
              2020, <https://dl.acm.org/doi/10.1145/3405672.3409489>.

   [NOVA]     Gao, K., Xiang, Q., Wang, X., Yang, Y.R., and J. Bi, "An
              Objective-Driven On-Demand Network Abstraction for
              Adaptive Applications", IEEE/ACM Transactions on
              Networking (TON) Vol. 27, Issue 2, pp. 805-818,
              DOI 10.1109/TNET.2019.2899905, April 2019,
              <https://doi.org/10.1109/TNET.2019.2899905>.

   [RESA]     Xiang, Q., Zhang, J., Wang, X., Liu, Y., Guok, C., Le, F.,
              MacAuley, J., Newman, H., and Y.R. Yang, "Fine-Grained,
              Multi-Domain Network Resource Abstraction as a Fundamental
              Primitive to Enable High-Performance, Collaborative Data
              Sciences", SC18: International Conference for High
              Performance Computing, Networking, Storage and Analysis,
              pp. 58-70, DOI 10.1109/SC.2018.00008, November 2018,
              <https://ieeexplore.ieee.org/document/8665783>.

   [RFC2216]  Shenker, S. and J. Wroclawski, "Network Element Service
              Specification Template", RFC 2216, DOI 10.17487/RFC2216,
              September 1997, <https://www.rfc-editor.org/info/rfc2216>.

   [RFC4271]  Rekhter, Y., Ed., Li, T., Ed., and S. Hares, Ed., "A
              Border Gateway Protocol 4 (BGP-4)", RFC 4271,
              DOI 10.17487/RFC4271, January 2006,
              <https://www.rfc-editor.org/info/rfc4271>.

   [RFC9113]  Thomson, M., Ed. and C. Benfield, Ed., "HTTP/2", RFC 9113,
              DOI 10.17487/RFC9113, June 2022,
              <https://www.rfc-editor.org/info/rfc9113>.

   [RFC9114]  Bishop, M., Ed., "HTTP/3", RFC 9114, DOI 10.17487/RFC9114,
              June 2022, <https://www.rfc-editor.org/info/rfc9114>.

   [SENSE]    ESnet, "Software Defined Networking (SDN) for End-to-End
              Networked Science at the Exascale", 2019,
              <https://www.es.net/network-r-and-d/sense/>.

   [SEREDGE]  Contreras, L., Baliosian, J., Martínez-Julia, P., and J.
              Serrat, "Computing at the Edge: But, what Edge?",
              Proceedings of NOMS 2020 - 2020 IEEE/IFIP Network
              Operations and Management Symposium, pp. 1-9,
              DOI 10.1109/NOMS47738.2020.9110342, April 2020,
              <https://ieeexplore.ieee.org/document/9110342>.

   [SWAN]     Hong, C., Kandula, S., Mahajan, R., Zhang, M., Gill, V.,
              Nanduri, M., and R. Wattenhofer, "Achieving high
              utilization with software-driven WAN", Proceedings of the
              ACM SIGCOMM 2013 conference on SIGCOMM (SIGCOMM '13), New
              York, NY, pp. 15-26, DOI 10.1145/2486001.2486012, August
              2013, <https://dl.acm.org/doi/10.1145/2486001.2486012>.

   [UNICORN]  Xiang, Q., Wang, T., Zhang, J., Newman, H., Yang, Y.R.,
              and Y. Liu, "Unicorn: Unified resource orchestration for
              multi-domain, geo-distributed data analytics", Future
              Generation Computer Systems, Volume 93, pp. 188-197,
              DOI 10.1016/j.future.2018.09.048, April 2019,
              <https://www.sciencedirect.com/science/article/abs/pii/
              S0167739X18302413?via%3Dihub>.

   [XQuery]   Robie, J., Ed., Dyck, M., Ed., and J. Spiegel, Ed.,
              "XQuery 3.1: An XML Query Language", W3C Recommendation,
              March 2017, <https://www.w3.org/TR/xquery-31/>.

Acknowledgments

   The authors would like to thank Andreas Voellmy, Erran Li, Haibin
   Song, Haizhou Du, Jiayuan Hu, Tianyuan Liu, Xiao Shi, Xin Wang, and
   Yan Luo for fruitful discussions.  The authors thank Greg Bernstein,
   Dawn Chen, Wendy Roome, and Michael Scharf for their contributions to
   earlier draft versions of this document.

   The authors would also like to thank Tim Chown, Luis Contreras, Roman
   Danyliw, Benjamin Kaduk, Erik Kline, Suresh Krishnan, Murray
   Kucherawy, Warren Kumari, Danny Lachos, Francesca Palombini, Éric
   Vyncke, Samuel Weiler, and Qiao Xiang, whose feedback and suggestions
   were invaluable for improving the practicability and conciseness of
   this document; and Mohamed Boucadair, Martin Duke, Vijay Gurbani, Jan
   Seedorf, and Qin Wu, who provided great support and guidance.

Authors' Addresses

   Kai Gao
   Sichuan University
   No.24 South Section 1, Yihuan Road
   Chengdu
   610000
   China
   Email: kaigao@scu.edu.cn


   Young Lee
   Samsung
   Republic of Korea
   Email: younglee.tx@gmail.com


   Sabine Randriamasy
   Nokia Bell Labs
   Route de Villejust
   91460 Nozay
   France
   Email: sabine.randriamasy@nokia-bell-labs.com


   Yang Richard Yang
   Yale University
   51 Prospect Street
   New Haven, CT 06511
   United States of America
   Email: yry@cs.yale.edu


   Jingxuan Jensen Zhang
   Tongji University
   4800 Caoan Road
   Shanghai
   201804
   China
   Email: jingxuan.n.zhang@gmail.com