]> Sichuan University

No.24 South Section 1, Yihuan Road Chengdu 610000 China kaigao@scu.edu.cn

Deutsche Telekom

Ida-Rhodes-Straße 2 Darmstadt 64295 Germany Roland.Schott@telekom.de

Yale University

51 Prospect Street New Haven CT USA yry@cs.yale.edu

Yale University

51 Prospect Street New Haven 3408 USA lauren.delwiche@yale.edu

Yale University

51 Prospect Street New Haven 3408 USA lachlan.keller@yale.edu

Transport Area ALTO The ALTO Protocol (RFC 7285) leverages HTTP/1.1 and is designed for the simple, sequential request-reply use case, in which an ALTO client requests a sequence of information resources and the server responds with the complete content of each resource one at a time. ALTO incremental updates using Server-Sent Events (SSE) (RFC 8895) defines a multiplexing protocol on top of HTTP/1.x, so that an ALTO server can incrementally push resource updates to clients whenever monitored network information resources change, allowing the clients to monitor multiple resources at the same time. However, HTTP/2 and later versions already support concurrent, non-blocking transport of multiple streams in the same HTTP connection. To take advantage of newer HTTP features, this document introduces the ALTO Transport Information Publication Service (TIPS). TIPS uses an incremental RESTful design to give an ALTO client the new capability to explicitly, concurrently (non-blocking) request (pull) specific incremental updates using native HTTP/2 or HTTP/3, while still functioning for HTTP/1.1. Discussion Venues Discussion of this document takes place on the Application-Layer Traffic Optimization Working Group mailing list (alto@ietf.org), which is archived at . Source for this draft and an issue tracker can be found at .

Introduction Application-Layer Traffic Optimization (ALTO) provides means for network applications to obtain network status information. So far, two transport protocols have been designed:

1. The ALTO base protocol , which is designed for the simple use case in which an ALTO client requests a network information resource, and the server sends the complete content of the requested information (if any) resource to the client.
2. ALTO incremental updates using Server-Sent Events (ALTO/SSE) , which is designed for an ALTO client to indicate to the server that it wants to receive updates for a set of resources and the server can then concurrently and incrementally push updates to that client whenever monitored resources change.

Both protocols are designed for HTTP/1.1 , but HTTP/2 and HTTP/3 can support HTTP/1.1 workflows. However, HTTP/2 and HTTP/3 provide features that can improve certain properties of ALTO and ALTO/SSE.

• First, consider the ALTO base protocol, which is designed to transfer only complete information resources. A client can run the base protocol on top of HTTP/2 or HTTP/3 to request multiple information resources in concurrent streams, but each request must be for a complete information resource: there is no capability it transmits incremental updates. Hence, there can be a large overhead when the client already has an information resource and then there are small changes to the resource.
• Next, consider ALTO/SSE . Although ALTO/SSE can transfer incremental updates, it introduces a customized multiplexing protocol on top of HTTP, assuming a total-order message channel from the server to the client. The multiplexing design does not provide naming (i.e., a resource identifier) to individual incremental updates. Such a design cannot use concurrent data streams available in HTTP/2 and HTTP/3, because both cases require a resource identifier. Additionally, ALTO/SSE is a push-only protocol, which denies the client flexibility in choosing how and when it receives updates.

To mitigate these concerns, this document introduces a new ALTO service called the Transport Information Publication Service (TIPS). TIPS uses an incremental RESTful design to provide an ALTO client with a new capability to explicitly, concurrently issue non-blocking requests for specific incremental updates using native HTTP/2 or HTTP/3, while still functioning for HTTP/1.1. Despite the benefits, however, ALTO/SSE , which solves a similar problem, has its advantages. First, SSE is a mature technique with a well-established ecosystem that can simplify development. Second, SSE does not allow multiple connections to receive updates for multiple objects over HTTP/1.1. HTTP/2 and HTTP/3 also specify server push, which might enhance TIPS. We discuss push-mode TIPS as an alternative design in . Specifically, this document specifies:

• Extensions to the ALTO Protocol for dynamic subscription and efficient uniform update delivery of an incrementally changing network information resource.
• A new resource type that indicates the TIPS updates graph model for a resource.
• URI patterns to fetch the snapshots or incremental updates.

discusses to what extent the TIPS design adheres to the Best Current Practices for building protocols with HTTP .

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 when, and only when, they appear in all capitals, as shown here.

Notations This document uses the same syntax and notations as introduced in to specify the extensions to existing ALTO resources and services.

TIPS Overview

Transport Requirements The ALTO Protocol and its extensions support two transport mechanisms: First, a client can directly request an ALTO resource and obtain a complete snapshot of that ALTO resource, as specified in the base protocol ; Second, a client can subscribe to incremental changes of one or multiple ALTO resources using the incremental update extension , and a server pushes the updates to the client through Server Sent Events (SSE). However, the current transport mechanisms are not optimized for storing, transmitting, and processing (incremental) updates of ALTO information resources. Specifically, the new transport mechanism must satisfy the following requirements:

Incremental updates:

Incremental updates can reduce both the data storage on an ALTO server and the transmission time of the updates, especially when the change of an ALTO resource is minor. The base protocol does not support incremental updates and the current incremental update mechanism in has limitations (as discussed below).

Concurrent, non-blocking update transmission:

When a client needs to receive and apply multiple incremental updates, it is desired to transmit the updates concurrently to fully utilize the bandwidth and to reduce head-of-line blocking. The ALTO incremental update extension , unfortunately, does not satisfy this requirement -- even though the updates can be multiplexed by the server to avoid head-of-line blocking between multiple resources, the updates are delivered sequentially and can suffer from head-of-line blocking inside the connection, for example, when there is a packet loss.

Prefetching updates:

Prefetching updates can reduce the time to send the request, making it possible to achieve sub-RTT transmission of ALTO incremental updates. In , this requirement is fulfilled using server-sent event (SSE) and is still desired in the ALTO new transport.

Backward compatibility:

While some of the previous requirements are offered by HTTP/2 and HTTP/3 , it is desired that the ALTO new transport mechanism can work with HTTP/1.1 as many development tools and current ALTO implementations are based on HTTP/1.1.

The ALTO new transport specified in this document satisfies all the design requirements and hence improves the efficiency of continuous dissemination of ALTO information. The key idea is to introduce a unified data model to describe the changes (snapshots and incremental updates) of an ALTO resource, referred to as a TIPS view. Along with the data model, this document also specifies a unified naming for the snapshots and incremental updates, independent of the HTTP version. Thus, these updates can be concurrently requested. Prefetching is realized using long polling. This document assumes the deployment model discussed in .

TIPS Terminology In addition to the terms defined in , this document uses the following terms:

Transport Information Publication Service (TIPS):

Is a new type of ALTO service, as specified in this document, to enable a uniform transport mechanism for updates of an incrementally changing ALTO network information resource.

Network information resource:

Is a piece of retrievable information about network state, per .

TIPS view (tv):

Is defined in this document to be the container of incremental transport information about the network information resource. The TIPS view has one basic component, updates graph (ug), but may include other transport information.

Updates graph (ug):

Is a directed, acyclic graph whose nodes represent the set of versions of an information resource, and edges the set of update items to compute these versions. An ALTO map service (e.g., Cost Map, Network Map) may need only a single updates graph. A dynamic network information service (e.g., Filtered Cost Map) may create an updates graph (within a new TIPS view) for each unique request.

Version:

Represents a historical content of an information resource. For an information resource, each version is associated with and uniquely identified by a monotonically and consecutively increased sequence number. This document uses the term "version s" to refer to the version associated with sequence number "s".

Start sequence number (start-seq):

Is the smallest non-zero sequence number in an updates graph.

End sequence number (end-seq):

Is the largest sequence number in an updates graph.

Snapshot:

Is a full replacement of a resource and is contained within an updates graph.

Incremental update:

Is a partial replacement of a resource contained within an updates graph, codified in this document as a JSON Merge Patch or JSON Patch. An incremental update is mandatory if the source version (i) and target version (j) are consecutive, i.e., i + 1 = j, and optional or a shortcut otherwise. Mandatory incremental updates are always in an updates graph, while optional/shortcut incremental updates may or may not be included in an updates graph.

Update item:

Refers to the content on an edge of the updates graph, which can be either a snapshot or an incremental update. An update item can be considered as a pair (op, data) where op denotes whether the item is an incremental update or a snapshot, and data is the content of the item.

ID#i-#j:

Denotes the update item on a specific edge in the updates graph to transition from version i to version j, where i and j are the sequence numbers of the source node and the target node of the edge, respectively.

Overview of ALTO TIPS

shows an example illustrating an overview of the ALTO TIPS service. The server provides the TIPS service of two information resources (#1 and #2) where #1 is an ALTO map service, and #2 is a filterable service. There are 3 ALTO clients (Client 1, Client 2, and Client 3) that are connected to the ALTO server. Each client uses the TIPS view to retrieve updates. Specifically, a TIPS view (tv1) is created for the map service #1, and is shared by multiple clients. For the filtering service #2, two different TIPS views (tv2 and tv3) are created upon different client requests with different filter sets. A heartbeat mechanism is used to detect the liveness of clients using TIPS. It allows the ALTO server to identify unused TIPS views, e.g., because of accidentally crashed ALTO clients, and release the resources of such views.

TIPS Updates Graph In order to provide incremental updates for a resource, an ALTO server creates an updates graph, which is a directed, acyclic graph that contains a sequence of incremental updates and snapshots (collectively called update items) of a network information resource.

Basic Data Model of Updates Graph For each resource (e.g., a cost map, a network map), the incremental updates and snapshots can be represented using the following directed acyclic graph model, where the server tracks the change of the resource maps with version IDs that are assigned sequentially (i.e., incremented by 1 each time):

• Each node in the graph is a version of the resource, where a tag identifies the content of the version (tag is valid only within the scope of resource). Version 0 is reserved as the initial state (empty/null).
• Each edge is an update item. In particular, the edge from i to j is the update item to transit from version i to version j.
• Version is path-independent (different paths arrive at the same version/node has the same content)

A concrete example is shown in . There are 7 nodes in the graph, representing 7 different versions of the resource. Edges in the figure represent the updates from the source version to the target version. Thick lines represent mandatory incremental updates (e.g., ID103-104), dotted lines represent optional incremental updates (e.g., ID103-105), and thin lines represent snapshots (e.g., ID0-103). Note that node content is path independent: the content of node v can be obtained by applying the updates from any path that ends at v. For example, assume the latest version is 105 and a client already has version 103. The base version of the client is 103 as it serves as a base upon which incremental updates can be applied. The target version 105 can either be directly fetched as a snapshot, computed incrementally by applying the incremental updates between 103 and 104, then 104 and 105, or if the optional update from 103 to 105 exists, computed incrementally by taking the "shortcut" path from 103 to 105.

TIPS Model Example | 101 | | | +======+ | | ID101-102 || | | \/ | | +======+ | | tag: 6431234 -> | 102 | | | +======+ | | ID102-103 || | | \/ | | +======+ / | +--------------+ tag: 0881080 -> | 103 |<--------/ | | Base Version | =======> +======+ ID0-103 | +--------------+ 103-104 || .. | \/ .. | +======+ .. | tag: 6452654 -> | 104 | .. ID103 | +======+ .. -105 | ID104-105 || .. | ID0-105 \/ |._ / +======+ / tag: 7838392 -> | 105 |<-----------/ +======+ ID105-106 || \/ +======+ tag: 6470983 -> | 106 | +======+ ]]>

Updates Graph Modification Invariants A server may change its updates graph (to compact, to add nodes, etc.), but it must ensure that any resource state that it makes available is reachable by clients, either directly via a snapshot (that is, relative to 0) or indirectly by requesting an earlier snapshot and a contiguous set of incremental updates. Additionally, to allow clients to proactively construct URIs for future update items, the ID of each added node in the updates graph must increment contiguously by 1. More specifically, the updates graph MUST satisfy the following invariants:

• Continuity: At any time, let ns denote the smallest non-zero version (i.e., start-seq) in the update graph and ne denote the latest version (i.e., end-seq). Then any version in between ns and ne must also exist. This implies that the incremental update from ni to ni + 1 exists for any ns <= ni <= ne, and all versions in the update graph (except 0) is an integer interval [ns, ne].
• Feasibility: Let ns denote the start-seq in the update graph. The server must provide a snapshot of ns and, in other words, there is always a direct link to ns in the update graph.
• "Right shift" only: Assume a server provides versions in [n1, n2] at time t and versions in [n1', n2'] at time t'. If t' > t, then n1' >= n1 and n2' >= n2.

For example, consider the case that a server compacts a resource's updates graph to conserve space, using the example model in . Assume at time 0, the server provides the versions {101, 102, 103, 104, 105, 106}. At time 1, both {103, 104, 105, 106} and {105, 106} are valid sets. However, {102, 103, 104, 105, 106} and {104, 105, 106} are not valid sets as there is no snapshot to version 102 or 104 in the update graph. Thus, there is a risk that the right content of version 102 (in the first example) or 104 (in the second example) cannot be obtained by a client that does not have the previous version 101 or 103, respectively.

TIPS Workflow and Resource Location Schema

Workflow At a high level, an ALTO client first uses the TIPS service (denoted as TIPS-F and F is for frontend) to indicate the information resource(s) that the client wants to monitor. For each requested resource, the server returns a JSON object that contains a URI, which points to the root of a TIPS view (denoted as TIPS-V), and a summary of the current view, which contains the information to correctly interact with the current view. With the URI to the root of a TIPS view, clients can construct URIs (see ) to fetch incremental updates and to report liveness. An example workflow is shown in . After the TIPS-F service receives the request from the client to monitor the updates of an ALTO resource, it creates a TIPS view service and returns the corresponding information to the client. The URI points to that specific TIPS-V instance and the summary contains a heartbeat interval, the start-seq and end-seq of the update graph, and a server-recommended edge to consume first, e.g., from i to j. An ALTO client can then continuously pull each additional update with the information. For example, the client in first fetches the update from i to j, and then from j to j+1. Note that the update item at <tips-view-uri>/ug/<j>/<j+1> may not yet exist, so the server holds the request until the update becomes available (long polling). In the meantime, the client must periodically send a heartbeat signal by making a POST request to /<tips-view-path>/hb to indicate that the client is still active. Otherwise, the server may close the TIPS view to release the resources. See details in . Once the client no longer needs the TIPS view, it can send a DELETE request to the TIPS-V service.

ALTO TIPS Workflow Supporting Client Pull |.-.-.-.-.-.-.-> | | , . | | <-------------------------------------| <-.-.-.-.-.-.-.| | . | GET //ug// . |------------------------------------------------------> | | content on edge i to j | | <------------------------------------------------------| | . | GET //ug// . |------------------------------------------------------> | | | | POST //hb | |------------------------------------------------------> | | | | content on edge j to j+1 | | <------------------------------------------------------| | . | DELETE . |------------------------------------------------------> | | o o ]]>

Resource Location Schema The resource location schema defines how a client constructs URI to fetch incremental updates and to report liveness through heartbeat requests. To access each update in an updates graph, consider the model represented as a "virtual" file system (adjacency list), contained within the root of a TIPS view URI (see for the definition of tips-view-uri). For example, assuming that the update graph of a TIPS view is as shown in , the location schema of this TIPS view will have the format as in .

Location Schema Example // root path to a TIPS view |_ ug // updates graph | |_ 0 | | |_ 101 // full 101 snapshot | | |_ 103 | | \_ 105 | |_ 101 | | \_ 102 // 101 -> 102 incremental update | |_ 102 | | \_ 103 | |_ 103 | | |_ 104 | | \_ 105 // optional shortcut 103 -> 105 incr. update | |_ 104 | | \_ 105 | \_ 105 | \_ 106 |_ hb // heartbeat \_ ... ]]>

TIPS uses this directory schema to generate template URIs which allow clients to construct the location of incremental updates after receiving the tips-view-uri from the server. The generic template for the location of the update item on the edge from node 'i' to node 'j' in the updates graph is: /ug// ]]> Due to the sequential nature of the update item IDs, a client can long poll a future update that does not yet exist (e.g., the incremental update from 106 to 107) by constructing the URI for the next edge that will be added, starting from the sequence number of the current last node (denoted as end-seq) in the graph to the next sequential node (with the sequence number of end-seq + 1): /ug// ]]> Incremental updates of a TIPS view are read-only. Thus, they are fetched using the HTTP GET method. In the meantime, the URI of the heartbeat service of a TIPS view has the following format: /hb ]]> To avoid the message hitting an HTTP cache, the heartbeat request must use the POST method.

TIPS Information Resource Directory (IRD) Announcement To announce a TIPS information resource in the information resource directory (IRD), an ALTO server MUST specify the "media-type", "capabilities" and "uses" as follows.

Media Type The media type of the Transport Information Publication Service resource is "application/alto-tips+json".

Capabilities The capabilities field of TIPS is modeled on that defined in Section 6.3 of . Specifically, the capabilities are defined as an object of type TIPSCapabilities:

TIPSCapabilities String; } IncrementalUpdateMediaTypes; ]]>

with field:

incremental-change-media-types:

If a TIPS can provide updates with incremental changes for a resource, the "incremental-change-media-types" field has an entry for that resource-id, and the value is the supported media types of incremental changes, separated by commas. For the implementation of this specification, this MUST be "application/ merge-patch+json", "application/json-patch+json", or "application/ merge-patch+json,application/json-patch+json", unless defined by a future extension. When choosing the media types to encode incremental updates for a resource, the server MUST consider the limitations of the encoding. For example, when a JSON merge patch specifies that the value of a field is null, its semantics are that the field is removed from the target and hence the field is no longer defined (i.e., undefined). This, however, may not be the intended result for the resource, when null and undefined have different semantics for the resource. In such a case, the server MUST choose JSON patch over JSON merge patch if JSON patch is indicated as a capability of the TIPS. If the server does not support JSON patch to handle such a case, the server then needs to send a full replacement.

Uses The "uses" attribute MUST be an array with the resource-ids of every network information resource for which this TIPS can provide service. This set may be any subset of the ALTO server's network information resources and may include resources defined in linked IRDs. However, it is RECOMMENDED that the ALTO server selects a set that is closed under the resource dependency relationship. That is, if a TIPS' "uses" set includes resource R1 and resource R1 depends on ("uses") resource R0, then the TIPS' "uses" set SHOULD include R0 as well as R1. For example, if a TIPS provides a TIPS view for a cost map, it SHOULD also provide a TIPS view for the network map upon which that cost map depends. If the set is not closed, at least one resource R1 in the "uses" field of a TIPS depends on another resource R0 which is not in the "uses" field of the same TIPS. Thus, a client cannot receive incremental updates for R0 from the same TIPS service. If the client observes in an update of R1 that the version tag for R0 has changed, it must request the full content of R0, which is likely to be less efficient than receiving the incremental updates of R0.

An Example Extending the IRD example in Section 8.1 of , is the IRD of an ALTO server supporting ALTO base protocol, ALTO/SSE, and ALTO TIPS.

Example of an ALTO Server Supporting ALTO Base Protocol, ALTO/SSE, and ALTO TIPS

Note that it is straightforward for an ALTO server to run HTTP/2 and support concurrent retrieval of multiple resources such as "my- network-map" and "my-routingcost-map" using multiple HTTP/2 streams. The resource "update-my-costs-tips" provides an ALTO TIPS service, and this is indicated by the media-type "application/ alto-tips+json".

TIPS Management Upon request, a server sends a TIPS view to a client. This TIPS view may be created at the time of the request or may already exist (either because another client has already created a TIPS view for the same requested network resource or because the server perpetually maintains a TIPS view for an often-requested resource). The server MAY keep track of which clients are subscribing to each TIPS view to determine whether or not it should delete a TIPS view and its corresponding updates graph and associated data.

Open Request An ALTO client requests that the server provide a TIPS view for a given resource by sending an HTTP POST body with the media type "application/alto-tipsparams+json". That body contains a JSON object of type TIPSReq, where:

TIPSReq

with the following fields:

resource-id:

The resource-id of an ALTO resource and MUST be in the TIPS' "uses" list (). If a client does not support all incremental methods from the set announced in the server's capabilities, the client MUST NOT use the TIPS service.

tag:

If the resource-id is a GET-mode resource with a version tag (or "vtag"), as defined in Section 10.3 of , and the ALTO client has previously retrieved a version of that resource from ALTO, the ALTO client MAY set the "tag" field to the tag part of the client's version of that resource. The server MAY use the tag when calculating a recommended starting edge for the client to consume. Note that the client MUST support all incremental methods from the set announced in the server's capabilities for this resource.

input:

If the resource is a POST-mode service that requires input, the ALTO client MUST set the "input" field to a JSON object with the parameters that the resource expects.

Open Response The response to a valid request MUST be a JSON object of type AddTIPSResponse, denoted as media type "application/alto-tips+json":

AddTIPSResponse

with the following fields:

tips-view-uri:

URI to the requested TIPS view. The value of this field must have the following format: where schema must be "http" or "https" unless specified by a future extension, and host, port and path are as specified in Sections 3.2.2, 3.2.3, and 3.3 in . The URI MUST be unique per client session and may be de-aliased by the server to refer to the actual location of the TIPS view which may be shared by other clients.

When creating the URI for the TIPS view, TIPS MUST NOT use other properties of an HTTP request, such as cookies or the client's IP address, to determine the TIPS view. Furthermore, TIPS MUST NOT reuse a URI for a different object in the same user session.

It is expected that there is an internal mechanism to map a tips- view-uri to the TIPS view to be accessed. For example, TIPS may assign a unique, internal state id to each TIPS view instance. However, the exact mechanism is left to the TIPS provider.

tips-view-summary:

Contains an updates-graph-summary and a heartbeat-interval. The updates-graph-summary field contains the starting sequence number (start-seq) of the updates graph and the last sequence number (end-seq) that is currently available, along with a recommended edge to consume (start-edge-rec). How the server calculates the recommended edge depends on the implementation. Ideally, if the client does not provide a version tag, the server should recommend the edge of the latest snapshot available. If the client does provide a version tag, the server should calculate the cumulative size of the incremental updates available from that version onward and compare it to the size of the complete resource snapshot. If the snapshot is bigger, the server should recommend the first incremental update edge starting from the client's tagged version. Otherwise, the server should recommend the latest snapshot edge. The heartbeat-interval field contains the interval length of the heartbeat. The time unit is second and the value MUST be interpreted as a floating number. The client is expected to send a heartbeat message (see ) for each TIPS view that it is currently using, at least every heartbeat-interval seconds. It is RECOMMENDED that the ALTO server picks an appropriate value for the heartbeat-interval field: if the value is too big, it may waste additional resources to maintain unused TIPS views; but if the value is too small, the normal use of the TIPS service might be disrupted.

If the request has any errors, the TIPS service must return an HTTP "400 Bad Request" to the ALTO client; the body of the response follows the generic ALTO error response format specified in Section 8.5.2 of . Hence, an example ALTO error response has the format shown in .

ALTO Error Example

Note that "field" and "value" are optional fields. If the "value" field exists, the "field" field MUST exist.

• If the TIPS request does not have a "resource-id" field, the error code of the error message MUST be E_MISSING_FIELD and the "field" field SHOULD be "resource-id". The TIPS service MUST NOT create any TIPS view.
• If the "resource-id" field is invalid or is not associated with the TIPS, the error code of the error message MUST be E_INVALID_FIELD_VALUE. The "field" field SHOULD be the full path of the "resource-id" field, and the "value" field SHOULD be the invalid resource-id.
• If the resource is a POST-mode service that requires input, the client MUST set the "input" field to a JSON object with the parameters that that resource expects. If the "input" field is missing or invalid, TIPS MUST return the same error response that resource would return for missing or invalid input (see ).

Furthermore, it is RECOMMENDED that the server uses the following HTTP codes to indicate other errors, with the media type "application/alto-error+json".

• 429 (Too Many Requests): when the number of TIPS views open requests exceeds the server threshold. The server may indicate when to re-try the request in the "Re-Try After" headers.

Open Example For simplicity, assume that the ALTO server is using the Basic authentication. If a client with username "client1" and password "helloalto" wants to create a TIPS view of an ALTO Cost Map resource with resource ID "my-routingcost-map", it can send the request depicted in .

Open Example

If the operation is successful, the ALTO server returns the message shown in .

Response Example

Heartbeat Request and Response Upon creating a TIPS view, the ALTO client must indicate its liveness to that specific TIPS view by sending a heartbeat request every heartbeat-interval second(s). The heartbeat request MUST have the following format: /heartbeat HTTP/1.1 HOST: ]]> The body of the request must be empty. The response to a heartbeat request must have status code 200 if the server has acknowledged the heartbeat message, and the corresponding error code if there is any error. When the request is successful, the body of the response must be empty. It is RECOMMENDED that the server uses the following HTTP codes to indicate errors, with the media type "application/alto-error+json", regarding TIPS view heartbeat requests.

• 404 (Not Found): if the requested TIPS view does not exist or is closed.

An ALTO server should close a TIPS view if multiple consecutive heartbeat requests are not received for that TIPS view. The maximum number of missing heartbeat requests is specific to the server implementation, e.g., statically configured for the server, specific to the resource and/or client, or dynamically set based on the server load, etc.

Close Request and Response An ALTO client can indicate it no longer desires to pull/receive updates for a specific network resource by "deleting" the TIPS view using the returned tips-view-uri and the HTTP DELETE method. The DELETE request MUST have the following format: HTTP/1.1 HOST: ]]> The response to a valid request must be 200 if successful, and the corresponding error code if there is any error. It is RECOMMENDED that the server uses the following HTTP codes to indicate errors, with the media type "application/alto-error+json", regarding TIPS view close requests.

• 404 (Not Found): if the requested TIPS view does not exist or is closed.

It must be noted that whether or not the server releases the resources associated with the TIPS view is implementation-dependent. For example, an ALTO server may maintain a set of clients that subscribe to the TIPS view of a resource: a client that deletes the view is removed from the set, and the TIPS view is only removed when the dependent set becomes empty. See other potential implementations in .

TIPS Data Transfers - Client Pull TIPS allows an ALTO client to retrieve the content of an update item from the updates graph, with an update item defined as the content (incremental update or snapshot) on an edge in the updates graph.

Request The client sends an HTTP GET request, where the media type of an update item resource MUST be the same as the "media-type" field of the update item on the specified edge in the updates graph. The GET request MUST have the following format: /ug// HOST: ]]> For example, consider the updates graph in . If the client wants to query the content of the first update item (0 -> 101) whose media type is "application/alto- costmap+json", it must send a request to "/tips/2718281828459/ug/0/101" and set the "Accept" header to "application/alto- costmap+json, application/alto-error+json". See for a concrete example.

Response If the request is valid (ug/<i>/<j> exists), the response is encoded as a JSON object whose data format is indicated by the media type. A client may conduct proactive fetching of future updates, by long polling updates that have not been listed in the directory yet. For long-poll prefetch, the client must have indicated the media type that may appear. It is RECOMMENDED that the server allows for at least the prefetch of <end-seq> -> <end-seq + 1> Hence, the server processing logic SHOULD be:

• If ug/<i>/<j> exists: return content using encoding.
• Else if ug/<i>/<j> pre-fetch is acceptable: put request in a backlog queue.
• Else: return error.

It is RECOMMENDED that the server uses the following HTTP codes to indicate errors, with the media type "application/alto-error+json", regarding update item requests.

• 404 (Not Found): if the requested TIPS view does not exist or is closed.
• 410 (Gone): if an update has a seq that is smaller than the start- seq.
• 415 (Unsupported Media Type): if the media type(s) accepted by the client does not include the media type of the update chosen by the server.
• 425 (Too Early): if the seq exceeds the server prefetch window
• 429 (Too Many Requests): when the number of pending (long-poll) requests exceeds the server threshold. The server may indicate when to re-try the request in the "Re-Try After" headers.

Example Assume the client wants to get the contents of the update item on edge 0 to 101. The format of the request is shown in .

GET Example

The response is shown in .

Response to a GET Request

New Next Edge Recommendation While intended TIPS usage is for the client to receive a recommended starting edge in the TIPS summary, consume that edge, then construct all future URIs by incrementing the sequence count by 1, there may be cases in which the client needs to request a new next edge to consume. For example, if a client has an open TIPS view yet has not polled in a while, the client may request the next logical incremental URI but the server has compacted the updates graph so it no longer exists. Thus, the client must request a new next edge to consume based on its current version of the resource.

Request An ALTO client requests that the server provide a next edge recommendation for a given TIPS view by sending an HTTP POST request with the media type "application/alto-tipsparams+json". The request has the form: /ug HTTP/1.1 HOST: ]]> The POST body has the following form, where providing the version tag of the resource the client already has is optional:

Response The response to a valid request MUST be a JSON object of type UpdatesGraphSummary (defined in but reproduced in as well), denoted as media type "application/alto-tips+json":

UpdatesGraphSummary

It is RECOMMENDED that the server uses the following HTTP codes to indicate errors, with the media type "application/alto-error+json", regarding new next edge requests.

• 404 (Not Found): if the requested TIPS view does not exist or is closed.
• 415 (Unsupported Media Type): if the media type(s) accepted by the client does not include the media type application/alto-tips+json.

Operation and Processing Considerations

Considerations for Choosing Updates When implementing TIPS, a developer should be cognizant of the effects of update schedule, which includes both the choice of timing (i.e., when/what to trigger an update on the updates graph) and the choice of message format (i.e., given an update, send a full replacement or an incremental change). In particular, the update schedule can have effects on both the overhead and the freshness of information. To minimize overhead, developers may choose to batch a sequence of updates for resources that frequently change by cumulative updates or a full replacement after a while. Developers should be cognizant that batching reduces the freshness of information and should also consider the effect of such delays on client behaviors. For incremental updates, this design allows both JSON patch and JSON merge patch for incremental changes. JSON merge patch is clearly superior to JSON patch for describing incremental changes to cost maps, endpoint costs, and endpoint properties. For these data structures, JSON merge patch is more space efficient, as well as simpler to apply. There is no advantage in allowing a server to use JSON patch for those resources. The case is not as clear for incremental changes to network maps. First, consider small changes, such as moving a prefix from one PID to another. JSON patch could encode that as a simple insertion and deletion, while JSON merge patch would have to replace the entire array of prefixes for both PIDs. On the other hand, to process a JSON patch update, the ALTO client would have to retain the indexes of the prefixes for each PID. Logically, the prefixes in a PID are an unordered set, not an array; aside from handling updates, a client does not need to retain the array indexes of the prefixes. Hence, to take advantage of JSON patch for network maps, ALTO clients would have to retain additional, otherwise unnecessary, data. Second, consider more involved changes, such as removing half of the prefixes from a PID. JSON merge patch would send a new array for that PID, while JSON patch would have to send a list of remove operations and delete the prefix one by one. Therefore, each TIPS instance may choose to encode the updates using JSON merge patch or JSON patch based on the type of changes in network maps.

Considerations for Load Balancing There are two levels of load balancing in TIPS. The first level is to balance the load of TIPS views for different clients, and the second is to balance the load of incremental updates. Load balancing of TIPS views can be achieved either at the application layer or at the infrastructure layer. For example, an ALTO server may set <tips-view-host> to different subdomains to distribute TIPS views, or simply use the same host of the TIPS service and rely on load balancers to distribute the load. TIPS allows clients to make concurrent pulls of incremental updates potentially through different HTTP connections. As a consequence, it introduces additional complexities when the ALTO server is being load balanced -- a feature widely used to build scalable and fault-tolerant web services. For example, a request may be incorrectly processed if the following two conditions both hold:

• the backend servers are stateful, i.e., the TIPS view is created and stored only on a single server;
• the ALTO server is using layer-4 load balancing, i.e., the requests are distributed based on the TCP 5-tuple.

Thus, additional considerations are required to enable correct load balancing for TIPS, including:

• Use a stateless architecture: One solution is to follow the stateless computing pattern: states about the TIPS view are not maintained by the backend servers but are stored in a distributed database. Thus, concurrent requests to the same TIPS view can be processed on arbitrary stateless backend servers, which all fetches data from the same database.
• Configure the load balancers properly: In case when the backend servers are stateful, the load balancers must be properly configured to guarantee that requests of the same TIPS view always arrive at the same server. For example, an operator or a provider of an ALTO server may configure layer-7 load balancers that distribute requests based on the tips-view-path component in the URI.

Considerations for Cross-Resource Dependency Scheduling Dependent ALTO resources result in cross-resource dependencies in TIPS. Consider the following pair of resources, where my-cost-map (C) is dependent on my-network-map (N). The updates graph for each resource is shown, along with links in between the respective updates graphs to show dependency:

Example Dependency Model |89 |-->|90 |-->|91 |-->|92 | +---+ +---+ +---+ +---+ +---+ | \ \ \ | \ \ \ +---+ +---+ +---+ +---+ +---+ my-cost-map (C) | 0 |-->|101|-->|102|-->|103|-->|104| +---+ +---+ +---+ +---+ +---+ |_______________________| ]]>

In , the cost-map versions 101 and 102 (denoted as C101 and C102) are dependent on the network-map version 89 (denoted as N89). The cost-map version 103 (C103) is dependent on the network-map version 90 (N90), and so on. Thus, the client must decide the order in which to receive and apply the updates. The order may affect how fast the client can build a consistent view and how long the client needs to buffer the update.

• Example 1: The client requests N89, N90, N91, C101, C102 in that order. The client either gets no consistent view of the resources or has to buffer N90 and N91.
• Example 2: The client requests C101, C102, C103, N89. The client either gets no consistent view or has to buffer C103.

Therefore, the client is RECOMMENDED to request and process updates in the ascending order of the smallest dependent tag, e.g., {C101, C102, N89} before {C103, N90}

Considerations for Client Processing Updates In general, when an ALTO client receives a full replacement for a resource, the ALTO client should replace the current version with the new version. When an ALTO client receives an incremental update for a resource, the ALTO client should apply those updates to the current version of the resource. However, because resources can depend on other resources (e.g., cost maps depend on network maps), an ALTO client must not use a dependent resource if the resource on which it depends has changed. There are at least two ways an ALTO client can do that. The following paragraphs illustrate these techniques by referring to network and cost map messages, although these techniques apply to any dependent resources. Note that when a network map changes, the server should send the network map update message before sending the updates for the dependent cost maps. One approach is for the ALTO client to save the network map update message in a buffer and continue to use the previous network map and the associated cost maps until the ALTO client receives the update messages for all dependent cost maps. The ALTO client then applies all network and cost map updates atomically. Alternatively, the ALTO client may update the network map immediately. In this case, the cost maps using the network map become invalid because they are inconsistent with the current network map; hence, the ALTO client must mark each such dependent cost map as temporarily invalid and must not use each such cost map until the ALTO client receives a cost map update indicating that it is based on the new network map version tag. Though a server should send update items sequentially, it is possible that a client receives the update items out of order (in the case of a retransmitted update item or a result of concurrent fetch). The client must buffer the update items if they arrive out of order and then apply them sequentially (based on the sequence numbers) due to the operation of JSON merge patch and JSON patch.

Considerations for Updates to Filtered Cost Maps If TIPS provides updates to a Filtered Cost Map that allows constraint tests, then an ALTO client may request updates to a Filtered Cost Map request with a constraint test. In this case, when a cost changes, the updates graph MUST have an update if the new value satisfies the test. If the new value does not, whether there is an update depends on whether the previous value satisfies the test. If it did not, the updates graph SHOULD NOT have an update. But if the previous value did, then the updates graph MUST add an update with a "null" value to inform the ALTO client that this cost no longer satisfies the criteria. TIPS can avoid having to handle such complicated behavior by offering TIPS only for Filtered Cost Maps that do not allow constraint tests.

Considerations for Updates to Ordinal Mode Costs For an ordinal mode cost map, a change to a single cost point may require updating many other costs. As an extreme example, suppose the lowest cost changes to the highest cost. For a numerical mode cost map, only that one cost changes. But for an ordinal mode cost map, every cost might change. While this document allows TIPS to offer incremental updates for ordinal mode cost maps, TIPS implementors should be aware that incremental updates for ordinal costs are more complicated than for numerical costs, and that small changes of the original cost value may result in large updates. A TIPS implementation can avoid this complication by only offering full replacements as updates in the updates graph for ordinal cost maps.

Considerations for Managing Shared TIPS Views From a client's point of view, it sees only one copy of the TIPS view for any resource. However, on the server side, there are different implementation options, especially for common resources (e.g., network map or cost map) that may be frequently queried by many clients. Some potential options are listed below:

• An ALTO server creates one TIPS view of the common resource for each client. When the client deletes the view, the server deletes the view in the server storage.
• An ALTO server maintains one copy of the TIPS view for each common resource and all clients requesting the same resources use the same copy. There are two ways to manage the storage for the shared copy:
  • the ALTO server maintains the set of clients that subscribe to the TIPS view, and only removes the view from the storage when the set becomes empty.
  • the TIPS view is never removed from the storage.

Developers may choose different implementation options depending on criteria such as request frequency, available resources of the ALTO server, the ability to scale, and programming complexity.

Considerations for Offering Shortcut Incremental Updates Besides the mandatory stepwise incremental updates (from i to i+1), an ALTO server may optionally offer shortcut incremental updates, or simple shortcuts, between two non-consecutive versions i and i+k (k > 1). Such shortcuts offer alternative paths in the update graph and can potentially speed up the transmission and processing of incremental updates, leading to faster synchronization of ALTO information, especially when the client has limited bandwidth and computation. However, implementors of an ALTO server must be aware that:

1. Optional shortcuts may increase the size of the update graph, in the worst case being the square of the number of updates (i.e., when a shortcut is offered for each version to all future versions).
2. Optional shortcuts require additional storage on the ALTO server.
3. Optional shortcuts may reduce concurrency when the updates do not overlap, e.g., when the updates apply to different parts of an ALTO resource. In such a case, the total size of the original updates is close to the size of the shortcut, but the original updates can be transmitted concurrently while the shortcut is transmitted in a single connection.

Security Considerations The security considerations (Section 15 of ) of the base protocol fully apply to this extension. For example, the same authenticity and integrity considerations (Section 15.1 of ) still fully apply; the same considerations for the privacy of ALTO users (Section 15.4 of ) also still fully apply. The additional services (addition of update read service and update push service) provided by this extension extend the attack surface described in Section 15.1.1 of . The following sub-sections discuss the additional risks and their remedies.

TIPS: Denial-of-Service Attacks Allowing TIPS views enables a new class of Denial-of-Service attacks. In particular, for the TIPS server, an ALTO client might create an excessive number of TIPS views. To avoid these attacks on the TIPS server, the server SHOULD choose to limit the number of active views and reject new requests when that threshold is reached. TIPS allows predictive fetching and the server SHOULD also choose to limit the number of pending requests. If a new request exceeds the threshold, the server SHOULD log the event and may return the HTTP status "429 Too many requests". It is important to note that the preceding approaches are not the only possibilities. For example, it may be possible for TIPS to use somewhat more clever logic involving IP reputation, rate-limiting, and compartmentalization of the overall threshold into smaller thresholds that apply to subsets of potential clients.

ALTO Client: Update Overloading or Instability The availability of continuous updates, when the client indicates receiving server push, can also cause overload for an ALTO client, in particular, an ALTO client with limited processing capabilities. The current design does not include any flow control mechanisms for the client to reduce the update rates from the server. For example, TCP, HTTP/2 and QUIC provide stream and connection flow control data limits, and PUSH stream limits, which might help prevent the client from being overloaded. Under overloading, the client MAY choose to remove the information resources with high update rates. Also, under overloading, the client may no longer be able to detect whether information is still fresh or has become stale. In such a case, the client should be careful in how it uses the information to avoid stability or efficiency issues.

Spoofed URI An outside party that can read the TIPS response or that can observe TIPS requests can obtain the TIPS view URI and use that to send fraudulent "DELETE" requests, thus disabling the service for the valid ALTO client. This can be avoided by encrypting the requests and responses (Section 15 of ).

IANA Considerations IANA is requested to register the following media types from the registry available at :

• application/alto-tips+json: as described in ;
• application/alto-tipsparams+json: as described in ;

• Note to the RFC Editor: Please replace This-Document with the RFC number to be assigned to this document.

application/alto-tips+json Media Type

Type name:

application

Subtype name:

alto-tips+json

Required parameters:

N/A

Optional parameters:

N/A

Encoding considerations:

Encoding considerations are identical to those specified for the "application/json" media type. See .

Security considerations:

See the Security Considerations section of This-Document.

Interoperability considerations:

This document specifies the format of conforming messages and the interpretation thereof.

Published specification:

of This-Document.

Applications that use this media type:

ALTO servers and ALTO clients either stand alone or are embedded within other applications.

Fragment identifier considerations:

N/A

Additional information:

Deprecated alias names for this type:

N/A

Magic number(s):

N/A

File extension(s):

This document uses the media type to refer to protocol messages and thus does not require a file extension.

Macintosh file type code(s):

N/A

Person and email address to contact for further information:

See Authors' Addresses section.

Intended usage:

COMMON

Restrictions on usage:

N/A

Author:

See Authors' Addresses section.

Change controller:

Internet Engineering Task Force (mailto:iesg@ietf.org).

Provisional registration?:

No

application/alto-tipsparams+json Media Type

Type name:

application

Subtype name:

alto-tipsparams+json

Required parameters:

N/A

Optional parameters:

N/A

Encoding considerations:

Encoding considerations are identical to those specified for the "application/json" media type. See .

Security considerations:

See the Security Considerations section of This-Document.

Interoperability considerations:

This document specifies the format of conforming messages and the interpretation thereof.

Published specification:

of This-Document.

Applications that use this media type:

ALTO servers and ALTO clients either stand alone or are embedded within other applications.

Fragment identifier considerations:

N/A

Additional information:

Deprecated alias names for this type:

N/A

Magic number(s):

N/A

File extension(s):

This document uses the media type to refer to protocol messages and thus does not require a file extension.

Macintosh file type code(s):

N/A

Person and email address to contact for further information:

See Authors' Addresses section.

Intended usage:

COMMON

Restrictions on usage:

N/A

Author:

See Authors' Addresses section.

Change controller:

Internet Engineering Task Force (mailto:iesg@ietf.org).

Provisional registration?:

No

References Normative References In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. Applications using the Internet already have access to some topology information of Internet Service Provider (ISP) networks. For example, views to Internet routing tables at Looking Glass servers are available and can be practically downloaded to many network application clients. What is missing is knowledge of the underlying network topologies from the point of view of ISPs. In other words, what an ISP prefers in terms of traffic optimization -- and a way to distribute it. The Application-Layer Traffic Optimization (ALTO) services defined in this document provide network information (e.g., basic network location structure and preferences of network paths) with the goal of modifying network resource consumption patterns while maintaining or improving application performance. The basic information of ALTO is based on abstract maps of a network. These maps provide a simplified view, yet enough information about a network for applications to effectively utilize them. Additional services are built on top of the maps. This document describes a protocol implementing the ALTO services. Although the ALTO services would primarily be provided by ISPs, other entities, such as content service providers, could also provide ALTO services. Applications that could use the ALTO services are those that have a choice to which end points to connect. Examples of such applications are peer-to-peer (P2P) and content delivery networks. JavaScript Object Notation (JSON) is a lightweight, text-based, language-independent data interchange format. It was derived from the ECMAScript Programming Language Standard. JSON defines a small set of formatting rules for the portable representation of structured data. This document removes inconsistencies with other specifications of JSON, repairs specification errors, and offers experience-based interoperability guidance. RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. The Application-Layer Traffic Optimization (ALTO) protocol (RFC 7285) provides network-related information, called network information resources, to client applications so that clients can make informed decisions in utilizing network resources. This document presents a mechanism to allow an ALTO server to push updates to ALTO clients to achieve two benefits: (1) updates can be incremental, in that if only a small section of an information resource changes, the ALTO server can send just the changes and (2) updates can be immediate, in that the ALTO server can send updates as soon as they are available. The Hypertext Transfer Protocol (HTTP) is a stateless application-level protocol for distributed, collaborative, hypertext information systems. This document specifies the HTTP/1.1 message syntax, message parsing, connection management, and related security concerns. This document obsoletes portions of RFC 7230. This specification describes an optimized expression of the semantics of the Hypertext Transfer Protocol (HTTP), referred to as HTTP version 2 (HTTP/2). HTTP/2 enables a more efficient use of network resources and a reduced latency by introducing field compression and allowing multiple concurrent exchanges on the same connection. This document obsoletes RFCs 7540 and 8740. The QUIC transport protocol has several features that are desirable in a transport for HTTP, such as stream multiplexing, per-stream flow control, and low-latency connection establishment. This document describes a mapping of HTTP semantics over QUIC. This document also identifies HTTP/2 features that are subsumed by QUIC and describes how HTTP/2 extensions can be ported to HTTP/3. A Uniform Resource Identifier (URI) is a compact sequence of characters that identifies an abstract or physical resource. This specification defines the generic URI syntax and a process for resolving URI references that might be in relative form, along with guidelines and security considerations for the use of URIs on the Internet. The URI syntax defines a grammar that is a superset of all valid URIs, allowing an implementation to parse the common components of a URI reference without knowing the scheme-specific requirements of every possible identifier. This specification does not define a generative grammar for URIs; that task is performed by the individual specifications of each URI scheme. [STANDARDS-TRACK] Informative References Applications often use HTTP as a substrate to create HTTP-based APIs. This document specifies best practices for writing specifications that use HTTP to define new application protocols. It is written primarily to guide IETF efforts to define application protocols using HTTP for deployment on the Internet but might be applicable in other situations. This document obsoletes RFC 3205.

A High-Level Deployment Model Conceptually, the TIPS system consists of three types of resources:

• (R1) TIPS frontend to manage (create/delete) TIPS views.
• (R2) TIPS view directory, which provides metadata (e.g., references) about the network resource data.
• (R3) The actual network resource data, encoded as complete ALTO network resources (e.g., cost map, network map) or incremental updates.

Sample TIPS Deployment Model | | | | | | | | | incremental | | | | |-------->| | | | | transfer | | | | | | | | | | resource | | | | | | | | | |<------------|-----------------------| | | | | |Client| | | | +-----+ +-----+ | | | "iget" base | | | | | | resource 2 | | | +-----+ +-----+ | | |-------------|---->| | | | | | | | | incremental | | | | | | | | | | transfer | +-----+ | | ------->| | | | | resource | | | | | | | |<------------|-----------------------| | | | | +------+ | +-----+ +-----+ | | | +------------------------------------------------+ ]]>

Design Point: Component Resource Location

• Design 1 (Single): all the three resource types at the same, single server (accessed via relative reference)
• Design 2 (Flexible): all three resource types can be at their own server (accessed via absolute reference)
• Design 3 (Dir + Data): R2 and R3 must remain together, though R1 might not be on the same server

This document supports Design 1 and Design 3. For Design 1, the TIPS service simply needs to always use the same host for the TIPS views. For Design 3, the TIPS service can set tips-view-host to a different server. Note that the deployment flexibility is at the logical level, as these services can be distinguished by different paths and potentially be routed to different physical servers by layer-7 load balancing. See for a discussion on load balancing considerations. Future documents may extend the protocol to support Design 2.

Conformance to "Building Protocols with HTTP" Best Current Practices This specification adheres fully to as further elaborated below:

• TIPS does not "redefine, refine, or overlay the semantics of generic protocol elements such as methods, status codes, or existing header fields" and instead focuses on "protocol elements that are specific to [the TIPS] application -- namely, [its] HTTP resources" ().
• There are no statically defined URI components ().
• No minimum version of HTTP is specified by TIPS which is recommended ().
• The TIPS design follows the advice that "When specifying examples of protocol interactions, applications should document both the request and response messages with complete header sections, preferably in HTTP/1.1 format" ().
• TIPS uses URI templates which is recommended ().
• TIPS follows the pattern that "a client will begin interacting with a given application server by requesting an initial document that contains information about that particular deployment, potentially including links to other relevant resources. Doing so ensures that the deployment is as flexible as possible (potentially spanning multiple servers), allows evolution, and also allows the application to tailor the "discovery document" to the client" ().
• TIPS uses existing HTTP schemes ().
• TIPS defines its errors "to use the most applicable status code" ().
• TIPS does not "make assumptions about the relationship between separate requests on a single transport connection; doing so breaks many of the assumptions of HTTP as a stateless protocol and will cause problems in interoperability, security, operability, and evolution" (). The only relationship between requests is that a client must first discover where a TIPS view of a resource will be served, which is consistent with the URI discovery in .

Push-mode TIPS using HTTP Server Push TIPS allows ALTO clients to subscribe to incremental updates of an ALTO resource, and the specification in this document is based on the current best practice of building such a service using native HTTP. Earlier versions of this document had investigated the possibility of enabling push-mode TIPS, i.e., by taking advantage of the server push feature in HTTP/2 and HTTP/3. In the ideal case, push-mode TIPS can potentially improve performance (e.g., latency) in more dynamic environments and use cases, with wait-free message delivery. Using native server push also results in minimal changes to the current protocol. While not adopted due to the lack of server push support and increased protocol complexity, push-mode TIPS remains a potential direction of protocol improvement.

Persistent HTTP Connections Previous versions of this document use persistent HTTP connections to detect the liveness of clients. This design is replaced with the heartbeat mechanism because it does not conform well with the best current practice of HTTP. For example, if an ALTO client is accessing a TIPS view over an HTTP proxy, the connection is not established directly between the ALTO client and the ALTO server, but between the ALTO client and the proxy and between the proxy and the ALTO server. Thus, using persistent connections may not correctly detect the right liveness state.

Acknowledgments The authors of this document would like to thank Mark Nottingham and Spencer Dawkins for providing invaluable reviews of earlier versions of this document, Adrian Farrel, Qin Wu, and Jordi Ros Giralt for their continuous feedback, Russ White, Donald Eastlake, Martin Thomson, Bernard Adoba, Spencer Dawkins, and Sheng Jiang for the directorate reviews, Martin Duke for the Area Director review, and Mohamed Boucadair for shepherding the document.