Simula Research Lab

ietf@bobbriscoe.net http://bobbriscoe.net/

Nokia Bell Labs

Antwerp Belgium koen.de_schepper@nokia.com https://www.bell-labs.com/usr/koen.de_schepper

Universidad Carlos III de Madrid

Av. Universidad 30 Leganes, Madrid 28911 Spain 34 91 6249500 marcelo@it.uc3m.es http://www.it.uc3m.es

Transport Transport Area Working Group Internet-Draft I-D This document describes the L4S architecture for the provision of a new Internet service that could eventually replace best efforts for all traffic: Low Latency, Low Loss, Scalable throughput (L4S). It is becoming common for all (or most) applications being run by a user at any one time to require low latency. However, the only solution the IETF can offer for ultra-low queuing delay is Diffserv, which only favours a minority of packets at the expense of others. In extensive testing the new L4S service keeps average queuing delay under a millisecond for all applications even under very heavy load, without sacrificing utilization; and it keeps congestion loss to zero. It is becoming widely recognized that adding more access capacity gives diminishing returns, because latency is becoming the critical problem. Even with a high capacity broadband access, the reduced latency of L4S remarkably and consistently improves performance under load for applications such as interactive video, conversational video, voice, Web, gaming, instant messaging, remote desktop and cloud-based apps (even when all being used at once over the same access link). The insight is that the root cause of queuing delay is in TCP, not in the queue. By fixing the sending TCP (and other transports) queuing latency becomes so much better than today that operators will want to deploy the network part of L4S to enable new products and services. Further, the network part is simple to deploy - incrementally with zero-config. Both parts, sender and network, ensure coexistence with other legacy traffic. At the same time L4S solves the long-recognized problem with the future scalability of TCP throughput. This document describes the L4S architecture, briefly describing the different components and how the work together to provide the aforementioned enhanced Internet service.

It is increasingly common for all of a user's applications at any one time to require low delay: interactive Web, Web services, voice, conversational video, interactive video, interactive remote presence, instant messaging, online gaming, remote desktop, cloud-based applications and video-assisted remote control of machinery and industrial processes. In the last decade or so, much has been done to reduce propagation delay by placing caches or servers closer to users. However, queuing remains a major, albeit intermittent, component of latency. For instance spikes of hundreds of milliseconds are common. During a long-running flow, even with state-of-the-art active queue management (AQM), the base speed-of-light path delay roughly doubles. Low loss is also important because, for interactive applications, losses translate into even longer retransmission delays. It has been demonstrated that, once access network bit rates reach levels now common in the developed world, increasing capacity offers diminishing returns if latency (delay) is not addressed. Differentiated services (Diffserv) offers Expedited Forwarding for some packets at the expense of others, but this is not applicable when all (or most) of a user's applications require low latency. Therefore, the goal is an Internet service with ultra-Low queueing Latency, ultra-Low Loss and Scalable throughput (L4S) - for all traffic. A service for all traffic will need none of the configuration or management baggage (traffic policing, traffic contracts) associated with favouring some packets over others. This document describes the L4S architecture for achieving that goal. It must be said that queuing delay only degrades performance infrequently . It only occurs when a large enough capacity-seeking (e.g. TCP) flow is running alongside the user's traffic in the bottleneck link, which is typically in the access network. Or when the low latency application is itself a large capacity-seeking flow (e.g. interactive video). At these times, the performance improvement from L4S must be so remarkable that network operators will be motivated to deploy it. Active Queue Management (AQM) is part of the solution to queuing under load. AQM improves performance for all traffic, but there is a limit to how much queuing delay can be reduced by solely changing the network; without addressing the root of the problem. The root of the problem is the presence of standard TCP congestion control (Reno ) or compatible variants (e.g. TCP Cubic ). We shall call this family of congestion controls 'Classic' TCP. It has been demonstrated that if the sending host replaces Classic TCP with a 'Scalable' alternative, when a suitable AQM is deployed in the network the performance under load of all the above interactive applications can be stunningly improved. For instance, queuing delay under heavy load with the example DCTCP/DualQ solution cited below is roughly 1 millisecond (1 ms) at the 99th percentile without losing link utilization. This compares with 5 to 20 ms on average with a Classic TCP and current state-of-the-art AQMs such as fq_CoDel  or PIE . Also, with a Classic TCP, 5 ms of queuing is usually only possible by losing some utilization. It has been convincingly demonstrated that it is possible to deploy such an L4S service alongside the existing best efforts service so that all of a user's applications can shift to it when their stack is updated. Access networks are typically designed with one link as the bottleneck for each site (which might be a home, small enterprise or mobile device), so deployment at a single node should give nearly all the benefit. The L4S approach requires component mechanisms in different parts of an Internet path to fulfill its goal. This document presents the L4S architecture, by describing the different components and how they interact to provide the scalable low-latency, low-loss, Internet service.

There are three main components to the L4S architecture (illustrated in ): The L4S service traffic needs to be isolated from the queuing latency of the Classic service traffic. However, the two should be able to freely share a common pool of capacity. This is because there is no way to predict how many flows at any one time might use each service and capacity in access networks is too scarce to partition into two. So a 'semi-permeable' membrane is needed that partitions latency but not bandwidth. The Dual Queue Coupled AQM is an example of such a semi-permeable membrane.Per-flow queuing such as in could be used, but it partitions both latency and bandwidth between every end-to-end flow. So it is rather overkill, which brings disadvantages (see ), not least that thousands of queues are needed when two are sufficient. A host needs to distinguish L4S and Classic packets with an identifier so that the network can classify them into their separate treatments. considers various alternative identifiers, and concludes that all alternatives involve compromises, but the ECT(1) codepoint of the ECN field is a workable solution. Scalable congestion controls already exist. They solve the scaling problem with TCP first pointed out in . The one used most widely (in controlled environments) is Data Centre TCP (DCTCP ), which has been implemented and deployed in Windows Server Editions (since 2012), in Linux and in FreeBSD. Although DCTCP as-is 'works' well over the public Internet, most implementations lack certain safety features that will be necessary once it is used outside controlled environments like data centres (see later). A similar scalable congestion control will also need to be transplanted into protocols other than TCP (SCTP, RTP/RTCP, RMCAT, etc.)

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in . In this document, these words will appear with that interpretation only when in ALL CAPS. Lower case uses of these words are not to be interpreted as carrying RFC-2119 significance. COMMENT: Since this will be an information document, This should be removed. The 'Classic' service is intended for all the congestion control behaviours that currently co-exist with TCP Reno (e.g. TCP Cubic, Compound, SCTP, etc). The 'L4S' service is intended for traffic from scalable TCP algorithms such as Data Centre TCP. But it is also more general—it will allow a set of congestion controls with similar scaling properties to DCTCP (e.g. Relentless ) to evolve.Both Classic and L4S services can cope with a proportion of unresponsive or less-responsive traffic as well (e.g. DNS, VoIP, etc). A congestion control where the packet flow rate per round trip (the window) is inversely proportional to the level (probability) of congestion signals. Then, as flow rate scales, the number of congestion signals per round trip remains invariant, maintaining the same degree of control. For instance, DCTCP averages 2 congestion signals per round-trip whatever the flow rate. A congestion control with a flow rate compatible with standard TCP Reno . With Classic congestion controls, as capacity increases enabling higher flow rates, the number of round trips between congestion signals (losses or ECN marks) rises in proportion to the flow rate. So control of queuing and/or utilization becomes very slack. For instance, with 1500 B packets and an RTT of 18 ms, as TCP Reno flow rate increases from 2 to 100 Mb/s the number of round trips between congestion signals rises proportionately, from 2 to 100. The default congestion control in Linux (TCP Cubic) is Reno-compatible for most Internet access scenarios expected for some years. For instance, with a typical domestic round-trip time (RTT) of 18ms, TCP Cubic only switches out of Reno-compatibility mode once the flow rate approaches 1 Gb/s. For a typical data centre RTT of 1 ms, the switch-over point is theoretically 1.3 Tb/s. However, with a less common transcontinental RTT of 100 ms, it only remains Reno-compatible up to 13 Mb/s. All examples assume 1,500 B packets. The original proposed standard Explicit Congestion Notification (ECN) protocol , which requires ECN signals to be treated the same as drops, both when generated in the network and when responded to by the sender. A home, mobile device, small enterprise or campus, where the network bottleneck is typically the access link to the site. Not all network arrangements fit this model but it is a useful, widely applicable generalisation.

The L4S architecture is composed of the following elements. Protocols:The L4S architecture encompasses the two protocol changes (an unassignment and an assignment) that we describe next: An essential aspect of a scalable congestion control is the use of explicit congestion signals rather than losses, because the signals need to be sent immediately and frequently—too often to use drops. 'Classic' ECN requires an ECN signal to be treated the same as a drop, both when it is generated in the network and when it is responded to by hosts. L4S needs networks and hosts to support two separate meanings for ECN. So the standards track needs to be updated to allow L4S packets to depart from the 'same as drop' constraint. has been prepared as a standards track update to relax specific requirements in RFC 3168 (and certain other standards track RFCs), which clears the way for the experimental changes proposed for L4S. also explains why the original experimental assignment of the ECT(1) codepoint as an ECN nonce is being reclassified as historic (it was never deployed, and it offers no security benefit now that deployment is optional). recommends ECT(1) is used as the identifier to classify L4S packets into a separate treatment from Classic packets. This satisfies the requirements for identifying an alternative ECN treatment in . Network components:The Dual Queue Coupled AQM has been specified as generically as possible as a 'semi-permeable' membrane without specifying the particular AQMs to use in the two queues. An informational appendix of the draft is provided for pseudocode examples of different possible AQM approaches. Initially a zero-config variant of RED called Curvy RED was implemented, tested and documented. The aim is for designers to be free to implement diverse ideas. So the brief normative body of the draft only specifies the minimum constraints an AQM needs to comply with to ensure that the L4S and Classic services will coexist. For instance, a variant of PIE called Dual PI Squared has been implemented and found to perform better than Curvy RED over a wide range of conditions, so it has been documented in a second appendix of . Host mechanisms: The L4S architecture includes a number of mechanisms in the end host that we enumerate next: Data Centre TCP is the most widely used example of a scalable congestion control. It is being documented in the TCPM WG as an informational record of the protocol currently in use . It will be necessary to define a number of safety features for a variant usable on the public Internet. A draft list of these, known as the TCP Prague requirements, has been drawn up (see Appendix A of ). The list also includes some optional performance improvements. Transport protocols other than TCP use various congestion controls designed to be friendly with Classic TCP. Before they can use the L4S service, it will be necessary to implement scalable variants of each of these congestion control behaviours. The following standards track RFCs currently define these protocols: ECN in TCP , in SCTP , in RTP , and in DCCP . Not all are in widespread use, but those that are will eventually need to be updated to allow a different congestion response, which they will have to indicate by using the ECT(1) codepoint. Scalable variants are under consideration for some new transport protocols that are themselves under development, e.g. QUIC and certain real-time media congestion avoidance techniques (RMCAT) protocols. ECN feedback is sufficient for L4S in some transport protocols (RTCP, DCCP) but not others: For the case of TCP, the feedback protocol for ECN embeds the assumption from Classic ECN that an ECN mark is the same as a drop, making it unusable for a scalable TCP. Therefore, the implementation of TCP receivers will have to be upgraded . Work to standardize more accurate ECN feedback for TCP (AccECN ) is in progress. ECN feedback is only roughly sketched in an appendix of the SCTP specification. A fuller specification has been proposed , which would need to be implemented and deployed before SCTCP could support L4S.

Explicit congestion signalling is a key part of the L4S approach. In contrast, use of drop as a congestion signal creates a tension because drop is both a useful signal (more would reduce delay) and an impairment (less would reduce delay). Explicit congestion signals can be used many times per round trip, to keep tight control, without any impairment. Under heavy load, even more explicit signals can be applied so the queue can be kept short whatever the load. Whereas state-of-the-art AQMs have to introduce very high packet drop at high load to keep the queue short. Further, when using ECN TCP's sawtooth reduction can be smaller, and therefore return to the operating point more often, without worrying that this causes more signals (one at the top of each smaller sawtooth). The consequent smaller amplitude sawteeth fit between a very shallow marking threshold and an empty queue, so delay variation can be very low, without risk of under-utilization. All the above makes it clear that explicit congestion signalling is only advantageous for latency if it does not have to be considered 'the same as' drop (as required with Classic ECN ). Therefore, in a DualQ AQM, the L4S queue uses a new L4S variant of ECN that is not equivalent to drop , while the Classic queue uses either classic ECN or drop, which are equivalent.Before Classic ECN was standardized, there were various proposals to give an ECN mark a different meaning from drop. However, there was no particular reason to agree on any one of the alternative meanings, so 'the same as drop' was the only compromise that could be reached. RFC 3168 contains a statement that: "An environment where all end nodes were ECN-Capable could allow new criteria to be developed for setting the CE codepoint, and new congestion control mechanisms for end-node reaction to CE packets. However, this is a research issue, and as such is not addressed in this document." Using just two queues is not essential to L4S (more would be possible), but it is the simplest way to isolate all the L4S traffic that keeps latency low from all the legacy Classic traffic that does not.Similarly, coupling the congestion notification between the queues is not necessarily essential, but it is a clever and simple way to allow senders to determine their rate, packet-by-packet, rather than be overridden by a network scheduler. Because otherwise a network scheduler would have to inspect at least transport layer headers, and it would have to continually assign a rate to each flow without any easy way to understand application intent. Once there are at least two separate treatments in the network, hosts need an identifier at the IP layer to distinguish which treatment they intend to use. A scalable congestion control keeps the signalling frequency high so that rate variations can be small when signalling is stable, and rate can track variations in available capacity as rapidly as possible otherwise.

All the following approaches address some part of the same problem space as L4S. In each case, it is shown that L4S complements them or improves on them, rather than being a mutually exclusive alternative: Diffserv addresses the problem of bandwidth apportionment for important traffic as well as queuing latency for delay-sensitive traffic. L4S solely addresses the problem of queuing latency (as well as loss and throughput scaling). Diffserv will still be necessary where important traffic requires priority (e.g. for commercial reasons, or for protection of critical infrastructure traffic). Nonetheless, if there are Diffserv classes for important traffic, the L4S approach can provide low latency for all traffic within each Diffserv class (including the case where there is only one Diffserv class).Also, as already explained, Diffserv only works for a small subset of the traffic on a link. It is not applicable when all the applications in use at one time at a single site (home, small business or mobile device) require low latency. Also, because L4S is for all traffic, it needs none of the management baggage (traffic policing, traffic contracts) associated with favouring some packets over others. This baggage has held Diffserv back from widespread end-to-end deployment. AQMs such as PIE and fq_CoDel give a significant reduction in queuing delay relative to no AQM at all. The L4S work is intended to complement these AQMs, and we definitely do not want to distract from the need to deploy them as widely as possible. Nonetheless, without addressing the large saw-toothing rate variations of Classic congestion controls, AQMs alone cannot reduce queuing delay too far without significantly reducing link utilization. The L4S approach resolves this tension by ensuring hosts can minimize the size of their sawteeth without appearing so aggressive to legacy flows that they starve them. Similarly per-flow queuing is not incompatible with the L4S approach. However, one queue for every flow can be thought of as overkill compared to the minimum of two queues for all traffic needed for the L4S approach. The overkill of per-flow queuing has side-effects: fq makes high performance networking equipment costly (processing and memory) - in contrast dual queue code can be very simple; fq requires packet inspection into the end-to-end transport layer, which doesn't sit well alongside encryption for privacy - in contrast the use of ECN as the classifier for L4S requires no deeper inspection than the IP layer; fq isolates the queuing of each flow from the others but not from itself so, unlike L4S, it does not support applications that need both capacity-seeking behaviour and very low latency.It might seem that self-inflicted queuing delay should not count, because if the delay wasn't in the network it would just shift to the sender. However, modern adaptive applications, e.g. HTTP/2 or the interactive media applications described in , can keep low latency objects at the front of their local send queue by shuffling priorities of other objects dependent on the progress of other transfers. They cannot shuffle packets once they have released them into the network. fq prevents any one flow from consuming more than 1/N of the capacity at any instant, where N is the number of flows. This is fine if all flows are elastic, but it does not sit well with a variable bit rate real-time multimedia flow, which requires wriggle room to sometimes take more and other times less than a 1/N share.It might seem that an fq scheduler offers the benefit that it prevents individual flows from hogging all the bandwidth. However, L4S has been deliberately designed so that policing of individual flows can be added as a policy choice, rather than requiring one specific policy choice as the mechanism itself. A scheduler (like fq) has to decide packet-by-packet which flow to schedule without knowing application intent. Whereas a separate policing function can be configured less strictly, so that senders can still control the instantaneous rate of each flow dependent on the needs of each application (e.g. variable rate video), giving more wriggle-room before a flow is deemed non-compliant. Also policing of queuing and of flow-rates can be applied independently. Yet again, L4S is not an alternative to ABE but a complement that introduces much lower queuing delay. ABE alters the host behaviour in response to ECN marking to utilize a link better and give ECN flows a faster throughput, but it assumes the network still treats ECN and drop the same. Therefore ABE exploits any lower queuing delay that AQMs can provide. But as explained above, AQMs still cannot reduce queuing delay too far without losing link utilization (to allow for other, non-ABE, flows).

A transport layer that solves the current latency issues will provide new service, product and application opportunities. With the L4S approach, the following existing applications will immediately experience significantly better quality of experience under load in the best effort class: Gaming; VoIP; Video conferencing; Web browsing; (Adaptive) video streaming; Instant messaging. The significantly lower queuing latency also enables some interactive application functions to be offloaded to the cloud that would hardly even be usable today: Cloud based interactive video; Cloud based virtual and augmented reality. The above two applications have been successfully demonstrated with L4S, both running together over a 40 Mb/s broadband access link loaded up with the numerous other latency sensitive applications in the previous list as well as numerous downloads - all sharing the same bottleneck queue simultaneously . For the former, a panoramic video of a football stadium could be swiped and pinched so that, on the fly, a proxy in the cloud could generate a sub-window of the match video under the finger-gesture control of each user. For the latter, a virtual reality headset displayed a viewport taken from a 360 degree camera in a racing car. The user's head movements controlled the viewport extracted by a cloud-based proxy. In both cases, with 7 ms end-to-end base delay, the additional queuing delay of roughly 1 ms was so low that it seemed the video was generated locally. Using a swiping finger gesture or head movement to pan a video are extremely latency-demanding actions—far more demanding than VoIP. Because human vision can detect extremely low delays of the order of single milliseconds when delay is translated into a visual lag between a video and a reference point (the finger or the orientation of the head sensed by the balance system in the inner ear (the vestibular system). Without the low queuing delay of L4S, cloud-based applications like these would not be credible without significantly more access bandwidth (to deliver all possible video that might be viewed) and more local processing, which would increase the weight and power consumption of head-mounted displays. When all interactive processing can be done in the cloud, only the data to be rendered for the end user needs to be sent. Other low latency high bandwidth applications such as: Interactive remote presence; Video-assisted remote control of machinery or industrial processes. are not credible at all without very low queuing delay. No amount of extra access bandwidth or local processing can make up for lost time.

The following use-cases for L4S are being considered by various interested parties: Where the bottleneck is one of various types of access network: DSL, cable, mobile, satellite Radio links (cellular, WiFi, satellite) that are distant from the source are particularly challenging. The radio link capacity can vary rapidly by orders of magnitude, so it is often desirable to hold a buffer to utilise sudden increases of capacity; cellular networks are further complicated by a perceived need to buffer in order to make hand-overs imperceptible; Satellite networks generally have a very large base RTT, so even with minimal queuing, overall delay can never be extremely low; Nonetheless, it is certainly desirable not to hold a buffer purely because of the sawteeth of Classic TCP, when it is more than is needed for all the above reasons. Private networks of heterogeneous data centres, where there is no single administrator that can arrange for all the simultaneous changes to senders, receivers and network needed to deploy DCTCP: a set of private data centres interconnected over a wide area with separate administrations, but within the same company a set of data centres operated by separate companies interconnected by a community of interest network (e.g. for the finance sector) multi-tenant (cloud) data centres where tenants choose their operating system stack (Infrastructure as a Service - IaaS) Different types of transport (or application) congestion control: elastic (TCP/SCTP); real-time (RTP, RMCAT); query (DNS/LDAP). Where low delay quality of service is required, but without inspecting or intervening above the IP layer : mobile and other networks have tended to inspect higher layers in order to guess application QoS requirements. However, with growing demand for support of privacy and encryption, L4S offers an alternative. There is no need to select which traffic to favour for queuing, when L4S gives favourable queuing to all traffic. If queuing delay is minimized, applications with a fixed delay budget can communicate over longer distances, or via a longer chain of service functions or onion routers.

The DualQ is, in itself, an incremental deployment framework for L4S AQMs so that L4S traffic can coexist with existing Classic "TCP-friendly" traffic. explains why only deploying a DualQ AQM in one node at each end of the access link will realize nearly all the benefit of L4S. L4S involves both end systems and the network, so suggests some typical sequences to deploy each part, and why there will be an immediate and significant benefit after deploying just one part. If an ECN-enabled DualQ AQM has not been deployed at a bottleneck, an L4S flow is required to include a fall-back strategy to Classic behaviour. describes how an L4S flow detects this, and how to minimize the effect of false negative detection.

DualQ AQMs will not have to be deployed throughout the Internet before L4S will work for anyone. Operators of public Internet access networks typically design their networks so that the bottleneck will nearly always occur at one known (logical) link. This confines the cost of queue management technology to one place. The case of mesh networks is different and will be discussed later. But the known bottleneck case is generally true for Internet access to all sorts of different 'sites', where the word 'site' includes home networks, small-to-medium sized campus or enterprise networks and even cellular devices (). Also, this known-bottleneck case tends to be true whatever the access link technology; whether xDSL, cable, cellular, line-of-sight wireless or satellite. Therefore, the full benefit of the L4S service should be available in the downstream direction when the DualQ AQM is deployed at the ingress to this bottleneck link (or links for multihomed sites). And similarly, the full upstream service will be available once the DualQ is deployed at the upstream ingress.

Deployment in mesh topologies depends on how over-booked the core is. If the core is non-blocking, or at least generously provisioned so that the edges are nearly always the bottlenecks, it would only be necessary to deploy the DualQ AQM at the edge bottlenecks. For example, some datacentre networks are designed with the bottleneck in the hypervisor or host NICs, while others bottleneck at the top-of-rack switch (both the output ports facing hosts and those facing the core). The DualQ would eventually also need to be deployed at any other persistent bottlenecks such as network interconnections, e.g. some public Internet exchange points and the ingress and egress to WAN links interconnecting datacentres.

For any one L4S flow to work, it requires 3 parts to have been deployed. This was the same deployment problem that ECN faced so we have learned from this. Firstly, L4S deployment exploits the fact that DCTCP already exists on many Internet hosts (Windows, FreeBSD and Linux); both servers and clients. Therefore, just deploying DualQ AQM at a network bottleneck immediately gives a working deployment of all the L4S parts. DCTCP needs some safety concerns to be fixed for general use over the public Internet (see Section 2.3 of ), but DCTCP is not on by default, so these issues can be managed within controlled deployments or controlled trials. Secondly, the performance improvement with L4S is so significant that it enables new interactive services and products that were not previously possible. It is much easier for companies to initiate new work on deployment if there is budget for a new product trial. If, in contrast, there were only an incremental performance improvement (as with Classic ECN), spending on deployment tends to be much harder to justify. Thirdly, the L4S identifier is defined so that intially network operators can enable L4S exclusively for certain customers or certain applications. But this is carefully defined so that it does not compromise future evolution towards L4S as an Internet-wide service. This is because the L4S identifier is defined not only as the end-to-end ECN field, but it can also optionally be combined with any other packet header or some status of a customer or their access link . Operators could do this anyway, even if it were not blessed by the IETF. However, it is best for the IETF to specify that they must use their own local identifier in combination with the IETF's identifier. Then, if an operator enables the optional local-use approach, they only have to remove this extra rule to make the service work Internet-wide - it will already traverse middleboxes, peerings, etc.

illustrates some example sequences in which the parts of L4S might be deployed. It consists of the following stages: Here, the immediate benefit of a single AQM deployment can be seen, but limited to a controlled trial or controlled deployment. In this example downstream deployment is first, but in other scenarios the upstream might be deployed first. If no AQM at all was previously deployed for the downstream access, the DualQ AQM greatly improves the Classic service (as well as adding the L4S service). If an AQM was already deployed, the Classic service will be unchanged (and L4S will still be added). In this stage, the name 'TCP Prague' is used to represent a variant of DCTCP that is safe to use in a production environment. If the application is primarily unidirectional, 'TCP Prague' at one end will provide all the benefit needed. Accurate ECN feedback (AccECN) is needed at the other end, but it is a generic ECN feedback facility that is already planned to be deployed for other purposes, e.g. DCTCP, BBR . The two ends can be deployed in either order, because TCP Prague only enables itself if it has negotiated the use of AccECN feedback with the other end during the connection handshake. Thus, deployment of TCP Prague on a server enables L4S trials to move to a production service in one direction, wherever AccECN is deployed at the other end. This stage might be further motivated by performance improvements between DCTCP and TCP Prague (see Appendix A.2 of ). This is a two-move stage to enable L4S upstream. The DualQ or TCP Prague can be deployed in either order as already explained. To motivate the first of two independent moves, the deferred benefit of enabling new services after the second move has to be worth it to cover the first mover's investment risk. As explained already, the potential for new interactive services provides this motivation. The DualQ AQM also greatly improves the upstream Classic service, assuming no other AQM has already been deployed. Note that other deployment sequences might occur. For instance: the upstream might be deployed first; a non-TCP protocol might be used end-to-end, e.g. QUIC, RMCAT; a body such as the 3GPP might require L4S to be implemented in 5G user equipment, or other random acts of kindness.

If L4S is enabled between two hosts but there is no L4S AQM at the bottleneck, any drop from the bottleneck will trigger the L4S sender to fall back to a classic ('TCP-Friendly') behaviour (see Appendix A.1.3 of ). Unfortunately, as well as protecting legacy traffic, this rule degrades the L4S service whenever there is a loss, even if the loss was not from a non-DualQ bottleneck (false negative). And unfortunately, prevalent drop can be due to other causes, e.g.: congestion loss at other transient bottlenecks, e.g. due to bursts in shallower queues; transmission errors, e.g. due to electrical interference; rate policing. Three complementary approaches are in progress to address this issue, but they are all currently research: In TCP Prague, ignore certain losses deemed unlikely to be due to congestion (using some ideas from BBR but with no need to ignore nearly all losses). This could mask any of the above types of loss (requires consensus on how to safely interoperate with drop-based congestion controls). A combination of RACK, reconfigured link retransmission and L4S could address transmission errors (no reference yet); Hybrid ECN/drop policers (see ). L4S deployment scenarios that minimize these issues (e.g. over wireline networks) can proceed in parallel to this research, in the expectation that research success will continually widen L4S applicability. Classic ECN support is starting to materialize (in the upstream of some home routers as of early 2017), so an L4S sender will have to fall back to a classic ('TCP-Friendly') behaviour if it detects that ECN marking is accompanied by greater queuing delay or greater delay variation than would be expected with L4S (see Appendix A.1.4 of ).

An L4S AQM uses the ECN field to signal congestion. So, in common with Classic ECN, if the AQM is within a tunnel or at a lower layer, correct functioning of ECN signalling requires correct propagation of the ECN field up the layers .

This specification contains no IANA considerations.

Because the L4S service can serve all traffic that is using the capacity of a link, it should not be necessary to police access to the L4S service. In contrast, Diffserv only works if some packets get less favourable treatment than others. So Diffserv has to use traffic policers to limit how much traffic can be favoured, In turn, traffic policers require traffic contracts between users and networks as well as pairwise between networks. Because L4S will lack all this management complexity, it is more likely to work end-to-end. During early deployment (and perhaps always), some networks will not offer the L4S service. These networks do not need to police or re-mark L4S traffic - they just forward it unchanged as best efforts traffic, as they already forward traffic with ECT(1) today. At a bottleneck, such networks will introduce some queuing and dropping. When a scalable congestion control detects a drop it will have to respond as if it is a Classic congestion control (as required in Section 2.3 of ). This will ensure safe interworking with other traffic at the 'legacy' bottleneck, but it will degrade the L4S service to no better (but never worse) than classic best efforts, whenever a legacy (non-L4S) bottleneck is encountered on a path. Certain network operators might choose to restrict access to the L4S class, perhaps only to customers who have paid a premium. Their packet classifier (item 2 in ) could identify such customers against some other field (e.g. source address range) as well as ECN. If only the ECN L4S identifier matched, but not the source address (say), the classifier could direct these packets (from non-paying customers) into the Classic queue. Allowing operators to use an additional local classifier is intended to remove any incentive to bleach the L4S identifier. Then at least the L4S ECN identifier will be more likely to survive end-to-end even though the service may not be supported at every hop. Such arrangements would only require simple registered/not-registered packet classification, rather than the managed application-specific traffic policing against customer-specific traffic contracts that Diffserv requires.

The L4S service does rely on self-constraint - not in terms of limiting capacity usage, but in terms of limiting burstiness. It is hoped that standardisation of dynamic behaviour (cf. TCP slow-start) and self-interest will be sufficient to prevent transports from sending excessive bursts of L4S traffic, given the application's own latency will suffer most from such behaviour. Whether burst policing becomes necessary remains to be seen. Without it, there will be potential for attacks on the low latency of the L4S service. However it may only be necessary to apply such policing reactively, e.g. punitively targeted at any deployments of new bursty malware.

As mentioned in , L4S should remove the need for low latency Diffserv classes. However, those Diffserv classes that give certain applications or users priority over capacity, would still be applicable. Then, within such Diffserv classes, L4S would often be applicable to give traffic low latency and low loss. WIthin such a class, the bandwidth available to a user or application is often limited by a rate policer. Similarly, in the default Diffserv class, rate policers are used to partition shared capacity. A classic rate policer drops any packets exceeding a set rate, usually also giving a burst allowance (variants exist where the policer re-marks non-compliant traffic to a discard-eligible Diffserv codepoint, so they may be dropped elsewhere during contention). In networks that deploy L4S and use rate policers, it will be preferable to deploy a policer designed to be more friendly to the L4S service, This is currently a research area. It might be achieved by setting a threshold where ECN marking is introduced, such that it is just under the policed rate or just under the burst allowance where drop is introduced. This could be applied to various types of policer, e.g. , or the 'local' (non-ConEx) variant of the ConEx congestion policer . Otherwise, whenever L4S traffic encounters a rate policer, it will experience drops and the source will fall back to a Classic congestion control, thus losing the benefits of L4S. Further discussion of the applicability of L4S to the various Diffserv classes, and the design of suitable L4S rate policers will require a separate dedicated document.

Receiving hosts can fool a sender into downloading faster by suppressing feedback of ECN marks (or of losses if retransmissions are not necessary or available otherwise). Various ways to protect TCP feedback integrity have been developed. For instance: The sender can test the integrity of the receiver's feedback by occasionally setting the IP-ECN field to the congestion experienced (CE) codepoint, which is normally only set by a congested link. Then the sender can test whether the receiver's feedback faithfully reports what it expects . A network can enforce a congestion response to its ECN markings (or packet losses) by auditing congestion exposure (ConEx) . The TCP authentication option (TCP-AO ) can be used to detect tampering with TCP congestion feedback. The ECN Nonce was proposed to detect tampering with congestion feedback, but it is being reclassified as historic. Appendix C.1 of gives more details of these techniques including their applicability and pros and cons.

Thanks to Wes Eddy, Karen Nielsen and David Black for their useful review comments.

Data Centre TCP (DCTCP) was designed to provide predictably low queuing latency, near-zero loss, and throughput scalability using explicit congestion notification (ECN) and an extremely simple marking behaviour on switches. However, DCTCP does not co-exist with existing TCP traffic---throughput starves. So, until now, DCTCP could only be deployed where a clean-slate environment could be arranged, such as in private data centres. This specification defines `DualQ Coupled Active Queue Management (AQM)' to allow scalable congestion controls like DCTCP to safely co-exist with classic Internet traffic. The Coupled AQM ensures that a flow runs at about the same rate whether it uses DCTCP or TCP Reno/Cubic, but without inspecting transport layer flow identifiers. When tested in a residential broadband setting, DCTCP achieved sub-millisecond average queuing delay and zero congestion loss under a wide range of mixes of DCTCP and `Classic' broadband Internet traffic, without compromising the performance of the Classic traffic. The solution also reduces network complexity and eliminates network configuration. This specification defines the identifier to be used on IP packets for a new network service called low latency, low loss and scalable throughput (L4S). It is similar to the original (or 'Classic') Explicit Congestion Notification (ECN). 'Classic' ECN marking was required to be equivalent to a drop, both when applied in the network and when responded to by a transport. Unlike 'Classic' ECN marking, for packets carrying the L4S identifier, the network applies marking more immediately and more aggressively than drop, and the transport response to each mark is reduced and smoothed relative to that for drop. The two changes counterbalance each other so that the throughput of an L4S flow will be roughly the same as a 'Classic' flow under the same conditions. However, the much more frequent control signals and the finer responses to them result in ultra-low queuing delay without compromising link utilization, even during high load. Examples of new active queue management (AQM) marking algorithms and examples of new transports (whether TCP-like or real- time) are specified separately. The new L4S identifier is the key piece that enables them to interwork and distinguishes them from 'Classic' traffic. It gives an incremental migration path so that existing 'Classic' TCP traffic will be no worse off, but it can be prevented from degrading the ultra-low delay and loss of the new scalable transports.

PSC Bell Labs Simula Research Lab Bell Labs BT (Under submission) Bell Labs Simula Research Lab Bell Labs BT BT Bell Labs Nokia Research Centre Google Google Google Google Google Simula Research Lab Bell Labs Bell Labs BT

The following table includes all the items that will need to be standardized to provide a full L4S architecture. The table is too wide for the ASCII draft format, so it has been split into two, with a common column of row index numbers on the left. The columns in the second part of the table have the following meanings: The IETF WG most relevant to this requirement. The "tcpm/iccrg" combination refers to the procedure typically used for congestion control changes, where tcpm owns the approval decision, but uses the iccrg for expert review ; Applicable to all forms of TCP congestion control; Applicable to Data Centre TCP as currently used (in controlled environments); Applicable to an future Data Centre TCP congestion control intended for controlled environments; Applicable to a Scalable variant of XXX (TCP/SCTP/RMCAT) congestion control. Req # Requirement Reference 0 ARCHITECTURE 1 L4S IDENTIFIER 2 DUAL QUEUE AQM 3 Suitable ECN Feedback , . SCALABLE TRANSPORT - SAFETY ADDITIONS 4-1 Fall back to Reno/Cubic on loss S.2.3, 4-2 Fall back to Reno/Cubic if classic ECN bottleneck detected S.2.3 4-3 Reduce RTT-dependence S.2.3 4-4 Scaling TCP's Congestion Window for Small Round Trip Times S.2.3, SCALABLE TRANSPORT - PERFORMANCE ENHANCEMENTS 5-1 Setting ECT in TCP Control Packets and Retransmissions 5-2 Faster-than-additive increase (Appx A.2.2) 5-3 Faster Convergence at Flow Start (Appx A.2.2) # WG TCP DCTCP DCTCP-bis TCP Prague SCTP Prague RMCAT Prague 0 tsvwg Y Y Y Y Y Y 1 tsvwg Y Y Y Y 2 tsvwg n/a n/a n/a n/a n/a n/a 3 tcpm Y Y Y Y n/a n/a 4-1 tcpm Y Y Y Y Y 4-2 tcpm/ iccrg? Y Y ? 4-3 tcpm/ iccrg? Y Y Y ? 4-4 tcpm Y Y Y Y Y ? 5-1 tcpm Y Y Y Y n/a n/a 5-2 tcpm/ iccrg? Y Y Y ? 5-3 tcpm/ iccrg? Y Y Y ?