MOQ Extensions for Relays to Support High-Throughput Low-Latency Traffic

Introduction Media flows can be carried over wireless networks, which can be challenging especially for applications with high-throughput and low latency requirements, such as video conferencing and Extended Reality (XR, presented for example in ). Wireless networks can implement techniques to improve network capacity and energy efficiency, as well as reduce the impact of packet losses on users, as illustrated here for XR traffic support in 5G:

The network can handle groups of packets based on how critical they are to the user experience. Some groups of data packets hold application data units that are handled together (e.g., decoded) by the application. 3GPP defines the term "PDU set" to identify these groups of data packets carrying the payload of an application data unit , which can correspond to the data packets of a Network Application Layer (NAL) unit. Application data units can depend on other application data units to be handled or decoded by the application (e.g., P-frames depend on I-frames, and higher layers depend on lower layers). The network can perform differentiated handling of groups of data packets, e.g., prioritize some groups over others in case of congestion. The network can also selectively drop data packets that depend on an already lost application data unit.
The network can limit wake-up time (of radios) to transmit and receive data. For this, it should know as much as possible about traffic patterns. The scheduler can benefit from information on the size and periodicity of traffic, as well as delay budget and expected jitter specific to the application.

Traffic handling of high-throughput low-latency traffic therefore includes differentiated handling of groups of packets, as well as configuring lower-layer scheduling. To perform this, a network node can act as, or communicate with, a MOQ relay to obtain the metadata it needs, associated with media data. It is also necessary for the media sender to identify application data units that correspond to different desired traffic handling (e.g., different layers within a frame), and provide associated metadata. describes a MOQ relay controlling traffic handling in an access network (for example, for media streams sent by B towards A and C). For privacy and security, it is desirable that the MOQ relay, which can be operated by a network or service operator, does not have access to media data and to metadata not related to its operation. For interoperability, it is also desirable for these mechanisms to not be codec specific.

Traffic Handling by Access Networks using a MOQ Relay. | |<---->| B | +---+ +----+------+ | | +---+ +---------+ | | MOQ Relay | +---------+ | +---+ +----+------+ | | +---+ | C |<-|Access Node|->| |<---->| D | +---+ +-----------+ +------------+ +---+ ]]>

XR Traffic Handling in 5G Networks A MOQ relay located at the ingress point of a wireless network, for example within User Plane Function (UPF) or 5G device, can collect metadata and use it to handle XR traffic:

Convey the collected metadata to the Radio Access Network (RAN), using GTP-U headers encapsulating the data packets it forwards to the RAN (e.g., for dynamic metadata such as packet sequence number within the group, priority/importance, dependency information, size, group ID). This makes it possible for the RAN to perform differentiated handling on the packets. The MOQ relay can also convey some metadata related to a flow in control messages to the RAN (e.g., periodicity, delay budget, jitter range). This makes it possible for the RAN to configure efficient scheduling for the flow.
Use the collected metadata to perform some local processing (on the UPF or 5G device): locally prioritize groups of packets based in case of congestion or drop packets that may no longer contribute to a user's perceived QoE, because they can't be decoded (missing dependency) or exceeded a delay budget; select a path when using multipath.

Based on the outcome of a study (reaching completion the time of writing), 3GPP identifies the following metadata for handling XR traffic (the agreed text is available at until its integration in the report). Data plane metadata is expected to be obtained, for the time being, from RTP/SRTP and their extensions headers, or alternatively, from other methods not standardized by 3GPP. Note that it's possible that future releases of the 3GPP standard reference other methods, such as one based on MOQ.

The following metadata was agreed to be used in the data plane:
- ID of a group of packets that share similar handling by the network (a "PDU set")
- Indication of the first and last data packet in a group (optional)
- A sequence number of individual packets within the group
- Size of a group in number of octets or data packets (optional)
- Group importance relative to other groups
Some data plane metadata was not agreed for release 18 but may be considered in future releases, including dependency information between groups
The following metadata was agreed to be used in the control plane, where it is provisioned by the service operator.
- QoS parameters: target delay budget and error rate for groups
- Burst periodicity
- Whether all data packets are needed to process the application data unit carried by the group

Related Handling by RTP Switches The RTP extension defined in enables an RTP switch (a common short term for various types of middleboxes) to access critical information about video frames. The processing by the RTP switch, similarly to the traffic handling discussed above for access networks, applies to packets associated with encrypted application data units. Groups of packets typically carry a specific layer in a frame. The examples of processing given in the framemarking draft are:

Start/stop forwarding streams at specified boundaries (e.g., on an intra-frame).
Determine which frame to drop with minimal impact to video quality.
For scalable streams with dependent layers, selectively forward specific layers to specific recipients due to recipient bandwidth of decoder limits.

The metadata enabling this behavior is also similar what has been identified for XR traffic handling in 5G:

Frame start and end flags (indication of the first and last data packet in a group)
Priority/importance of groups of data packets is expressed using flags: Independent Frame flag, and Discardable Frame flag.
Dependency information between groups of data packets is expressed through: Temporal Layer 0 Picture Index, Base Layer Sync flag, Temporal ID, Layer ID.

An Extension for Traffic Handling of High-Throughput Low-Latency Traffic A MOQ extension is proposed to implement this type of handling by the network, since the required metadata will not be needed in all cases. Note: it may be sufficient to mark the related metadata fields as "optional", however using an explicit extension helps telling endpoints which fields are expected by relays on the path. A MOQ protocol extension (or multiple extensions sharing similar design characteristics) can be defined to cover multiple use cases as discussed in 2.1 and 2.2.

Endpoint Behavior To use the proposed extension, the endpoints are expected to perform the following:

The application sends media data units, defined in an application specific way. Groups can be identified by stream ID, datagram context ID, or using a dedicated group ID present in datagrams.
The application also sends metadata related to the service:
- "Per data unit" metadata is expected to change from one media data unit to the next, and can include a data unit ID, size, importance, dependency information, etc. The relay should receive this metadata before its related media data. If media data is transported in streams, per data unit metadata is sent before media in the stream. If media data is transported in datagrams, per data unit metadata is sent in a reliable message, which helps reduce the overhead per datagram.
- "Per flow" metadata is sent at the beginning of the media flow, and may be updated later during the flow lifetime. This can include metadata related to delay budget, error rate, burst periodicity.
- When media data is sent in datagrams, the application sends metadata associated with individual datagrams. This includes metadata that can vary per datagram, such as a datagram sequence number within the media data unit, indication of first or last datagram. Some per data unit metadata may also be sent in individual datagrams if it enables a more efficient handling without too much overhead.
All related metadata is accessible by the MOQ relays providing the service (i.e., cleartext field not end-to-end encrypted but still protected by the QUIC connection).

Impact on the Base MOQ Protocol A few requirements are proposed on the base MOQ protocol to enable the type of protocol extensions described in this document.

The base MOQ protocol should support negotiating the use of a protocol extension.
- An extension corresponds to a set of supported messages, fields and associated endpoint behavior
- The client should learn of the extensions supported by a MOQ relay out-of-band
- MOQ relays should be involved in extension negotiation, at least to be able to reject a service request if a necessary extension is not present in the request

Potential mechanism: extensions can be negotiated using HTTP headers in the extended CONNECT request that initiates the WebTransport session.

Assuming a stream based MOQ base protocol:
- The extension mechanism should support defining cleartext fields (i.e., not end-to-end encrypted) in a stream, placed before encrypted media data.
Assuming a datagram based MOQ base protocol:
- The extension mechanism should support defining cleartext fields in a stream. This stream contains a cleartext ID that is also present in all related datagrams. This ID can be the datagram context ID, or another group ID present in all datagrams.
- The extension mechanism should support defining cleartext fields present in individual datagrams.
The extension mechanism should support defining cleartext fields in reliable messages, that relate to a media flow (e.g., using a cleartext ID that is present in all streams/datagrams carrying data for this media flow)

Potential mechanism: cleartext fields (TLVs or frames) can be defined as part of the extension and sent by endpoints in streams and datagrams (before the encrypted media payload if any).

Detailed Description TBD (for example: define extension name, define new cleartext fields needed by the extension, make some optional MOQ fields mandatory when the extension is used, define new messages, etc.)

Security Considerations To enable support for the feature described in this document, the application exposes metadata to a MOQ relay under the control of a network or service operator. The encrypted media is not exposed to the MOQ relay. The level of exposure is similar to the Frame Marking RTP extension .

Acknowledgments Thanks to Jaya Rao, Ghyslain Pelletier and Renan Krishna for the discussions and comments.