]> Mozilla

mt@lowentropy.net

Private Octopus Inc.

huitema@huitema.net

Fastly

kazuhooku@gmail.com

Meta

matt.joras@gmail.com

Ericsson

marcus.ihlar@ericsson.com

Web and Internet Transport SCONE locomotive pastry This document describes a protocol where on-path network elements can communicate their perspective on the maximum sustainable throughput for QUIC flows to endpoints. This throughput advice suggests an upper bound on long-term average throughput, independent of and complementary to real-time congestion control signals. About This Document The latest revision of this draft can be found at . Status information for this document may be found at . Discussion of this document takes place on the SCONE Working Group mailing list (), which is archived at . Subscribe at . Source for this draft and an issue tracker can be found at .

Introduction Many networks have known rate limits or apply these limits by policy to constrain data rates. This is often done without any ability to indicate rate limits to applications. The result can be that application performance is degraded, because throughput limits can manifest in ways that are incompatible with the rate estimation or congestion control algorithms used at endpoints. Having the network indicate what throughput limits apply, in a way that is accessible to endpoints, allows applications to use this information when adapting their send rate. The Standard Communication with Network Elements (SCONE) protocol is negotiated by QUIC endpoints. SCONE provides a means for a network to signal its present best estimate for maximum sustainable throughput, or throughput advice, associated with the flows of UDP datagrams that QUIC exchanges. Any network function that is able to update the content of UDP datagrams qualifies as a network element that can use SCONE packets to provide throughput advice to QUIC endpoints. Networks with rate limits can use SCONE to send throughput advice to cooperating endpoints to limit overall network usage. Where congestion control signals -- such as Explicit Congestion Notification (ECN) , delays and loss -- operate on a timescale of a round trip time, throughput advice operates over a much longer period. This has benefits in some networks as endpoints can adapt network usage to better suit network conditions. For example, radio networks and battery-powered devices perform better with short, bursty exchanges, rather than constant transmission at a fixed rate. For endpoints, SCONE throughput advice makes network policies visible, which can reduce wasteful probing beyond those limits.

Overview QUIC endpoints can negotiate the use of SCONE by including a transport parameter () in the QUIC handshake. Endpoints then occasionally send SCONE packets, which are always coalesced with ordinary QUIC packets that they send. Networks that have rate limiting policies, or known throughput constraints, can detect flows that include SCONE packets. The network, via an on-path network element, can indicate a maximum sustainable throughput by modifying the SCONE packet as it transits the network element. The propagation of SCONE packets, including the throughput advice that is updated by a network element, is shown in .

Propagation of SCONE signal | SCONE(advice) | | +QUIC +---- +QUIC ---->| | | | Validate QUIC packet | | | and record advice | | | ]]>

QUIC endpoints that receive SCONE packets observe the indicated version and extract the rate signal. They record any throughput advice only if the coalesced QUIC packet is successfully processed; see . Throughput advice only applies to the direction and path for which it is received. A connection that migrates or uses multipath cannot assume that throughput advice from one path applies to new paths. Advice for the client-to-server direction and the server-to-client direction of each path are independent, and are expected to be different, for reasons including asymmetric link capacity and path diversity. Applications can use SCONE in either or both directions of each path at the discretion of endpoints.

Applicability This protocol can provide throughput advice only for QUIC flows where endpoints send SCONE packets (). The operation of the SCONE protocol depends on network elements that are able to modify packets as they are forwarded. This provides endpoints strong evidence that the network element has the power to apply a rate limiting policy; though see for potential limitations on this. The throughput advice that this protocol carries is independent of congestion signals, limited to a single path and UDP packet flow, unidirectional, and strictly advisory. A companion document addresses applicability, operational, and deployment considerations in more detail.

Independent of Congestion Signals SCONE throughput advice is not a substitute for congestion feedback or congestion control. They are complementary. Congestion signals, such as acknowledgments or ECN markings , provide real-time information on loss and delay for a network path. A congestion controller needs to detect changed conditions and change sending behavior more quickly than SCONE allows. In comparison, SCONE throughput advice operates over a much longer period. SCONE provides an explicit signal regarding potential throughput constraints on a path. This allows applications to set throughput targets quickly, without waiting to estimate achievable throughput from observed performance. Congestion signals can indicate a throughput limit that is different from the signaled throughput advice. Endpoints cannot assume that the rate indicated in throughput advice is achievable if congestion signals indicate otherwise. Congestion could be experienced at a different point on the network path than the network element that signals throughput advice. Therefore, endpoints need to respect the send rate constraints that are set by a congestion controller. Networks can use SCONE to communicate throughput advice for reasons other than rate limiting policies. For example, a network element in an access network could provide reduced throughput advice to guide application use of network capacity during periods of unusually high usage. Throughput advice can indicate temporary increases in available capacity or temporarily reduced capacity. This includes persistent overuse, equipment faults, or other transient issues. Providing advice is applicable if increases or reductions are expected to last for more than one monitoring period; see .

Unspecified Scope Just because a network element can set throughput advice, that does not prove that the flow can achieve that rate. Nor is it a commitment to providing that throughput. It is a hint that exceeding that rate is unlikely to be successful, from the perspective of a specific network element. Throughput advice is signaled for a specific flow, but the underlying throughput constraint could pertain to a collection of flows, rather than a single flow. The scope of the underlying throughput constraint is not carried in the signal. For instance, policy limits might apply at a network subscription level, such that multiple flows receive the same signal and combined usage contributes to a shared capacity limit. Endpoints can therefore be more confident in the throughput signal as an indication of the maximum achievable throughput than as any indication of expected throughput. In addition to endpoints respecting congestion signals (see ), networks might need to monitor and enforce policies, even where applications attempt to follow advice (see ), since the combined usage of multiple flows might exceed the shared capacity limit. The advised throughput will likely only be achievable when the application is the only entity consuming bandwidth in the scope of the underlying constraint. In the presence of multiple flows, achievable throughput could be lower than indicated by the throughput advice, with the throughput limited by the congestion controller or application rate control algorithm. This implies that signals can most usefully be applied to a downlink flow in access networks, close to an endpoint. In that case, capacity is less likely to be split between multiple active flows.

Per-Flow Signal The same address tuple (IP version, source and destination IP addresses and UDP ports) might be used for multiple QUIC connections. A single signal might be lost or only reach a single application endpoint. Network elements can apply SCONE throughput advice to all QUIC connections that include SCONE packets to ensure that advice is received by all application endpoints. The signaled advice applies to the flow of packets on the same address tuple for the duration of the current monitoring period, unless it is updated earlier or the flow ends; see for details on the monitoring period. Rate limiting policies often apply on the level of a device or subscription, but endpoints cannot assume that this is the case. A separate signal can be sent for each flow. When network elements provide throughput advice to a QUIC flow that encapsulates tunneled flows (such as ) they can only provide the advice to the outermost flow. Endpoints can apply the throughput advice to packets in flows that are subsequently encapsulated, but following that advice can have privacy implications; see .

Unidirectional Signal Throughput advice is signaled with SCONE packets that are transmitted as part of the flow that the advice applies to. Carrying signals in the affected flow, in the same way that ECN signals are conveyed, ensures that there is no ambiguity about what flow is affected. However, this means that the endpoint that receives throughput advice is not the endpoint that needs to adapt its sending behavior. A receiving endpoint might need to communicate the value it receives to the sending peer in order to ensure that the limit is respected. This document does not define how that communication occurs as this is specific to the application in use.

Advisory Signal Throughput advice indicates what one part of the network expects to be achievable for flows that transit that portion of the network. It is possible that very different throughput is achievable -- either higher or lower than the advice -- as determined by congestion control. Endpoints that receive this signal therefore need to treat the information as advisory. The fact that an endpoint requests throughput advice does not necessarily mean that it will adhere to advice; in some cases, the endpoint cannot. For example, a flow could initially be used to serve video chunks, with the client selecting chunks of different bitrates based on received advice, but later switch to a bulk download that cannot be similarly controlled. Composite flows from multiple applications, such as tunneled flows, might only have a subset of the involved applications that are capable of handling SCONE signals. Therefore, when a network element detects that throughput exceeds the advertised throughput advice, it might apply rate limiting. Network conditions and rate-limit policies can change in ways that make previously signaled advice obsolete. For example, routing changes can cause a flow to move to a different network path. There are no guarantees that updated advice will be sent at such events.

Application Use of Advice Applications that choose to follow throughput advice do so in the way that best suits their needs. The most obvious way to follow throughput advice is to inform the sending peer of the advice so that the peer can adjust sending rates as necessary. This document does not provide specific guidance on how applications might adapt their use of network capacity in response to advice. Some applications offer options for rate control that can offer improved performance when following advice. For instance, real-time and streaming video applications can often dynamically adapt their network usage. Typical HTTP Live Streaming or Dynamic Adaptive Streaming over HTTP clients are provided with manifests that allow them to adjust the bitrate and quality of media segments based on available network capacity. Low priority bulk transfer applications, such as software updates, might also choose to follow advice. Following throughput advice could reduce the impact of an application on other network users, reserves capacity for high-priority activities, and could avoid potential enforcement action by the network; see .

Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here. In SCONE:

• a "network element" is any device in the network that can produce throughput advice.
• "Throughput advice" refers the information communicated to endpoints by network elements in the form of a rate signal; see .
• A "monitoring period" is the time over which rate signals apply; see .

SCONE Packet A SCONE packet is a QUIC long header packet that follows the QUIC invariants; see . It only includes the header, with no payload. shows the format of the SCONE packet using the conventions from .

SCONE Packet Format

The most significant bit (0x80) of the packet indicates that this is a QUIC long header packet. The next bit (0x40) is reserved and can be set according to . The Rate Signal High Bits field consists of the low six bits (0x3f) of the first byte. Together with the most significant bit of the Version field, this forms the 7-bit Rate Signal. Values for the Rate Signal are described in . The Version field contains either 0x6f7dc0fd or 0xef7dc0fd. The only difference between these two values is the most significant bit, which also contributes to the Rate Signal. All other bits are identical, which facilitates detection and modification of SCONE packets. The Destination Connection ID field MUST be set to the same value as the Destination Connection ID field of the other packets in the same datagram; see . The Source Connection ID field is set to match the Source Connection ID field of any packet that follows. If the next packet in the datagram does not have a Source Connection ID field, which is the case for packets with a short header (), the Source Connection ID field MUST be empty and the Source Connection ID Length field MUST be 0. SCONE packets MUST be coalesced with other QUIC packets (see ) and MUST be included as the first packet in a datagram. This is primarily to simplify the process of updating throughput advice in network elements. This is also necessary in many cases for QUIC versions 1 and 2 because packets with a short header cannot precede any other packets. A sender MUST NOT include more than one SCONE packet in a datagram.

Rate Signals A Rate Signal is a 7-bit unsigned integer (0-127). The high six bits are the Rate Signal High Bits, and the least significant bit is the most significant bit of the Version field. When sent by a QUIC endpoint, the Rate Signal is set to 127. Receiving a value of 127 indicates that throughput advice is unknown, either because network elements on the path are not providing advice or they do not support SCONE. All other values (0 through 126) represent the ceiling of rates advised by the network element(s) on the path. Throughput advice follows a logarithmic scale defined as:

• Base rate (b_min) = 100 kbit/s (100,000 bits per second)
• Bitrate at value n = b_min * 10^(n/20)

where n is an integer between 0 and 126 represented by the Rate Signal. lists some of the values for signals and the corresponding bitrate for each. Examples of SCONE signals and corresponding rates

Bitrate	Rate Signal
100 kbit/s	0
112 kbit/s	1
126 kbit/s	2
141 kbit/s	3
1 Mbit/s	20
1.12 Mbit/s	21
10 Mbit/s	40
11.2 Mbit/s	41
100 Mbit/s	60
112 Mbit/s	61
1 Gbit/s	80
1.12 Gbit/s	81
10 Gbit/s	100
11.2 Gbit/s	101
100 Gbit/s	120
112 Gbit/s	121
199.5 Gbit/s	126
Unknown	127

Monitoring Period The time over which throughput advice applies is defined to be a period of 67 seconds. Protocol participants can use a different period, depending on their role. Senders can limit their send rate over any time period up to 67 seconds. Network elements can monitor and apply limits to send rates using time period of at least 67 seconds. The choice of 67 seconds is a compromise between competing interests. Longer periods allow applications more flexibility in terms of how to allocate bandwidth over time. Shorter periods allow networks to administer policies more tightly, or react more quickly to changes in their capacity. A shorter period also allows applications to increase send rates sooner when rates increase. The choice of 67 seconds, as a prime number, also helps avoid synchronization with other periodic effects that are commonly measured in whole seconds. This includes segment length or key frame intervals in video applications, but also includes timers for Network Address Translation (NAT) devices; see . Any repeating phenomenon at a 67-second interval is therefore unlikely to be due to other periodic effects.

Endpoint Processing of SCONE Packets Processing a SCONE packet involves reading the value from the Rate Signal field. However, throughput advice MUST be ignored unless another packet from the same datagram is successfully processed. A SCONE packet therefore always needs to be coalesced with other QUIC packets; see for why this is necessary. What successful processing means depends on the QUIC version associated with the coalesced packets, but this minimally includes successfully authenticating the packet as valid. SCONE packets MUST be ignored if the QUIC packets that follow are discarded or ignored by the endpoint, which especially includes packets that are discarded as potential duplicates. A SCONE packet is defined by the use of the long header bit (0x80 in the first byte) and the SCONE protocol version (0x6f7dc0fd or 0xef7dc0fd in the next four bytes). The 7-bit Rate Signal can be extracted by combining the low 6 bits of the first byte with the most significant bit of the version field. A SCONE packet MUST be discarded if the Destination Connection ID is not consistent with those coalesced packets, as specified in . Similarly, if the Source Connection ID is inconsistent, the SCONE packet MAY be discarded. When discarding a SCONE packet due to inconsistent Connection IDs, endpoints MAY also discard the QUIC packets that were coalesced into the same datagram. A receiver MAY discard a datagram that contains more than one SCONE packet. A receiver MUST discard a SCONE packet if the rate signal is unknown (127). If a connection uses multiple Differentiated Services Code Point (DSCP) markings , the throughput advice that is received on datagrams with one marking might not apply to datagrams that have different markings.

Following Throughput Advice Endpoints that receive throughput advice can advise their peer of the limit so that the peer might limit the amount of data it sends over any monitoring period (). Alternatively, the endpoint might change its own behavior to effect a similar outcome indirectly, which might use flow control or changes to request patterns. An endpoint that receives throughput advice might receive multiple different values. Where advice is reported to applications, endpoints MUST report the lowest throughput advice received during the previous monitoring period; see . After a monitoring period () without receiving any throughput advice, the previous advice expires. Endpoints can remove any constraints that resulted from the expired throughput advice. This does not mean that there are no limits, either in policy or due to network conditions, only that these limits are now unknown. Other constraints on usage will still apply, which necessarily includes congestion control and might include other, application-specific constraints. Allowing advice to expire ensures that changes in routing do not cause stale advice to persist indefinitely when network elements on a new path do not provide advice. This approach ensures that network elements are able to reduce the frequency with which they send updated signals to as low as once per monitoring period. However, applying signals at a low frequency risks endpoints discarding throughput advice if no SCONE packet is available for providing updated advice (), or packets carrying advice are lost. Sending the signal multiple times per monitoring period increases the likelihood that the signal is received.

Negotiating SCONE A QUIC endpoint indicates that it is able to receive SCONE packets by including the scone_supported transport parameter (0x219e). Each endpoint independently indicates willingness to receive SCONE packets. An endpoint that does not include the scone_supported transport parameter can send SCONE packets if their peer includes the transport parameter. The scone_supported transport parameter MUST be empty. Receiving a non-zero-length scone_supported transport parameter MUST be treated as a connection error of type TRANSPORT_PARAMETER_ERROR; see . This transport parameter is valid for QUIC versions 1 and 2 and any other version that recognizes the versions, transport parameters, and frame types registries established in Sections , , and of . Endpoints MUST NOT remember whether the scone_supported transport parameter was present on the previous connection when using 0-RTT; see . That is, SCONE packets cannot be sent on a connection until the transport parameter is received.

Indicating Support on New Flows All new flows that are initiated by a client that supports SCONE MUST include bytes with values 0xc8 and 0x13 as the last two bytes of the payload of the UDP datagrams that commence a new flow, if the QUIC version in use permits the inclusion of data after packets. For example, in QUIC version 1, these datagrams contain QUIC packets with a long header, including a length prefix (). The UDP datagrams sent by a client can contain: one or more QUIC version 1 Initial packets, zero or more 0-RTT packets, padding or other data that is discarded on receipt, and the indication bytes (0xc8, 0x13) as the final bytes of the UDP payload. The SCONE indication bytes MUST be sent in every datagram until the client receives any datagram from the server, at which point the client can reasonably expect that the indication was received. A client that uses a QUIC version that sends length-delimited packets during the handshake, which includes QUIC versions 1 and 2 , can include an indicator of SCONE support outside of the QUIC packets at the end of datagrams that start a flow. The handshakes of these protocols ensures that the indication can be included in every datagram the client sends until it receives a response -- of any kind -- from the server.

Limitations of Indication This indication does not mean that SCONE signals will be respected, only that the client is able to negotiate SCONE. A server might not support SCONE and either endpoint might choose not to send SCONE packets. Finally, applications might be unable to apply throughput advice or choose to ignore it. There is a non-negligible risk of collision with other protocols or even QUIC usage without SCONE indications. The indicator is just two bytes, which could be sent by chance on non-SCONE flows. This means that the indication alone is not sufficient to indicate that a flow is QUIC with the potential for SCONE support. Despite these limitations, having an indication might allow network elements to change their starting posture with respect to their enforcement of their rate limit policies.

Indications for Migrated Flows In QUIC version 1 and 2, the two byte indicator () cannot be used on migration to a new path. Sending a SCONE packet for the first few packets on a new path gives network elements on that path the ability to recognize the flow as being able to receive throughput advice. The SCONE packet also gives the network element an opportunity to provide throughput advice for the new flow. To enable this indication, even if an endpoint would not otherwise send SCONE packets, endpoints can send a SCONE packet any time they send a QUIC PATH_CHALLENGE or PATH_RESPONSE frame. This applies to both client and server endpoints, but only if the peer has sent the transport parameter; see .

Avoiding Ossification When Reading the Indicator A network element could classify all 5-tuples where the first observed UDP datagram ends in the indicator bytes as potential SCONE. However, this is not a strong indicator (see ), so network elements are encouraged to apply further criteria to further reduce the set of flows that are identified as potentially supporting SCONE to reduce the likelihood of false positives. However, it SHOULD NOT apply criteria that reduce the ability of new QUIC versions to employ SCONE. SCONE operates independently of any specific QUIC version, so any criteria should consult the QUIC version invariants in .

Network Deployment QUIC endpoints can enable the use of the SCONE protocol by sending SCONE packets (). Network elements can then update the Rate Signal field () based on network policies or information about available capacity.

Applying Throughput Advice Signals A network element detects a SCONE packet by observing that a packet has a QUIC long header and one of the SCONE protocol versions (0x6f7dc0fd or 0xef7dc0fd). A network element then conditionally replaces the most significant bit of the Version field and the Rate Signal High Bits field with values of its choosing. A network element might receive a packet that already includes a rate signal. The network element replaces the rate signal if it wishes to signal a lower value for throughput advice; otherwise, the original values are retained, preserving the signal from the network element with the lower policy. A network element MUST NOT replace a rate signal with a higher or unknown value. The following pseudocode indicates how a network element might detect a SCONE packet and replace the existing rate signal (packet_signal) with a new rate signal (target_signal) that encodes the throughput advice of this network element. > 31) if target_signal < packet_signal: packet[0] = (packet[0] & 0xc0) | (target_signal >> 1) packet[1] = (packet[1] & 0x7f) | ((target_signal & 1) << 7) ]]> Once the throughput advice is updated, the network element updates the UDP checksum for the datagram. An incremental update process is described in and includes additional information relating to IPv6.

When To Avoid Updating Throughput Advice A network element MUST NOT alter datagrams to add SCONE packets or synthesize datagrams that contain SCONE packets. The latter will not be accepted and the former, even if they do not exceed the path MTU as a result, can be detected by applications and could be ignored. This document does not define a mechanism to support detection, but one might be added in future. Network elements MUST only update the content of datagrams on a given address tuple a few times each monitoring period. Network elements MAY update more often immediately after a change in their throughput advice, to reduce the reaction time from senders. If too many datagrams are altered, that could interfere with UDP protocols that are not QUIC; see .

Ensuring Throughput Advice Availability To avoid throughput advice expiring, a network element needs to ensure that it updates throughput advice in SCONE packets with no more than a monitoring period () between each update. Because this depends on the availability of SCONE packets and packet loss can cause signals to be missed, network elements might need to update more often. Ideally, network elements update advice in SCONE packets at least twice per monitoring period, to match endpoint behavior (see ). At the start of a flow, network elements are encouraged to update the rate signal of the first few SCONE packets it observes so that endpoints can obtain throughput advice early. Senders that send a SCONE packet or network elements that update SCONE packets every 20–30 seconds are likely sufficient to ensure that throughput advice is not lost. To reduce the risk of synchronization across multiple senders, which could cause network elements to miss updates, senders can include a small random delay.

Monitoring Flows Providing throughput advice is optional for any network. A network that updates SCONE packets to provide throughput advice might, also optionally, choose to monitor flows to determine whether applications are following advice. This section outlines a method that a network element could use to determine whether a flow exceeds the value from provided throughput advice. Network deployments that choose to monitor are free to follow any monitoring regime that suits their needs. This documented approach to monitoring is largely illustrative; there is no interoperability impact from choosing an alternative approach. However, monitoring any more strictly than the following could mean that an application might be incorrectly classified as not following advice. A looser monitoring approach, such as monitoring over a longer time window than the monitoring period (67s) or using a higher rate than is signaled, has no risk of incorrect classification. When a network changes the throughput advice it intends to provide, applications need time to adjust their sending behavior. As a result, any monitoring needs to allow time for SCONE packets to be updated, for those packets to be received by endpoints, and for applications to adapt. A network element can then monitor affected flows to determine whether the provided throughput advice was followed. The simplest monitoring approach bases monitoring on the maximum value that the network element was configured to apply to SCONE packets during the preceding two monitoring periods. This allows an additional monitoring period to compensate for additional delays as SCONE packets are not delivered reliably. Relative to an additional monitoring period, other delays in propagating throughput advice are expected to be negligible. If the network element cannot update the throughput advice in every SCONE packet (or can only do so periodically), a longer period might be used.

Deployment of Monitoring Functions Any monitoring and policy enforcement could be implemented in different network elements than the ones that signal throughput advice. This enables more flexible allocation of responsibilities between nodes in the same administrative domain. SCONE packets can cross administrative boundaries, which means that network elements might observe throughput advice on SCONE packets that have transited other networks. A network element MUST NOT enforce throughput limits based on throughput advice that is observed in SCONE packets received from other entities. Any enforcement action needs to be the result of configuration or other authorized administrative action, not unauthenticated network signals. Unlike endpoints, network elements do not have the capability to validate other QUIC packets contained in the same datagram; see .

Flows That Exceed Throughput Advice A network could deploy policy enforcement that drops or delays packets to ensure that applications do not exceed throughput limits set in policy. SCONE allows networks to provide advice to applications, so that there is less need to enforce throughput limits on flows. Strict enforcement through dropping or delaying packets can be inefficient and lead to poor application performance. Some applications will not support SCONE. Other applications either will not or cannot follow throughput advice. Networks can monitor flows to determine if applications follow advice; see . A network could choose to either disable or loosen policy enforcement for flows where SCONE is active, but re-enable or tighten enforcement if monitoring indicates that throughput advice is not being respected.

Endpoint Usage The SCONE protocol defines two versions (0x6f7dc0fd and 0xef7dc0fd) that combined carry throughput advice that covers a range of bitrates between 100 kbit/s and 199.5 Gbit/s.

Providing Opportunities to Apply Throughput Advice Signals Endpoints that wish to offer network elements the option to provide throughput advice signals can send SCONE packets at any time. This is a decision that a sender makes when constructing datagrams. As specified in , endpoints include a SCONE packet as the first packet in a datagram, coalesced with additional packets. Upon confirmation that the peer is willing to receive SCONE packets, an endpoint SHOULD include SCONE packets in the first few UDP datagrams that it sends. Doing so increases the likelihood of eliciting early throughput advice from network elements, allowing applications to apply that advice from the early stages of the data transfer. After that, endpoints that seek to receive throughput advice on a flow MUST send a SCONE packet at least twice each monitoring period; see . Sending SCONE packets more often might be necessary to:

Avoid missing advice:

If SCONE packets are not sent, updated, and received for an entire monitoring period, an application might assume that no throughput advice is being provided.

Reduce latency:

The time between SCONE packets determines the maximum delay between changes in throughput advice and when that advice can be received and acted upon.

A sender can track the receipt of the coalesced QUIC packet and send another SCONE packet when loss is detected. However, it is likely simpler to send SCONE packets more often. Sending a SCONE packet every 20–30 seconds is likely sufficient to ensure that throughput advice is not lost, though endpoints might send a packet every few seconds to improve responsiveness. This period could be determined by how quickly an application is able to respond to a change in throughput advice. For example, a streaming application that fetches video segments that are 5 seconds in length might send SCONE packets on a similar cadence. A real-time conferencing application might send more often. In either case, the length of the monitoring period () limits how fast any application can react. Though sending SCONE packets more than once each round trip time might help reduce exposure to packet loss, it is better to spread updates over time rather than to send multiple SCONE packets in less frequent bursts. The main cost associated with sending SCONE packets is the reduction in available space in datagrams for application data. A sender MUST NOT send a SCONE packet unless a QUIC packet needs to be sent for other reasons. It is safe to allow throughput advice to lapse if the connection is otherwise idle, whereas induced activity wastes resources by keeping connections alive unnecessarily. A network element that wishes to signal updated throughput advice waits for the next SCONE packet in the desired direction; see .

Feedback To Sender About Signals Information about throughput advice is intended for the sending application. Any signal from network elements can be propagated to the receiving application using an implementation-defined mechanism. This document does not define a means for indicating what was received. The expectation is that any signal is propagated to the application for handling, rather than being handled automatically by the transport layer. How a receiving application communicates throughput advice to a sending application will depend on the application in use. Different applications can choose different approaches. For example, in an application where a receiver drives rate adaptation, it might not be necessary to define additional signaling. A sender can use any acknowledgment mechanism provided by the QUIC version in use to learn whether datagrams containing SCONE packets were likely received. This might help inform whether to send additional SCONE packets in the event that a datagram is lost. For instance, if a UDP datagram carrying both a SCONE packet and an ack-eliciting QUIC packet is acknowledged, the sender knows the SCONE packet was also received. However, rather than relying solely on transport-layer acknowledgments, an application-layer mechanism might better indicate what has been received and acted upon. SCONE packets could be stripped from datagrams in the network, which cannot be reliably detected. This could result in a sender falsely believing that no network element applied throughput advice. Senders will therefore proceed as though there was no advice.

Security Considerations SCONE throughput advice is not authenticated. Throughput advice might be incorrectly set in order to encourage endpoints to behave in ways that are not in their interests. Endpoints can ignore limits, though that can have consequences; see . The congestion controller employed by a sender provides real-time information about the rate at which the network path is delivering data. Similarly, if there is a strong need to ensure that throughput advice is respected, network elements cannot assume that the signaled advice will be respected by endpoints.

Off-Path Adversaries The modification of packets provides endpoints proof that a network element is in a position to drop datagrams and could apply a rate limit policy. states that endpoints only accept signals if the datagram contains a packet that it accepts to prevent an off-path attacker from inserting spurious throughput advice. Some off-path attackers could be able to both observe traffic and inject packets. Attackers with such capabilities could observe packets sent by an endpoint, create datagrams coalescing an arbitrary SCONE packet and the observed packet, and send these datagrams such that they arrive at the peer endpoint before the original packet. Spoofed packets that seek to advertise a higher limit than might otherwise be permitted also need to bypass any rate limiters. The attacker will thus get arbitrary SCONE packets accepted by the peer, with the result being that the endpoint receives a false or misleading rate limit. The recipient of throughput advice therefore cannot guarantee that the signal was not generated by an on-path network element. The capabilities required of an off-path attacker are substantially similar to those of on-path elements. An off-path attacker can generate throughput advice that will be accepted by an endpoint if it has the ability to damage packets in a way that could be able to affect throughput capacity of the flow. The one exception is an off-path attacker that is only capable of generating spoofed copies of packets with modified throughput rates that reach endpoints ahead of the original.

Fake SCONE Packets Attackers that can inject packets could compose arbitrary "SCONE-like" packets by selecting a pair of IP addresses and ports, an arbitrary rate signal, a valid SCONE version number, an arbitrary "destination connection ID", and an arbitrary "source connection ID". A coalesced "1RTT" packet will start with a plausible first octet, and continue with the selected destination connection ID followed by a sufficiently long series of random bytes, mimicking the content of an encrypted packet. Endpoints will reject such packets because they do not contain valid QUIC packets, but network elements cannot detect this. All the network elements between the injection point and the destination will have to process these packets. Attackers could send a high volume of these "fake" SCONE packets in a denial of service (DOS) attempt against network elements. The attack will force the intermediaries to process the fake packets. If network elements are keeping state for ongoing SCONE flows, this might exhaust memory resources. The mitigation is the same as for other distributed DOS attacks: limit the rate of SCONE packets that a network element is willing to process; possibly, implement logic to distinguish valid SCONE packets from fake packets; or, use generic protection against Distributed DOS attacks. Attackers could also try to craft the fake SCONE packets in ways that trigger a processing error at network elements. For example, they might pick connection identifiers of arbitrary length. Network elements can mitigate these attacks with an implementation that fully conforms to the specification of .

Damage to Other Protocols Network elements that update SCONE packet fields might do that for datagrams exchanged in other protocols. If the first five bytes of the datagram match the QUIC long header byte and SCONE version, the network element might modify the signal, resulting in damage to those protocols. The most serious damage occurs when every datagram matches and is subsequently modified, because that could mean that the protocol is effectively unable to operate end-to-end. To avoid unrecoverable damage to non-QUIC protocols, network elements only update a limited number of datagrams in each monitoring period; see . In addition, some heuristics might be used to detect SCONE-compatible QUIC flows. This includes identification of a QUIC handshake on the flow, the presence of indications (), or other heuristics. If these heuristics indicate a non-QUIC flow, the safest option is for network elements to disable updating of datagrams.

Privacy Considerations The focus of this analysis is the extent to which observing SCONE packets could be used to gain information about endpoints. This might be leaking details of how applications using QUIC operate or leaks of endpoint identity when using additional privacy protection, such as a VPN. Any network element that can observe the content of that packet can read the throughput advice that was applied. Any signal is visible on the path, from the point at which it is applied to the point at which it is consumed at an endpoint. On path elements can also alter the SCONE signal to try trigger specific reactions and gain further knowledge. In the general case of a client connected to a server through the Internet, SCONE does not provide much advantage to attackers. The identities of the clients and servers are already visible through their IP addresses. Traffic analysis tools already provide more information than the throughput advice set by SCONE. There are two avenues of attack that require more analysis:

• that the passive observation of SCONE packets might help identify or distinguish endpoints; and
• that active manipulation of SCONE signals might help reveal the identity of endpoints that are otherwise hidden behind VPNs or proxies.

Passive Attacks If only a few clients and server pairs negotiate the usage of SCONE, the occasional observation of SCONE packets will "stick out". That observation could be combined with observation of timing and volume of traffic to help identify the endpoint or categorize the application that they are using. A variation of this issue occurs if SCONE is widely implemented, but only used in some specific circumstances. In that case, observation of SCONE packets reveals information about the state of the endpoint. If multiple servers are accessed through the same front facing server, Encrypted Client Hello (ECH) can prevent outside parties from identifying which specific server a client is using. However, if only a few of these servers use SCONE, any SCONE packets will help identify which specific server a client is using. This issue will be mitigated if SCONE becomes widely implemented, and if the usage of SCONE is not limited to the type of applications that make active use of the signal. QUIC implementations are therefore encouraged to make the feature available unconditionally. Endpoints might send SCONE packets whenever a peer can accept them.

Active Attacks Suppose a configuration in which multiple clients use a VPN or proxy service to access the same server. The attacker sees the IP addresses in the packets behind VPN and proxy and also between the users and the VPN, but it does not know which VPN address corresponds to what user address. Suppose now that the attacker selects a flow on the link between the VPN/proxy and server. The attacker applies throughput advice to SCONE packets in that flow. The attacker chooses a bandwidth that is lower than the "natural" bandwidth of the connection. A reduction in the rate of flows between client and VPN/proxy might allow the attacker to link the altered flow to the client.

Client identification attack on VPN or proxy | | +--------+ +--------+ | VPN |<==========>| | | Client |------------->| / |<==========>| Server | +--------+ | Proxy |<==========>| | .---->| | ^ +--------+ +--------+ / +-------+ | | Client |======' | +--------+ ^ Apply throughput advice signal \ \ Observe change ]]>

An attacker that can manipulate SCONE headers might cause an observable change in sending behavior; see . Though clients that use a VPN or proxy might choose to disable SCONE, removing SCONE signals is of little help against this form of attack. Lost or ECN-marked packets are likely to produce a congestion control response, which are alternative methods available to an attacker seeking to match flows. An effective, but wasteful, defense is to provide cover traffic between the client and intermediary to mask changes in sending rate on tunneled flows.

IANA Considerations This document registers new QUIC versions () and a QUIC transport parameter ().

SCONE Versions This document registers the following entries to the "QUIC Versions" registry maintained at https://www.iana.org/assignments/quic, following the guidance from .

Value:

0x6f7dc0fd

Status:

permanent

Specification:

This document

Change Controller:

IETF (iesg@ietf.org)

Contact:

QUIC Working Group (quic@ietf.org)

Notes:

SCONE Protocol - Even Signal Values

Value:

0xef7dc0fd

Status:

permanent

Specification:

This document

Change Controller:

IETF (iesg@ietf.org)

Contact:

QUIC Working Group (quic@ietf.org)

Notes:

SCONE Protocol - Odd Signal Values

scone_supported Transport Parameter This document registers the scone_supported transport parameter in the "QUIC Transport Parameters" registry maintained at https://www.iana.org/assignments/quic, following the guidance from .

Value:

0x219e

Parameter Name:

scone_supported

Status:

Permanent

Specification:

This document

Change Controller:

IETF (iesg@ietf.org)

Contact:

QUIC Working Group (quic@ietf.org)

Notes:

(none)

References Normative References This document defines the core of the QUIC transport protocol. QUIC provides applications with flow-controlled streams for structured communication, low-latency connection establishment, and network path migration. QUIC includes security measures that ensure confidentiality, integrity, and availability in a range of deployment circumstances. Accompanying documents describe the integration of TLS for key negotiation, loss detection, and an exemplary congestion control algorithm. This document defines the properties of the QUIC transport protocol that are common to all versions of the protocol. In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. This document describes a method for negotiating the ability to send an arbitrary value for the second-most significant bit in QUIC packets. This document specifies QUIC version 2, which is identical to QUIC version 1 except for some trivial details. Its purpose is to combat various ossification vectors and exercise the version negotiation framework. It also serves as a template for the minimum changes in any future version of QUIC. Note that "version 2" is an informal name for this proposal that indicates it is the second version of QUIC to be published as a Standards Track document. The protocol specified here uses a version number other than 2 in the wire image, in order to minimize ossification risks. Informative References This memo specifies the incorporation of ECN (Explicit Congestion Notification) to TCP and IP, including ECN's use of two bits in the IP header. [STANDARDS-TRACK] Alibaba Inc. Uber Technologies Inc. University of Mons (UMONS) UCLouvain and WELRI Private Octopus Inc. Ericsson This document specifies a multipath extension for the QUIC protocol to enable the simultaneous usage of multiple paths for a single connection. It introduces explicit path identifiers to create, delete, and manage multiple paths. This document does not specify address discovery or management, nor how applications using QUIC schedule traffic over multiple paths. Verizon Nokia Meta Verizon This document describes the Applicability and Manageability considerations for providing throughput guidance to application endpoints. This guidance is specifically addressed within the context of telecommunications service provider networks utilizing the Standard Communication with Network Elements (SCONE) protocol. The goal of this document is to describe the potential benefits of applications using a transport that enables Explicit Congestion Notification (ECN). The document outlines the principal gains in terms of increased throughput, reduced delay, and other benefits when ECN is used over a network path that includes equipment that supports Congestion Experienced (CE) marking. It also discusses challenges for successful deployment of ECN. It does not propose new algorithms to use ECN nor does it describe the details of implementation of ECN in endpoint devices (Internet hosts), routers, or other network devices. This document describes how to proxy UDP in HTTP, similar to how the HTTP CONNECT method allows proxying TCP in HTTP. More specifically, this document defines a protocol that allows an HTTP client to create a tunnel for UDP communications through an HTTP server that acts as a proxy. This document describes a protocol for transferring unbounded streams of multimedia data. It specifies the data format of the files and the actions to be taken by the server (sender) and the clients (receivers) of the streams. It describes version 7 of this protocol. This document defines basic terminology for describing different types of Network Address Translation (NAT) behavior when handling Unicast UDP and also defines a set of requirements that would allow many applications, such as multimedia communications or online gaming, to work consistently. Developing NATs that meet this set of requirements will greatly increase the likelihood that these applications will function properly. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. This document defines the IP header field, called the DS (for differentiated services) field. [STANDARDS-TRACK] This memo describes an updated technique for incremental computation of the standard Internet checksum. It updates the method described in RFC 1141. This memo provides information for the Internet community. This memo does not specify an Internet standard of any kind. This document specifies version 6 of the Internet Protocol (IPv6). It obsoletes RFC 2460.

Acknowledgments made significant contributions to the original TRAIN specification that forms the basis for a large part of this document. The following people also contributed significantly to the development of the protocol: , , , , , , and .