| Internet-Draft | Network Collaboration Requirements | September 2024 |
| Kaippallimalil, et al. | Expires 30 March 2025 | [Page] |
Collaborative signaling from client-to-network and server-to-network can improve the user experience by informing the network about the nature and relative importance of packets (frames, streams, etc.) without having to disclose the content of the packets. Moreover, the collaborative signaling may be enabled so that clients and servers are aware of the network's treatment of incoming packets. Also, client-to-network collaboration can be put in place without revealing the identity of the remote servers. This collaboration allows for differentiated services at the network (e.g., packet discard preference), the sender (e.g., adaptive transmission), or through cooperation of server/client and the network.¶
This document lists some use cases that illustrate the need for a mechanism to share metadata and outlines requirements for both client-to-network and server-to-network. The document focuses on UDP flows bound to the same user.¶
This note is to be removed before publishing as an RFC.¶
Status information for this document may be found at https://datatracker.ietf.org/doc/draft-kwbdgrr-tsvwg-net-collab-rqmts/.¶
Discussion of this document takes place on the TSVWG Working Group mailing list (mailto:tsvwg@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/tsvwg/. Subscribe at https://www.ietf.org/mailman/listinfo/tsvwg/.¶
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 30 March 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.¶
Wireless networks, including 5G and WLAN, inherently experience large variations in link quality over sub-RTT (round-trip time) intervals and on the other hand, applications such as interactive media demand both low latency and high bandwidth.¶
Superior service during adverse network events can be achieved by the sender conveying packet behavior preferences to the network for packets within a single UDP 4-tuple flow. During adverse network events this allows the network to be informed about the least-impactful packets to drop (or delay) in the same UDP 4-tuple flow. Without such signaling, the network can only indiscriminately drop (or delay) packets. With such capability, loss-tolerant and delay-tolerant transport protocols such as RTP [RFC3550], QUIC [RFC9000], and Unreliable QUIC [RFC9221] can inform the network and provide a superior end user experience.¶
Some of the complications that are induced by adverse network events may be eliminated by adequate dimensioning and upgrades. However, such upgrades may not be always (immediately) possible or justified. Complementary mitigations are thus needed to soften these complications by introducing some collaboration between endpoints and networks to adjust their behaviors.¶
Section 3 discusses the rationale for per-packet metadata.¶
Section 5 outlines use cases to illustrate the issues and the need for additional information per flow to allow the network to optimize its handling. Section 6 describes the requirements for on-path media collaboration signals.¶
The document makes use of the following terms:¶
Refers to metadata that doesn't change often during the lifetime of a connection and thus can be exchanged once or as needed. This is communicated per flow (i.e., UDP 4-tuple) between client and network.¶
Examples of such metadata are client request to honor per-packet metadata and preferences.¶
Refers to metadata that varies packet to packet within the same flow, often capturing the nature and characteristics of the traffic each packet carries. This needs to be communicated on a per packet basis.¶
Examples of such metadata are Packet Priority and tolerance to delay¶
Network management actions that are undertaken as a reaction to unplanned overload events. Concretely, this includes policies which react to congestion events with very short to very long durations (e.g., varying wireless and mobile air interface conditions) or protection policies to soften the impact of ongoing attacks.¶
Maximizing network utilization and enhancing perceived end user experience under adverse network conditions are challenging. Factors that affect wireless networks include change in channel conditions, interference between proximate cells, and end user mobility. These variations in link quality can be in the order of a millisecond or less [_5G-Lumos] while congestion control takes several tens of milliseconds (more than one RTT) to estimate data rate over a specific path. Similarly, application servers that encode and serve live or interactive content (media, typically) take time to adjust the encoding level and other processes to match the network rate. End-to-end congestion control algorithms are far from being optimal when the link quality is highly variable in sub-RTT timeframes and the application demands both low latency and high bandwidth (e.g., Section 2.1 of [RFC6077]). In these conditions, applications settle for a lower throughput when latency is prioritized, or for higher throughput at the expense of much higher delays.¶
While rate control based on feedback for a flow (UDP 4-tuple) is evidently not able to adapt for sub-RTT changes in available wireless channel resources, an application server can provide information on a per-packet basis that a network shaper may use to allocate the available resources more effectively. For example, [_5G-Octopus] has shown for volumetric video packets and a rate controller that errs on the side of overestimation, that the network shaper can drop low priority frames of a group of pictures (corresponding to transient wireless link bandwidth drops) and still achieve significantly better performance than state-of-the-art based on feedback. With not fully encrypted packets, networks may use heuristics to build an "implicit signal" gleaned from a packet to prioritize or otherwise shape flows. However, implicit signals are not desirable as they lead to ossification of protocols as result of introducing unintended dependencies [RFC9419]. When packet contents are encrypted, the approach of using implicit signals is no longer viable.¶
Bandwidth constraints exist most predominantly at the access network (e.g., radio access networks). Users who are serviced via these networks use clients which run various applications; each having different connectivity needs for an optimal user experience. These needs are not frozen but change over time depending on the application and even depending on how an application is used (e.g., user's preferences). An explicit signal to the client can help to manage the use of available bandwidth better and better share it with requesting applications.¶
Other applications like interactive media can demand both high throughput and low latency and, in some cases, carry different streams (e.g., audio and video) in a single transport connection (e.g., WebRTC [RFC8825]).¶
With RTP [RFC3550], the media type could be examined and used as an implicit signal for determining relative priority. However, [RFC9335] defines a new mechanism that completely encrypts RTP header extensions and Contributing sources (CSRCs). Furthermore, a full encrypted transport (e.g., QUIC [RFC9000]) does not expose any media header information that on-path network elements can use for forwarding decisions.¶
Figure 1 shows where such bandwidth and performance constraints usually exist with a "B" (for Bottleneck) in 3GPP/mobile networks and WLAN/ISP networks. When a bottleneck exists temporarily, the network has no choice but to discard or delay packets -- which can harm certain flows and, thus, lead to suboptimal perceived experience. In this document, this is termed 'Reactive Management'.¶
Figure 2 shows a bottleneck (access) router on the path of packets from Server to Client. A network shaper in the router manages QoS of flows of multiple users and can buffer (delay), discard, or apply other flow control rules. Application layer signaling and feedback between Client - Server (A - in the figure) adjust transmission rate over a period of several RTTs using feedback and congestion control algorithms. Congestion control algorithms (CCAs) are generally conservative and settle to a steady rate that avoids excessive packet loss. In networks where link conditions (between Client and Router) vary significantly at timescales well below the RTT, this results in unused (wasted) bandwidth at short timescales.¶
There is some research (e.g., [_5G-Octopus]) to indicate that media applications can obtain better Quality of Experience (QoE) when sending at a higher rate (less conservative than current CCA) and the media application is willing to tolerate some packet loss or delay of low priority packets. Packet priority and tolerance to delay of packets in such a case would be provided on-path in a side channel associated to the downstream packet (B - in the figure). The requirements for this server-to-network (S2N) metadata are described in Section 6.2.¶
The client may provide information to an (access) router to drop lower priority marked packets of a flow (UDP 4-tuple) temporarily which can in turn allocate available bandwidth to other flows of that network attachment, especially during a reactive management event.¶
Network shapers observe flows and apply policies to maximize performance but are not aware whether there is a high preference for one flow (UDP 4-tuple) over another flow belonging to the same user and network attachment (e.g., a subscriber connection, a 3GPP PDU Session. See Appendix A for more details). Clients may provide an (access) router with preferences of which flow (UDP 4-tuple) has relative priority among flows of that network attachment using H2N (C - in the figure). Clients may also provide information to an (access) router to drop, when resources are congested, 'lower priority'-marked packets of a flow (UDP 4-tuple) temporarily which can in turn allocate bandwidth to other flows of that same network attachment.¶
In summary, the rapid variation of wireless link quality and/or bandwidth limitations in networks along with interactive applications that demand low latency and high throughput can lead to suboptimal user experience.¶
Server indicates the importance of a packet within a flow. This allows the network to prioritize based on requirement and during reactive events. This priority value may also be used to indicate loss tolerance and the network elements may drop loss tolerant packets during reactive events.¶
This is a per-packet metadata requirement.¶
Metadata to indicate whether the packet can tolerate delay.¶
This is a per-packet metadata requirement.¶
User/Client indicating to the network to honor the application's metadata signaling.¶
This is a per-flow metadata requirement.¶
The ability of the receiver to change the priority by communicating to the network to prioritize one payload(metadata) over another within the flow -- without cooperation of the sender. Gives the sender the ability to have same metadata for all the connections without having to change based on the user preference, aids in scalability.¶
This is a per-flow metadata requirement.¶
API framework to facilitate signaling for applications.¶
Signaling to the network (Section 6.1, Section 6.2) will need to be facilitated by Application Programming Interfaces (APIs) for any application to use them. Signaling and retrieval of the signals may not be performed at a single layer (although not encouraged). Hence, a framework is required to abstract the underlying protocol(s) and allow the application(s) to retrieve/send signals using a single or a set of APIs independent of the channels that are used to convey the signals. The API framework is required even if one single channel is used so that any application on a client can consume the signals.¶
There might be many channels to signal the metadata such as (non-exhaustive list):¶
UDP Options [I-D.ietf-tsvwg-udp-options]¶
IPv6 Hop-by-Hop Options (Section 4.3 of [RFC8200])¶
QUIC CID mapping¶
ICMP messages¶
Streaming video contains the occasional key frame ("I-frame") containing a full video frame. These frames are necessary to rebuild receiver state after loss of delta frames. The key frames are therefore more critical to deliver to the receiver than delta frames.¶
Examples: live broadcast, on-demand video streaming.¶
Use cases:¶
Majority of streaming traffic is Audio and Video traffic. Video contains partial frames and full frames, which need to be distinguished so that full frames can be indicated to the network. Audio traffic is more critical than video for many applications. This differences in priority needs to be indicated to the network to ensure network (de)prioritizes (or even drop if deemed necessary) traffic appropriately.¶
Requirement: REQ-PACKET-PRIORITY.¶
Impact: With the above requirement met, better quality of service could be maintained in resource-constrained networks and during reactive events ensuring better user experience.¶
The server (or relay) sends the same stream to many receivers, including the same metadata (especially with media over QUIC). Different clients have different priorities for different types of traffic. This would result in change in priorities for the same type of traffic that a single server sends, based on the user/client.¶
Requirement: REQ-PAYLOAD-CLIENT-DECIDES.¶
Impact: With the above requirement met, each client/user preferences are prioritized accordingly while maintaining scalability on the server, since the metadata that the server sends still remains the same for all the connections.¶
In loss-prone networks or during Reactive Management events, if all packets of an application flow (UDP 4-tuple) such as live broadcast or on-demand video streaming are treated the same, it limits the ability to maximize network utilization and use the transiently available bandwidth. Dropping or delaying of (media) packets randomly is likely to lower network utilization and application performance.¶
Requirement: REQ-PACKET-PRIORITY, REQ-PACKET-DELAY.¶
Impact: By identifying packets that tolerate being dropped, congestion can be reduced leading to improved performance/quality of service.¶
Interactive media includes content that a user can actively engage with and results in input and response actions that can be highly delay-sensitive. This can also include mixed traffic based on the user activity and interaction.¶
Examples: VoIP (Peer-to-Peer (P2P), group conferencing), gaming, Remote Desktop Virtualization, eXtended Reality (XR).¶
Use cases:¶
A mobile/roaming user prioritizes audio over video during a VoIP call to have a seamless meeting experience.¶
Requirement: REQ-PAYLOAD-CLIENT-DECIDES.¶
Impact: With the above requirement met, each client/user preferences are prioritized accordingly while maintaining scalability on the server.¶
A remote desktop user prioritizes graphics updates over an on-going file copy operation. A user types in/interacts with a document/file after triggering a save file operation, while save operation is on-going.¶
Requirement: REQ-PACKET-PRIORITY, REQ-PACKET-DELAY.¶
Impact: By prioritizing graphic updates/interactive traffic, user interactivity is improved with lesser jitter.¶
A game or VoIP application may want to signal different metadata for the same type of packet in each direction. One user, in a VoIP conference call, wants to prioritize the slide deck being shared while the other wants to prioritize audio and other wants to prioritize video of the speaker. Each user's varied preferences can be catered with same type of metadata originating from the server.¶
Requirement: REQ-PACKET-PRIORITY, REQ-PAYLOAD-CLIENT-DECIDES.¶
Impact: With the above requirement met, each client/user preferences are prioritized accordingly while maintaining scalability on the server.¶
A network glitch while user is in an eXtended Reality application. The traffic comprises of haptic, video, audio, graphics update and keystrokes. During such reactive event, some packets need to be deprioritized/dropped to maintain interactivity.¶
Requirement: REQ-PACKET-PRIORITY, REQ-PACKET-DELAY.¶
Impact: By prioritizing high priority traffic, user's interactive experience is improved with lesser jitter.¶
Currently, some flows are granted higher priority over other flows because of a contractual agreement between the ISP and the content provider. These contracts could be extended to also allow per-packet prioritization within a single UDP 4-tuple, as desired by this document (Section 6.1 and Section 6.1.1).¶
For such applications to benefit from per-packet prioritization within a single UDP 4-tuple, the client needs to determine which per-packet markings are supported by the ISP's network (e.g., encoded into IPv6 Flow Label, UDP Option, or DSCP). Then it can indicate to the ISP's network that a certain UDP 4-tuple will have those markings.¶
Requirements: REQ-API-FRAMEWORK and REQ-CLIENT-DECIDES.¶
Impact: By signaling ISPs to honor the metadata for a particular flow, the client facilitates in the identification of traffic to prioritize for the ISPs. This would enable the ISPs to extend contracts to servers that cannot be identified by using IP addresses, thereby opening up such contracts to more content providers.¶
There are various approaches for collaborative signaling between the server/client and network including out-of-band signaling, client-centric metadata sharing and proxied connections. The requirements here focus on proxied metadata connections on path with the data traffic.¶
The path signals below should follow the principles of intentional distribution, protection of information, minimization and limiting impact as described in [RFC9419] and [RFC8558]. Leveraging previous experience ([RFC9049]), the metadata signals do not need application identity, application cause (or 'reason'), server identity or the inspection of client-to-server encrypted payload.¶
The metadata connections may be between server and network (in either direction) or between client and network (in either direction).¶
Some use cases benefit from server - network metadata exchanges (Section 6.2) and others need client involvement (Section 6.1).¶
For the requirements that follow, the assumption is that the client agrees to the exchange of metadata between the server and network, or between the client and network.¶
Due to contractual agreements (mentioned in Section 4 under REQ-CLIENT-DECIDES) between content providers and ISPs, not all the content providers' signals are honored. However, extending these contracts would disadvantage content providers and other servers that cannot obtain such a contract or have traffic that is difficult or impossible for the ISP to identify and provide such service. This drawback can be overcome using metadata defined in Section 6.1.1.¶
Client authorization signal provides information on whether the ISP should honor the signals sent by the current flow. This would enable the ISP to identify servers, which are difficult or impossible to identify by other available methods today.¶
This signal also serves as the medium to initiate the negotiation discussed in Section 5.3).¶
Encrypting/Obfuscating metadata information is recommended (Section 7.1). The client authorization signal provides the means to convey the key needed by the ISP to decrypt the metadata.¶
REQ-CLIENT-DECIDES is satisfied by signaling Client Flow Authorization as part of client-to-network signal.¶
Application flows (UDP 4-tuple) for live media, eXtended Reality (XR) and gaming require both high bandwidth and low latency and would as such benefit from being able to use the bandwidth available for the flow. In wireless networks, some of the bandwidth available for the flow is not possible to schedule using the feedback based rate control (due to the significant link variations at sub-RTT timescales). In such networks where variations in link quality is well below RTT, congestion control algorithms settle to a steady rate that avoids excessive packet loss. Feedback via ECN/L4S [RFC9331] provides an accurate signal but is also on an RTT timescale and thus does not provide finer resolution information of instantaneous bandwidth available.¶
If application packets can either tolerate delay or some loss of lower priority packets, the network traffic shaper and scheduler can use this information to provide a higher application quality of service. There is some research [_5G-Octopus] to indicate that media applications can obtain better measured application quality when sending at a higher rate (less conservative than current CCA) and allowing the network to delay or drop low priority packets.¶
The metadata in Section 6.2.1 should also satisfy constraints identified in Section 7. Privacy (Section 7.1) requires that metadata should not provide additional information to identify the application or the user. The application server can decide on the metadata values that provide the best handling for the packets of the flow and may not necessarily reflect the exact priority values that allow an on-path observer to perform traffic analysis. This metadata is advisory in nature and network traffic policy that restricts its use would not result in additional issues. Other constraints including scale (Appendix D.6) and continuity (Appendix D.3) are required for Section 6.2.1.¶
Realizing the additional bandwidth potential with these metadata may require a higher sending rate for the transport flow. This requires work that is not specified in this document. Similarly, the assumption is that network shapers and schedulers can use the metadata in Section 6.2.1 but further details are out of scope.¶
Previous work in [TR.23.700-70-3GPP] has identified the general problem in this section. However, the solution in [TS.23.501-3GPP] is specific to a 5G network. The metadata sent from a (dedicated 5G) application server identifies PDU set information and end-of-burst signals which are not understood by non-3GPP systems such as Wi-Fi or DOCSIS. Further, 3GPP functions and policy configurations are required since this is a 5G specific solution. The metadata disclosed in the 5G solution also identifies frame boundaries and does not fully conform to the constraints for privacy or minimality of data identified in Section 7.¶
Per-packet priority information provides the priority level of one packet relative to other packets within a transport flow (UDP 4-tuple). When a packet is marked with high priority, the expectation is that the network will give high importance to forwarding the packet and when a packet is marked with lower priority, the network will drop the packet in the presence of severe congestion or limited bandwidth. The application server can decide on the priority or importance values that provide the best handling for the packets of the transport flow and may not necessarily reflect the exact priority values that allow an on-path observer to perform traffic analysis. When more than one application stream (e.g., video, audio) is sent on the same transport flow, the application server decides the best allocation of priority values across the different streams of the flow.¶
Per-packet priority or importance determines the drop priority of a packet.¶
REQ-PACKET-PRIORITY is satisfied by signaling Packet Priority as part of server-to-network metadata.¶
Some packets of a media flow (UDP 4-tuple) can tolerate more delay over the wire than others (e.g., packets carrying live media frames require very low latency while packets carrying a background image for augmented reality can tolerate more delay). The objective of this metadata is to indicate that these packets can tolerate a limited amount of delay when there is severe congestion or limited bandwidth. Similar to the LE PHB [RFC8622] for flows, the expectation is that in this case, each packet marked with this metadata is dropped only when there is excessive delay. As with per-packet priority in Section 6.2.1, the application server can decide on the metadata values that provide the best handling for the packets of the transport flow and may not necessarily reflect the exact delay tolerance values that allow an on-path observer to perform traffic analysis.¶
REQ-PACKET-DELAY is satisfied by signaling Tolerance to Delay as part of server-to-network metadata.¶
Traffic policing and shaping are enforced in ingress/egress network points for various reasons (protect the network against attacks, ensure conformance with a traffic profile, etc.). Out-of-profile traffic may be discarded or assigned another class (e.g., using Lower Effort Per-Domain Behavior (LE PDB) [RFC3662]) a bandwidth limit among others. The exact behavior is policy-based and deployment-specific.¶
The entire set of operations to manage traffic is beyond the scope of this document. This section focuses on operational constraints that impact server-network, and client-network modes of sending metadata.¶
Encrypted media payloads along with temporary IPv6 addresses between a server and user (client) provide a measure of privacy for the content and the identity of the user. It should, however, be noted that media flows (e.g., encrypted video payloads in SRTP) exhibit a pattern of bursts and intervals that amounts to a signature and is vulnerable to frequency analysis.¶
To avoid this kind of frequency analysis, media sent by the server would need to be scheduled or multiplexed differently to each user/recipient. This may be possible in transports like QUIC which allows flexibility in scheduling each stream. Transports like QUIC also fully encrypt the entire stream and, therefore, no media headers are observable on-path either. The security aspects of the media payload / transport are not in the scope of this document and is described here only to provide context for metadata privacy.¶
Some metadata (e.g., the size of a burst of packets, sequence number, and timestamp) can be readily observed or inferred by entities along the network path. However, it is essential to recognize that while sequence numbers and timestamps are typically visible in the clear-text headers of protocols (e.g., TCP, RTP, or SRTP) they are not directly observable in encrypted protocols such as QUIC. All metadata sent from the server to the network, including these elements and others, are vulnerable to modification while in transit. Only an on-path attacker can modify on-path metadata. Such an attacker could engage in other malicious activities, like corrupting the checksum or dropping the packet. For instance, an active attacker could alter the metadata to mislabel important packets as unimportant, but such changes are detectable by the receiver.¶
It is recommended to encrypt or obfuscate the metadata information so it is only available to the server, client, and authorized network elements. The method of encryption or obfuscation is not described in this document but rather in other documents that specify how this metadata is encoded and exchanged amongst client, server, and network elements. The privacy implications of revealing metadata to network elements need to be thoroughly analyzed. This analysis should ensure that any exposure of metadata does not compromise user privacy or allow unauthorized entities to infer sensitive information about the data being transmitted while maintaining minimal resource consumption. There is a tension between resource consumption (and thus efficiency) of such encryption and the user's privacy and this tension depends on the threat model (Section 7.4 of [RFC6973]); the threat of a network provider building a subscriber profile of viewed video content is different from the threat of an interactive voice or video call. To mitigate traffic analysis, the sender might, for example, purposefully mis-mark metadata in some packets, add some randomness to avoid recurrent traffic patterns, etc.¶
REQ-PRIVACY-ADDITIONAL and REQ-SIGNALING-AVOIDANCE are satisfied by not revealing any information that could identify the application's identity, reason to signal, server identity, and securing the metadata.¶
Application feedback with measurements of packets lost and delay incurred may affect the sending rate and other behavior of the application. The requirements and specification to mitigate these aspects are not in the scope of this document.¶
None.¶
Security aspects for the metadata are discussed in Section 7.1. The principles outlined in [RFC8558], [RFC9049] and [RFC9419] contain security considerations and are referenced in Section 6.¶
Since the document focuses only on priorities within a flow (not specifying inter-flow priority), the document does not induce concerns related to a specific user or client declaring all flows or a subset of them as being more important. Such abuse concerns are thus not applicable.¶
This document has no other security considerations.¶
This document is a merge of [I-D.rwbr-tsvwg-signaling-use-cases] and [I-D.kaippallimalil-tsvwg-media-hdr-wireless].¶
T. Reddy contributed text and ideas to this document.¶
Xavier De Foy and the authors of this draft have discussed the similarities and differences of this draft with the MoQ draft for carrying media metadata.¶
The authors wish to thank Mike Heard, Sebastian Moeller and Tom Herbert for discussions on metadata fields, fragmentation and various transport aspects.¶
The authors appreciate input from Marcus Ilhar and Magnus Westerlund on the need to address privacy in general and Dan Druta to consider a common transport across various client/server to network signaling when possible. Ruediger Geib suggested that limiting the amount of state information that a wireless router has to keep for a flow should be minimized.¶
Ingemar Johansson's suggestions on fast fading (which L4S handles) and dramatic drops in wireless accesses have been helpful to identify the issues. Thanks to Hang Shi for the review and comments on client-to-network signaling. Thanks to Luis Miguel Contreras, Colin Kahn, Marcus Ilhar and Tianji Jiang for their review and comments.¶
A network attachment represents the communication link between clients and network (router) over which a connection policy (including QoS) is applied to flows within that network attachment.¶
A network attachment may be established using control plane signaling between the client and the network (access) router and is out of scope of this document. Transport flows over a network attachment may consist of multiple streams such as video or audio. Figure 4 shows a high level view of network attachments, flows, and QoS/policy discussed in Section 6.¶
The requirements in Section 6 apply to data units like frames within a flow, but not between flows. Specifically, this document does not discuss flows of distinct clients/users.¶
Figure 4 shows "Client-1" and "Client-2" which negotiate connection policy (e.g., QoS) and other aspects like mobility handling, charging applied to flows in that network attachment. "Client-1" has "flow-x1" and "flow-x2" over its network attachment while "Client-2" has "flow-x3". The requirements in this document focus on on-path collaboration signals that apply to data units such as media frames within flows like "flow-x1/x2/x3" but not between them.¶
Audio can be prioritized differently than video.¶
This requirement may be generalized to non-media packet types.¶
This is an enhanced requirement that requires e2e application layer signaling (out of scope here) to identify of frame boundaries and may not be suitable in cases which are sensitive to traffic analysis (see REQ-SIGNALING-AVOIDANCE and [RFC9049]). If the application provides frame boundaries, the client signals the enhanced application priority values in REQ-PAYLOAD-CLIENT-DECIDES.¶
This is a per-flow metadata requirement.¶
Video contains partial frames and full frames, which need to be distinguished so that full frames can be indicated to the network.¶
This is an enhanced requirement that requires e2e application layer signaling (out of scope here) to identify of frame boundaries and may not be suitable in cases which are sensitive to traffic analysis (see REQ-SIGNALING-AVOIDANCE and [RFC9049]). If the application provides frame boundaries, the client signals the enhanced application priority values in REQ-PAYLOAD-CLIENT-DECIDES.¶
This is a per-packet metadata requirement.¶
A mechanism to signal the available network throughput to interested clients, including changes to throughput.¶
The network shall inform the endpoint of the Rate limiting policies.¶
Means to expose the signal independent of the application should be considered. An example of such exposure is OS APIs.¶
During detected reactive events, the network implements a reactive traffic policy to reduce or offload some of the traffic.¶
This may involve utilizing alternative network attachments available to the client (e.g., Wi-Fi).¶
Handover from one radio or router to another should continue to provide same service level.¶
Signaling should support Multiple bottlenecks.¶
The network must identify multiple bottlenecks, including those within the ISP and subscriber networks, ensuring all bottlenecks benefit from network/client collaboration to enhance overall performance.¶
Means to characterize the scope of a shared metadata for the sake of better interoperability should be supported.¶
The metadata can be used by the network to prioritize traffic within a single 5-tuple connection and metadata cannot be leveraged for prioritization between different flows.¶
The network should use a single channel for sharing metadata to simplify the process and avoid the need for redundant functions.¶
The metadata and other state information that a router has to maintain for each additional flow it handles should be kept to a minimum or eliminated altogether.¶
Streaming video contains the occasional key frame ("I-frame") containing a full video frame. These frames are necessary to rebuild receiver state after loss of delta frames. The key frames are therefore more critical to deliver to the receiver than delta frames.¶
Examples:¶
Audio is more critical than video for many applications and should be prioritized differently than video.¶
Requirement: REQ-MEDIA-AV-SEPARATE.¶
Video contains partial frames and full frames, which need to be distinguished so that full frames can be indicated to the network.¶
Requirement: REQ-MEDIA-KEYFRAME.¶
There are cases (crisis) where "normal" network resources cannot be used at maximum and, thus, a network would seek to reduce or offload some of the traffic during these events -- often called 'reactive traffic policy'. An example of such use case is cellular networks that are overly used (and radio resources exhausted) such as a large collection of people (e.g., parade, sporting event), or such as a partial radio network outage (e.g., tower power outage). During such a condition, an alternative network attachment may be available to the client (e.g., Wi-Fi).¶
Network-to-client signals are useful to put in place adequate traffic distribution policies on the client (e.g., prefer the use of alternate paths, offload a network).¶
Requirement: REQ-NETWORK-SEEKS-LOAD-DOWN.¶
Bandwidth constraints exist most predominantly at the access network. This can be constraints in the network itself or a result of rate limiting due to various reasons.¶
Also, traffic exchanged over a network attachment may be subject to rate-limit policies. These policies may be intentional policies (e.g., enforced as part of the activation of the network attachment and typically agreed upon service subscription) or be reactive policies (e.g., enforced temporarily to manage an overload or during a DDoS attack mitigation).¶
Requirements: REQ-NETWORK-THROUGHPUT, REQ-NRLP.¶
Use cases:¶
Performance Optimization: Some applications support some forms of bandwidth measurements (e.g., [app-measurement]) which feed how the content is accessed to using ABR. Complementing or replacing these measurements with explicit signals will improve overall network performance and can help optimize the data transfer. Signaling bandwidth availability allows endpoints to avoid contributing to network congestion. When the network informs the endpoint about available bandwidth, the endpoint can dynamically adjust its data transmission rate. Knowing available bandwidth helps the endpoint allocate resources efficiently. Cloud-based applications can auto-scale based on available bandwidth.¶
Rate Limiting: Monthly data quotas on cellular networks can be easily exceeded by video streaming, in particular, if the client chooses excessively high quality or routinely abandons watching videos that were downloaded. The network can assist the client by informing the client of the network's bandwidth policy.¶
Applications that have access to a resource-quota information may adopt an aggressive behavior (compared to those that don't have access) if they assumed that a resource-quota like metadata is for the application, not for the client that runs the applications.¶
This is challenging for home networks where multiple clients may be running behind the same CPE, with each of them running a video application. The same challenge may apply when tethering is enabled.¶
Requirement: REQ-SIGNAL-EXPOSURE-FAIRNESS.¶
If distinct channels are used to share the metadata between a client and a network, a network that engages in the collaborative signaling approach will require sophisticated features to classify flows and decide which channel is used to share metadata so that it can consume that information.¶
Likewise, the network will require to implement redundant functions; for each signaling interface.¶
As such, application- and protocol-specific signaling channels are suboptimal. (REQ-SINGLE-CHANNEL)¶
Requirement: REQ-SINGLE-CHANNEL is preferred.¶
The frequency of handovers increases when a user moves faster or when the media session lasts longer. During handovers, there should be minimal delay incurred during handover in configuring/setting up the metadata of a media session in progress.¶
Requirement: REQ-CONTINUITY.¶
Whereas models often show a single bottleneck, there are frequently two (or more) bottlenecks: the ISP network and within the subscriber network (e.g., Wi-Fi link). As such, all bottlenecks near the subscriber should be able to benefit from network/client collaboration.¶
Requirement: REQ-MULTIPLE-BOTTLENECKS.¶
An operational challenge for sharing resource-quota like metadata (e.g., maximum bitrate) is that the network is generally not entitled to allocate quota per-application, per-flow, per-stream, etc. that delivered as part of an Internet connectivity service. However, the network has a visibility about the overall network attachment (e.g. inbound/outbound bandwidth discussed in [I-D.ietf-opsawg-teas-attachment-circuit]).¶
Hints about resource-like metadata is bound by default to the overall network attachment, not specific to a given application or flow.¶
It is out of the scope of this document to discuss setups (e.g., 3GPP PDU Sessions) where network attachments with Guaranteed Bit Rate (GBR) for specific flows is provided.¶
Requirement: REQ-SCOPED-METADATA.¶
There may be a large number of flows handled by the server and wireless/access router. Per flow information (state) at a wireless router for optimizing the flow can negate the advantages offered as the number of flows handled increases.¶
Requirement: REQ-ISP-SCALE.¶