| Internet-Draft | SCONE APIs | November 2024 |
| Eddy, et al. | Expires 7 May 2025 | [Page] |
This document describes API considerations to provide applications with network-supplied information about acceptable network flow rates. Since this information is expected to be signalled from the network within the stack below the application using the SCONE protocol (to be defined by the IETF), it needs to be made accessible to applications in order for them to pick proper video rates, or to otherwise confine the application behavior within network-defined limits.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 7 May 2025.¶
Copyright (c) 2024 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.¶
Video traffic is already 70% of all traffic on the Internet and is expected to grow to 80% by 2028. New formats like short form videos have seen tremendous growth in recent years. Both in developed and emerging markets video traffic forms 50-80% of traffic on mobile networks. These growth trends are likely to increase with new populations coming online on mobile-first markets and the observation that unlike text content, video content consumption is not being limited by literacy barriers. On the other hand, the electromagnetic spectrum is a limited resource. In order to ensure that mobile networks continue functioning in a healthy state despite this incredible growth, communication service providers (CSPs) will be required to make infrastructure investments such as more licensed spectrum, cell densification, massive MIMO etc. In order to flatten the rate of growth, CSPs in several markets attempt to identify and throttle video traffic based on user data plans. There are several problems with this kind of throttling:¶
CSPs can not explicitly measure the effect that throttling has on the end user’s quality of experience (QoE) making this an open loop approach.¶
Traffic detection and throttling for every flow is compute intensive for CSPs. With distributed UPF (user plane function) in 5G mobile networks more nodes in CSP network may need to support traffic detection and throttling. Traffic detection can have inaccuracies and these inaccuracies are expected to increase as the content delivery industry moves towards end-2-end encryption like TLS 1.3 and encrypted client hello (ECH).¶
The unpredictable and non-transparent behavior of traffic throttlers used by CSPs confuse the bandwidth estimation and congestion control protocols being used within end-2-end video delivery sessions between content server and client. This results in poor quality of experience (QoE) for the end user.¶
Content and Application Providers (CAPs) are designing algorithms to detect the presence of such traffic throttlers to counter their detrimental effects. These algorithms have their own inaccuracies in detection and add compute resources on the CAP side.¶
An alternative approach is for CAPs to self-adapt the traffic corresponding to video flows. Since CAPs control the client and server endpoints and can measure end user QoE, they are in a better position to do this self-adaptation in a close loop manner. This alternative approach has already been proven to improve user QoE in production deployments [YouTube].¶
For this alternative approach to work a standardized secure on-path network interface is required which will enable CSP controlled network elements to signal the desired traffic profile characteristics to the CAP client/server endpoints. The Standard Communication with Network Elements (SCONE) protocol (previously known as SADCDN and SCONEPRO) is an IETF working group motivated by this alternate approach.¶
The general problem statement for SCONE is described in the video optimization requirements document [I-D.joras-sconepro-video-opt-requirements], including the shaping or throttling that CSPs perform. While this problem currently has especially large impact on a few large content providers, solutions for SCONE are generally applicable to any applications that use QUIC [RFC9000] and are subject to throttling within CSP networks.¶
General use of SCONE metadata for any applications can be facilitated via an open Application Programming Interface (API) that could be implemented in appropriate QUIC stacks, web browsers, or other libraries.¶
There are two aspects to consider for an API:¶
How will applications learn about network information that is discovered by SCONE lower in the stack? This is a primary consideration in this document.¶
How will applications signal their type (e.g. video streaming) or other relevant properties to the stack, to indicate that they are SCONE-capable? This is a secondary consideration in this document, because currently networks that perform throttling have built-in methods to implicitly determine the appropriate flows to throttle.¶
Depending on how the SCONE signaling protocol works (that is to-be-defined by the IETF), the SCONE metadata may be available at various different places in the protocol stack spanning operating system, browser, and application code. This document tries to initially make no assumptions about how the SCONE signalling works, and so considers possibilities to integrate the metadata into APIs provided from OSes, QUIC libraries, web browsers, etc. There are open questions at the moment about SCONE signaling via on-path devices, what type of information is conveyed, and how it is represented. The API capabilities may be limited by the protocol decisions, and realistic concerns about signaling across network domain boundaries, etc.¶
During early SCONE discussions, there have been suggestions that the API might take inspiration from Explicit Congestion Notification (ECN), as ECN also exposes information from the network (congestion experienced codepoints) to end hosts, and the design contends with potential for abuse, crosses network domain boundaries, and has other desirable properties. Some key differences from ECN in usage relavent to a SCONE API have been pointed out:¶
1) ECN information is consumed either by transport protocols (e.g. TCP, MPTCP, and SCTP) or congestion control algorithms operating above e.g. UDP or UDP-Lite [RFC8303] [RFC8304], but typically below an application. For instance, within QUIC stacks used for video streaming, ECN can be consumed by the QUIC congestion control, but is not exposed to the application.¶
2) The exposure of SCONE metadata is intended to be at the level of data flows (e.g. to aid application decisions about what media quality to fetch), whereas ECN is consumed at the level of packets (within an individual flow).¶
While ECN is not a solution for SCONE [I-D.tomar-sconepro-ecn], it is productive to consider as an example based on similarities, including:¶
Signaling is coming from the network, and may cross different network domains.¶
Signaling points can also drop packets, and the signaling participation is indtended to avoid excessive packet drops.¶
The purpose of this document is only to demonstrate that general API exposure of the SCONE metadata is achievable without prescribing a specific API solution. It is envisioned that one or more specific API solutions will be defined either through IETF or W3C, to correspond with the SCONE protocol specification. At least in its current form, this document is not intended to be published as an RFC.¶
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.¶
There are a variety of different application stack designs that are relevant. The main assumption for SCONE in general is that QUIC is used.¶
Applications could, for instance, either use QUIC directly through a linked software library, or applications could be running within a browser, using QUIC indirectly via browser APIs.¶
Specific protocol solutions for SCONE will be defined by the working group. In general, the SCONE network rate-limiting information could be discovered by an end-system in several different ways; potential network signaling approaches are described in other documents (e.g. [I-D.ihlar-scone-masque-mediabitrate], [I-D.brw-sconepro-rate-policy-discovery], or others). General classes of information signaling include:¶
Via the QUIC stack, with the information inserted by an on-path SCONE network element.¶
Via other IP or transport methods below QUIC (e.g. IP extension headers, UDP options, etc.) that might be inserted on the path.¶
Via the OS, with the information coming in network advertisements separate from the transport connection (e.g. via Router Advertisements or DHCP).¶
These methods are not necessarily mutually exclusive. For instance, a DHCP option could indicate that certain classes of applications are throttled at a particular rate, while a SCONE on-path mechanism could also provide more explicit per-connection indications of when throttling is active as well.¶
| SCONE Solution Type | Information Visibility | API Definitions |
|---|---|---|
| QUIC-based | QUIC library or web browser | Specific to each QUIC-library, or browser APIs. |
| IP or UDP-based | OS and possibly QUIC library | Socket API extension may be needed, e.g. to expose via socket options or other means. For mobile apps using native stacks, OS extensions may be needed. |
| Other approaches that are not on-path or per-connection | OS | Per-OS and/or socket API extensions |
For solution types that are not QUIC-based, the QUIC implementation could use socket or OS API extensions in order to get the data and convey it up to the applications via its own API. However, QUIC library APIs are not standardized; they differ between common QUIC libraries, and so this document only suggests in a general sense of how the QUIC stack should convey this information to applications.¶
The following sections consider first the types of metadata that could be provided, and then how that SCONE metadata might be provided in different cases of APIs including potential:¶
There are existing collections of metadata that are available for some types of adaptive rate streaming. Examples include those defined by the CTA-5004 specification [CTA-5004] and CTA-5006 specification [CTA-5006], which standardize JSON objects or HTTP headers conveyed by streaming media clients and servers respectively. These CTA specifications are also expected to be usable by intermediaries.¶
Another example that could be built upon is the proposed flow metadata identifying the DownlinkBitrate [I-D.rwbr-sconepro-flow-metadata].¶
Several of the use cases defined by CTA-5006 are relevant, including:¶
Provide an estimate of the throughput available between an edge server and a player, such that the player can use that information to enrich its decision about which bitrate to select, especially the initial bitrate decision made at the start of playback.¶
Provide a means for a CDN to instruct all connected players to limit their upper bitrate, in response to an ISP request to reduce congestion on the last-mile network.¶
Allow a CDN to signal to a player a suggested playback bitrate in order to optimize collective QoE.¶
The "CDN" notion could be replaced with an ISP's on-path throttling device.¶
As a specific example of metadata for SCONE, CTA-5004 allows clients to indicate multiple types of rate metadata that are related to adaptive coded rate selection in the face of throttling. Which of these options are used depends upon the client, but could be one of:¶
| CTA Data | Units | CTA Description |
|---|---|---|
| Measured Throughput | Integer kilobits per second (kbps) | The throughput between client and server, as measured by the client and MUST be rounded to the nearest 100 kbps. This value, however derived, SHOULD be the value that the client is using to make its next Adaptive Bitrate switching decision. If the client is connected to multiple servers concurrently, it must take care to report only the throughput measured against the receiving server. If the client has multiple concurrent connections to the server, then the intent is that this value communicates the aggregate throughput the client sees across all those connections. |
| Requested maximum throughput | Integer kbps | The requested maximum throughput that the client considers sufficient for delivery of the asset. Values MUST be rounded to the nearest 100 kbps. ... Note: This can benefit clients by preventing buffer saturation through over-delivery ... |
The CTA-5006 also allows the server to convey a maximum suggested bitrate.¶
| CTA Data | Units | CTA Description |
|---|---|---|
| Max suggested bitrate | Integer kbps | The maximum bitrate value that the player SHOULD play in its Adaptive Bit Rate (ABR) ladder. If the player is playing a bitrate higher than this value, it SHOULD immediately switch to a bitrate lower than or equal to this value. |
There are two high-level approaches that might be viable in reusing CTA data types within an API that supports SCONE:¶
Append entire CTA-defined objects as optional components of dictionary or other data structures used in APIs, e.g. as a "cta5005" object that would be able to fully convey all of the CTA-defined metadata.¶
Append only selected fields from the CTA specs, reusing the definitions, but only including specific items of interest for SCONE's problem statement, and leaving other CTA-defined metadata to be conveyed through the existing CTA-defined methods.¶
The API can be agnostic to how the data is conveyed between network or on-path SCONE devices and either clients or servers, but the same API can convey the information to the applications, regardless of how the network signaling works. The semantics of the maxium suggested bitrate from the CTA-5006 match the needed semantics most closely, though since it is for an ABR ladder, it should be described carefully to apply the proper bitrate definition that the throttler is using for "on the wire" packets (including header overhead, etc.) versus the pure media payload stream bitrate.¶
For browser applications, there are multiple different browser APIs that might be extended to include SCONE metadata; notably including:¶
W3C Network Information¶
WebTransport using QUIC¶
HTTP Live Streaming (HLS) or Dynamic Adaptive Streaming over HTTP (DASH) client libraries¶
Any Javascript HTTP requests that directly or indirectly use HTTP/3.¶
In either of these cases, the corresponding W3C API definitions are the proper place for actual definition of API extensions, and this document is merely exploring possibilities.¶
The exploration is primarily around the ability to convey SCONE signaling information that is discovered from the network path up to applications. In addition, to indicate an application's desire to use SCONE signaling in the first place, some small API extension is also be required, unless relying totally on the underlying stack or network to infer which flows should be receiving SCONE treatment (e.g. as networks already infer which flows to throttle).¶
The W3C Network Information API [W3C-NetInfo] is supported to some extent by several, but not all, common web browsers today. It provides the possibility for an appliction to determine what underlying connection types or "effective" connection types (e.g. cellular. bluetooth, etc.) may be in use, with a corresponding set of performance parameter estimates including:¶
Round Trip Time (RTT) in milliseconds providing a delay estimate.¶
downlink in megabits per second providing an effective bandwidth estimate based on recently observed application layer throughput or properties of the underlying connectivity technology.¶
downlinkMax in megabits per second representing an upper bound on the downlink speed of the first network hop, based on the underlying connectivity technology.¶
The downlink and downlinkMax could be leveraged as places to put the SCONE-discovered rate limit for an application, since anything greater than the SCONE-discovered rate would not be expected to be usable for the application.¶
Alternatively, another field could be added to the NetworkInformation interface in order to specifically provide the SCONE metadata.¶
In any case, a good property of the Network Information API is that an application can hook event handlers to be notified of changes, so that if there are limits that kick-in or are lifted midway into an application's lifetime (e.g. due to some mobility, etc.), the application will be able to be easily notified and adapt.¶
In the future, WebTransport (WEBTRANS) might be used by SCONE's targeted types of applications, such as browser-based adaptive streaming. The IETF WEBTRANS working group is liasing with W3C as the IETF defines the protocol specification, whereas the W3C defines the API to use it. This case is similar to the IETF RTCWEB and W3C WebRTC WG coordination in the past. The same model of collaboration between IETF and W3C should work for SCONE metadata, and the information provided could be discussed with the WEBTRANS WG in the IETF and notified to the W3C later, either through common participation and/or formal liason statement.¶
The existing WebTrans API definition from W3C includes a "getStats()" method, that is defned in order to asynchronously gather and report statistics for a WebTransport underlying connection. It returns a WebTransportConnectionStats object that is defined as a dictionary, including a number of items such as:¶
bytesSent¶
packetsSent¶
bytesLost¶
packetsLost¶
bytesReceived¶
packetsReceived¶
smoothedRtt¶
rttVariation¶
minRtt¶
WebTransportDatagramStats datagrams¶
estimatedSendRate¶
The estimatedSendRate is an unsigned long long, in bits per second, calculated by the congestion control algorithm in the user agent. This would be in the "upstream" direction to a CSP, though, rather than the "downstream" from a CSP, so is not useful to a client application in receiving indication of a downstream throttling rate from the network.¶
Since other measurements are already included and the WebTransportConnectionStats is a dictionary, it seems natural to extend it to include additional optional fields, such as an allowed media rate, or other types of fields providing the application information that the underlying host or stack have discovered about the presence of throttling or explicit signaling of allowed media rate on a path.¶
Such extensions might be including at a "MAY" level of conformance statement (rather than "SHALL" as used by all of the currently-defined information elements), as the allowed media rate will not be universally present or even useful for all WebTransport applications. Alternatively, it could be set to a "null" value similar to how the estimatedSendRate is sent when it is unknown by the user agent.¶
Client libraries for HLS and DASH will use the underlying Javascript APIs or other underlying APIs, and might rely on them for SCONE metadata support, as discussed in the next subsection.¶
Typical HTTP adaptive streaming applications using existing browser API options would be ideal to support as directly as possible. There are different ways to transfer HTTP/3 data provided to JavaScript applications that might be applicable.¶
For instance, things like jQuery.ajax() or the "Fetch" API may be used. In this case, there is little or no information about the network path state provided, though there are either jqXHR or Response objects returned that (for instance) allow HTTP headers to be conveyed, and in both cases could be extended to include SCONE metadata. These are returned at the completion of the HTTP transfer, however. It might be more difficult to support more dynamic updates such as providing the metadata to an application mid-transfer so that an application might quickly switch to other media rates for future video segments being pre-fetched.¶
While there are no standard QUIC APIs, and there are multiple different styles in use, many QUIC implementations include objects in the API that represent the QUIC connections directly, and allow setting callbacks for connection-related events, or allow direct querying of connection state. SCONE metadata could either be supported as a type of callback event, triggered when the metadata is received, or it could be included within other connection state in a polled or interrogated data structure.¶
Other QUIC implementations may leave I/O and management of sockets or other aspects to the application, external to the QUIC library. In this case:¶
If the SCONE metadata is visible as part of the QUIC connection, then it could be provided through the QUIC implementation's API.¶
If the SCONE metadata is visible to the OS or as part of a socket API, it could be provided to the application via the underlying OS or socket abstraction libraries used by applications.¶
Regarding identification of the application flow type, options for a QUIC API may include adding "SCONE-capable" type of flag or an optional flow-type tag that can be set by applications. Compared to the complexity of existing QUIC APIs, these could be small additions.¶
While Media over QUIC (MoQ) is being defined, it is intended for media streaming over QUIC, which might be applicable to SCONE, in case adaptive rate streams are detected and throttled by CSPs. As yet, there is no standard MoQ API, an MoQ session is currently scoped either to a QUIC connection or a WebTransport session, so it should not be difficult to expose information learned by either transport stack to MoQ applications. Since MoQ applications are media flows, it may be very simple for an application flow type to be conveyed or inferred via an eventual MoQ API..¶
If SCONE signaling is implemented via QUIC packets, this section is likely to be moot. OS APIs would only be important to consider for SCONE metadata to be exposed in the cases that it is discovered through some lower-level possibilities such as IP extension headers or options, UDP options, DHCP, etc.¶
In applicable cases, the OSes supporting SCONE might extend their sets of socket options to support a getsockopt that permits reading SCONE-received metadata, such as a maximum bit rate. Another possibility would be to allow such information to be retrieved via an ioctl on relevant sockets. Several popular libraries and higher-level languages wrap system calls like getsockopt and ioctl already, and could be extended to include SCONE metadata. The "cmsg" interface [RFC2292] provices an example of how complex sets of ancillary data can be exchanged between applications and an OS kernel consistent with the way that socket options are implemented.¶
A challenge might be that the application would likely need to periodically poll the OS for this information, as there may not be a general way to register a callback or hander for when the OS notices a change. This may be one reason, among others, why supporting SCONE signaling via the QUIC connection itself is more desirable.¶
Another challenge relates to information such as a maximum rate per flow type discovered via DHCP or other means. In that case, it may be difficult for the application to know how the network will actually classify its packets, and so it may not be clear what rate is relevant to assume. This could be another argument favoring QUIC-based signaling, in terms of applications being able to directly make productive use of the information.¶
General SCONE security considerations are discussed in the other documents covering specific network-to-host signaling methods. Privacy concerns have also been discussed in [I-D.tomar-scone-privacy].¶
Existing APIs that expose information about the network path to applications have documented security considerations, especially with regard to user privacy. For instance, there may be concerns that such information can be used to assist in fingerprinting users, defeating anonymization, or otherwise exposing more information about the user to the application than the user might explicitly consent to. Such concerns have been document, for example, with regard to the Network Information API.¶
By providing additional information about throttling rate limits within the path, SCONE could increase the amount of information availble on top of that provided by the current APIs. For instance, if the rate information is very fine-grained, it could be useful in fingerprinting.¶
Generally, however, the CSP throttling information is currently very coarse grained, as it is used today. Additionally, the application providers authenticate their users, and their is not an expectation of anonymity in popular platforms today.¶
Beyond this, it is also the case that information provided by SCONE can already be learned by CAP endpoints through various other mechanisms (e.g. the effect of on-path throttlers is clearly visible by observing application traffic packet flows). SCONE simply makes the signaling explicit, rather than requiring it to be observed and inferred separately.¶
This document has no IANA actions.¶
This section to be removed prior to any potential publication within an RFC.¶
The "CSP" term is overloaded, especially with regard to web technology, and might be changed to "carrier", "network operator", etc. in the future, but would need to be consistent with terminology in other SCONE documents.¶