Congestion Exposure (ConEx) M. Kuehlewind, Ed.
Internet-Draft ETH Zurich
Intended status: Experimental R. Scheffenegger
Expires: February 7, 2016 NetApp, Inc.
August 6, 2015

TCP modifications for Congestion Exposure
draft-ietf-conex-tcp-modifications-09

Abstract

Congestion Exposure (ConEx) is a mechanism by which senders inform the network about expected congestion based on congestion feedback from previous packets in the same flow. This document describes the necessary modifications to use ConEx with the Transmission Control Protocol (TCP).

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on February 7, 2016.

Copyright Notice

Copyright (c) 2015 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.


Table of Contents

1. Introduction

Congestion Exposure (ConEx) is a mechanism by which senders inform the network about expected congestion based on congestion feedback from previous packets in the same flow. ConEx concepts and use cases are further explained in [RFC6789]. The abstract ConEx mechanism is explained in [draft-ietf-conex-abstract-mech]. This document describes the necessary modifications to use ConEx with the Transmission Control Protocol (TCP).

The markings for ConEx signaling are defined in the ConEx Destination Option (CDO) for IPv6 [draft-ietf-conex-destopt]. Specifically, the use of four flags is defined: X (ConEx-capable), L (loss experienced), E (ECN experienced) and C (credit).

ConEx signaling is based on loss or Explicit Congestion Notification (ECN) marks [RFC3168] as congestion indications. The sender collects this congestion information based on existing TCP feedback mechanisms from the receiver to the sender. No changes are needed at the receiver to implement ConEx signaling. Therefore no additional negotiation is needed to implement and use ConEx at the sender. This document specifies the sender's actions that are needed to provide meaningful ConEx information to the network.

Section 2 provides an overview of the modifications needed for TCP senders to implement ConEx. First congestion information has to be extracted from TCP's loss or ECN feedback as described in section 3. Section 4 details how to set the CDO marking based on this congestion information. Section 5 discusses loss of packets carrying ConEx information. Section 6 discusses timeliness of the ConEx feedback signal, given congestion is a temporary state.

This document describes congestion accounting for TCP with and without the Selective Acknowledgment (SACK) extension [RFC2018] (in section 3.1). However, ConEx benefits from the more accurate information that SACK provides about the number of bytes dropped in the network. It is therefore preferable to use the SACK extension when using TCP with ConEx. The detailed mechanism to set the L flag in response to loss-based congestion feedback signal is given in section 4.1.

Whereas loss has to be minimized, ECN can provide more fine-grained feedback information. ConEx-based traffic measurement or management mechanisms could benefit from this. Unfortunately, the current ECN feedback mechanism does not reflect multiple congestion markings if they occur within the same Round-Trip Time (RTT). A more accurate feedback extension to ECN (AccECN) is proposed in a separate document [draft-kuehlewind-tcpm-accurate-ecn], as this is also useful for other mechanisms.

Congestion accounting for both classic ECN feedback and AccECN feedback is explained in detail in section 3.2. Setting the E flag in response to ECN-based congestion feedback is again detailed in section 4.1.

1.1. Requirements Language

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 [RFC2119].

2. Sender-side Modifications

This section gives an overview of actions that need to be taken by a TCP sender modified to use ConEx signaling.

In the TCP handshake, a ConEx sender MUST negotiate for SACK and ECN preferably with AccECN feedback. Therefore a ConEx sender MUST also implement SACK and ECN. Depending on the capability of the receiver, the following operation modes exist:

A ConEx sender MUST expose all congestion information to the network according to the congestion information received by ECN or based on loss information provided by the TCP feedback loop. A TCP sender SHOULD count congestion byte-wise (rather than packet-wise; see next paragraph). After any congestion notification, a sender MUST mark subsequent packets with the appropriate ConEx flag in the IP header. Furthermore, a ConEx sender must send enough credit to cover all experienced congestion for the connection so far, as well as the risk of congestion for the current transmission (see Section 4.2).

With SACK the number of lost payload bytes is known, but not the number of packets carrying these bytes. With classic ECN only an indication is given that a marking occurred but not the exact number of payload bytes nor packets. As network congestion is usually byte-congestion [RFC7141], the byte-size of a packet marked with a CDO flag is defined to represent that number of bytes of congestion signalling [draft-ietf-conex-destopt]. Therefore the exact number of bytes should be taken into account, if available, to make the ConEx signal as exact as possible.

Detailed mechanisms for congestion counting in each operation mode are described in the next section.

3. Counting congestion

A ConEx TCP sender maintains two counters: one that counts congestion based on the information retrieved by loss detection, and a second that accounts for ECN based congestion feedback. These counters hold the number of outstanding bytes that should be ConEx marked with respectively the E flag or the L flag in subsequent packets.

The outstanding bytes for congestion indications based on loss are maintained in the loss exposure gauge (LEG), as explained in Section 3.1.

The outstanding bytes counted based on ECN feedback information are maintained in the congestion exposure gauge (CEG), as explained in Section 3.2.

When the sender sends a ConEx capable packet with the E or L flag set it reduces the respective counter by the byte-size of the packet. This is explained for both counters in Section 4.1.

Note that all bytes of an IP packet must be counted in the LEG or CEG to capture the right number of bytes that should be marked. Therefore the sender SHOULD take the payload and headers into account, up to and including the IP header. However, in TCP the information how large the headers of an lost or marked pacekt were is usually not available, as only payload data will be acknowledged.

If equal-sized packets, or at least equally distributed packet sizes can be assumed, the sender MAY only add and subtract TCP payload bytes. In this case there should be about the same number of ConEx marked packets as the original packets that were causing the congestion. Thus both contain about the same number of header bytes so they will cancel out. This case is assumed for simplicity in the following sections.

Otherwise, if a sender sends different sized packets (with unequally distributed packet sizes), the sender needs to memorize or estimate the number of lost or ECN-marked packets. If the sender has sufficient memory available, the most accurate way to reconstruct the number of lost or marked packets is to remember the sequence number of all sent but not acknowledged packets. In this case a sender is able to reconstruct the number of packets and thus the header bytes that were sent during the last RTT. Otherwise, if e.g. not enough memory is available, the sender should estimate the packet size, e.g. if the packet size distribution follows a certain known pattern, or by using the minimum packet size seen in the last RTT.

If the number of newly sent-out packets with the ConEx L or E flag set is smaller (or larger) than this estimated number of lost/ECN-marked packets, the additional header bytes should be added to (or can be subtracted from) the respective gauge.

3.1. Loss Detection

This section applies whether or not SACK support is available. The following subsection in addition handles the case when SACK is not available.

A TCP sender detects losses and subsequently retransmits the lost data. Therefore, ConEx sender can simply set the ConEx L flag on all retransmissions in order to at least cover the amount of bytes lost. If this aprroach is taken, no LEG is needed.

However, any retransmission may be spurious. In this case more bytes have been marked than necessary. To compensate this effect a ConEx sender can maintain a local signed counter, the (LEG), that indicats the number of outstanding bytes to be sent with the ConEx L flag and also can become negative.

Using the LEG, when a TCP sender decides that a data segment needs to be retransmitted, it will increase LEG by the size of the TCP payload bytes in the retransmission (assuming equal sized segments such that the retransmitted packet will have the same number of header bytes as the original ones):

For each retransmision:

LEG += payload

Note, how the LEG is reduced when the ConEx L marking are set is described in section Section 4.

Further to accommodate spurious restransmissions, a ConEx sender SHOULD make use of heuristics to detect such spurious retransmissions (e.g. F-RTO [RFC5682], DSACK [RFC3708], and Eifel [RFC3522], [RFC4015]) if are already available in a given implementation. If no mechanism for detecting spurious retransmissions is available, the ConEx sender MAY chose to implement one of the mechanism stated above. However, given the inaccuracy that ConEx may have anyway and the timeliness of ConEx information, a ConEx MAY also chose to not componsate for spurious retransmission. In this case if spurious retransmissions occur, the ConEx sender simple has sent too much ConEx signals which e.g would decrease the congestion allowance in a ConEx policer unnecessary.

If a heuristic to detect spurious retransmission is used and has determined that a certain number of packets were retransmitted erroneously, the ConEx sender subtracts the payload size of these TCP packets from LEG.

If a spurious reransmission is detected:

LEG -= payload

Note that the LEG can get negative, if too many L marking have already been sent. This case is further discussed in section Section 6.

3.1.1. Without SACK Support

If multiple losses occur within one RTT and SACK is not used, it may take several RTTs until all lost data is retransmitted. With the scheme described above, the ConEx information will be delayed considerably, but timeliness is important for ConEx. However, for ConEx it is not important to know which data got lost but only how much. During the first RTT after the initial loss detection, the amount of received data and thus also the amount of lost data can be estimated based on the number of received ACKs.

Therefore a ConEx sender can use the following algorithm to estimated the number of lost bytes with an additional delay of one RTT using an additional Loss Estimation Counter (LEC):

   flight_bytes:      current flight size in bytes
   retransmit_bytes:  payload size of the retransmission

   At the first retransmission in a congestion event LEC is set:

      LEC = flight_bytes - 3*SMSS 

      (At this point of time in the transmission, in the worst case, 
      all packets in flight minus three that trigged the dupACks 
      could have been lost.)

   Then during the first RTT of the congestion event:

      For each retransmission:
         LEG += retransmit_bytes
         LEC -= retransmit_bytes
   
      For each ACK:
         LEC -= SMSS


   After one RTT:

      LEG += LEC

      (The LEC now estimates the number of outstanding bytes  
      that should be ConEx L marked.)


   After the first RTT for each following retransmissions:

      if (LEC > 0): LEC -= retransmit_bytes
      else if (LEC==0): LEG += retransmit_bytes

      if (LEC < 0): LEG += -LEC

      (The LEG is not increased for those bytes that were 
      already counted.)

3.2. ECN

ECN [RFC3168] is an IP/TCP mechanism that allows network nodes to mark packets with the Congestion Experienced (CE) mark instead of dropping them when congestion occurs.

A receiver might support 'classic' ECN, the more accurate ECN feedback scheme (AccECN), or neither. In the case that ECN is not supported for a connection, of course, no ECN marks will occur; thus the sender will never set the E flag. Otherwise, a ConEx sender needs to maintain a signed counter, the congestion exposure gauge (CEG), for the number of outstanding bytes that have to be ConEx marked with the E flag.

The CEG is increased when ECN information is received from an ECN-capable receiver supporting the 'classic' ECN scheme or the accurate ECN feedback scheme. When the ConEx sender receives an ACK indicating one or more segments were received with a CE mark, CEG is increased by the appropriate number of bytes as described further below.

Unfortunately in case of duplicate acknowledgements the number of newly acknowledged bytes will be zero even though (CE marked) data has been received. Therefore, we increase the CEG by DeliveredData, as defined below:

DeliveredData = acked_bytes + SACK_diff + (is_dup)*1SMSS - (is_after_dup)*num_dup*1SMSS +

DeliveredData covers the number of bytes that has been newly delivered to the receiver. Therefore on each arrival of an ACK, DeliveredData will be increased by the newly acknowledged bytes (acked_bytes) as indicated by the current ACK, relative to all past ACKs. The formula depends on whether SACK is available: if SACK is not avaialble SACK_diff is always zero, whereas is ACK information is available is_dup and is_after_dup are always zero.

With SACK, DeliveredData is increased by the number of bytes provided by (new) SACK information (SACK_diff). Note, if less unacknowledged bytes are announced in the new SACK information than in the previous ACK, SACK_diff can be negative. In this case, data is newly acknowledged (in acked_bytes), that has previously already been accumulated into DeliveredData based on SACK information.

Otherwise without SACK, DeliveredData is increased by 1 SMSS on duplicate acknowledgements as duplicate acknowledgements do not acknowlegde any new data (and acked_bytes will be zero). For the subsequent partial or full ACK, acked_bytes cover all newly acknowledged bytes including the ones that where already accounted which the receiption of any duplicate acknowledgement. Therefore DeliveredData is reduced by one SMSS for each preceding duplicate ACK. Consequently, is_dup is one if the current ACK is a duplicated ACK without SACK, and zero otherwise. is_after_dup is only one for the next full or partial ACK after a number of duplicated ACKs without SACK and num_dup counts the number of duplicated ACKs in a row (which usually is 3 or more).

With classic ECN, one congestion marked packet causes continuous congestion feedback for a whole round trip, thus hiding the arrival of any further congestion marked packets during that round trip. A more accurate ECN feedback scheme (AccECN) is needed to ensure that feedback properly reflects the extent of congestion marking. The two cases, with and without a receiver capable of AccECN, are discussed in the following sections.

3.2.1. Accurate ECN feedback

With a more accurate ECN feedback scheme (AccECN) that is supported by the receiver either the number of marked packets or the number of marked bytes will be feed back from the receiver to the sender and is therefore know at sender-side. In the latter case the CEG can directly be increased by the number of marked bytes. Otherwise if D is assumed to be the number of marks, the gauge (CEG) will be conservatively increased by one SMSS for each marking or at max the number of newly acknowledged bytes:

CEG += min(SMSS*D, DeliveredData)

3.2.2. Classic ECN support

With classic ECN, as soon as a CE mark is seen at the receiver, it will feed this information back to the sender by setting the Echo Congestion Experienced (ECE) flag in the TCP header of subsequent ACKs. Once the sender receives the first ECE of a congestion notification, it sets the CWR flag in the TCP header once. When this packet with Congestion Window Reduced (CWR) flag in the TCP header arrives at the receiver, acknowledging its first ECE feedback, the receiver stops setting ECE.

If the ConEx sender fully conforms to the semantics of ECN signaling as defined by [RFC3168], it will receive one full RTT of ACKs with the ECE flag set whenever at least one CE mark was received by the receiver. As the sender cannot estimate how many packets have actually been CE marked during this RTT, the most conservative assumption MAY be taken, namely assuming that all packets were marked. This can be achieved by increasing the CEG by DeliveredData for each ACK with the ECE flag:

CEG += DeliveredData

Optionally a ConEx sender could implement the following technique (that not conforms to [RFC3168]), called advanced compatibility mode, to considerably improve its estimate of the number of ECN-marked packets:

To extract more than one ECE indication per RTT, a ConEx sender could set the CWR flag continuously to force the receiver to signal only one ECE per CE mark. Unfortunately, the use of delayed ACKs [RFC5681] (which is common) will prevent feedback of every CE mark; if a CWR confirmation is received before the ECE can be sent out on the next ACK, ECN feedback information could get lost (depeding on the actual receiver implementation). Thus a sender SHOULD set CWR only on those data segments that will presumably trigger a (delayed) ACK. The sender would need an additional control loop to estimated which data segments will trigger an ACK in order to extract more timely congestion notifications. Still the CEG SHOULD be increased by DeliveredData, as one or more CE marked packets could be acknowledged by one delayed ACK.

4. Setting the ConEx Flags

By setting the X flag, a packet is marked as ConEx-capable. All packets carrying payload MUST be marked with the X flag set, including retransmissions. Only if no congestion feedback information is (currently) available, the X flag SHOULD be zero, such as for control packets on a connection that has not sent any (user) data for some time e.g., sending only pure ACKs which are not carrying any payload.

4.1. Setting the E or the L Flag

As described in section Section 3.1, the sender needs to maintain a CEG counter and might maintain a LEG counter. If no LEG is used, all retransmission will be marked with the L flag.

Further, as long as the LEG or CEG counter is positive, the sender marks each ConEx-capable packet with L or E respectively, and decreases the LEG or CEG counter by the TCP payload bytes carried in the marked packet (assuming headers are not being counted because packet sizes are regular). No matter how small the value of LEG or CEG, if it is positive, the sender MUST NOT defer packet marking to ensure ConEx signals are timely. Therefore the value of LEG and CEG will commonly be negative.

If both LEG and CEG are positive, the sender MUST mark each ConEx-capable packet with both L and E. If a credit signal is also pending (see next section), the C flag can be set as well.

4.2. Setting the Credit Flag

The ConEx abstract mechanism [draft-ietf-conex-abstract-mech] requires that sufficient credit MUST be signaled in advance to cover the expected congestion during the feedback delay of one RTT.

To monitor the credit state at the audit, a ConEx sender needs to maintain a Credit State Counter (CSC) in bytes. If congestion occurs, credits will be consumed and the CSC is reduced by the number of bytes that where lost or estimated to be ECN-marked. If the risk of congestion was estimated wrongly and thus too few credits were sent, the CSC becomes zero but cannot go negative.

To be sure that the credit state in the audit never reaches zero, the number of credits should always equal the number of bytes in flight as all packets could potentially get lost or congestion marked. In this case a ConEx sender also monitors the number of bytes in flight F. If F ever becomes larger than CSC, the ConEx sender sets the C flag on each ConEx-capable packet and increase CSC by the payload size of each marked packet until CSC is no less than F again. However, a ConEx sender might also be less conservative and send fewer credits, if it e.g. assumes based on previous experience that the congestion will be low on a certain path.

Recall that CSC will be decreased whenever congestion occurs, therefore CSC will need to be replenished as soon as CSC drops below F. Also recall that the sender can set the C flag on a ConEx-capable packet whether or not the E or L flags are also set.

In TCP Slow Start, the congestion window might grow much larger than during the rest of the transmission. Likely, a sender could consider sending fewer than F credits but risking being penalized by an audit function. However, the credits should at least cover the increase in sending rate. Given the exponential increase as implemented in the TCP Slow Start algorithm which means that the sending rate doubles every RTT, a ConEx sender should at least cover half the number of packets in flight by credits.

Note that the number of losses or markings within one RTT does not solely depend on the sender's actions. In general, the behavior of the cross traffic, whether active queue management (AQM) is used and how it is parameterized influence how many packets might be dropped or marked. As long as any AQM encountered is not overly aggressive with ECN marking, sending half the flight size as credits should be sufficient whether congestion is signaled by loss or ECN.

To maintain half of the packet in flight as credits, of course half of the packet of the initial window must be C marked. In Slow Start marking every fourth packet introduces the correct amount of credit as can be seen in Figure 1.

                        in_flight  credits
RTT1  |------XC------>|     1         1
      |------X------->|     2         1
      |------XC------>|     3         2
      |               |
RTT2  |------X------->|     3         2
      |------X------->|     4         2
      |------X------->|     4         2
      |------XC------>|     5         3
      |------X------->|     5         3
      |------X------->|     6         3
      |               |
RTT3  |------X------->|     6         3
      |------XC------>|     7         4
      |------X------->|     7         4
      |------X------->|     8         4
      |------X------->|     8         4
      |------XC------>|     9         5
      |------X------->|     9         5
      |------X------->|    10         5
      |------X------->|    10         5
      |------XC------>|    11         6
      |------X------->|    11         6
      |------X------->|    12         6
      |      .        |
      |      :        |
      

Figure 1: Credits in Slow Start (with an initial window of 3)

It is possible that a TCP flow will encounter an audit function without relevant flow state, due to e.g. rerouting or memory limitations. Therefore, the sender needs to detect this case and resend credits. A ConEx sender might reset the credit counter CSC to zero if losses occur in subsequent RTTs (assuming that the sending rate was correctly reduced based on the received congestion signal and using a conservatively large RTT estimation).

This section proposes a concrete algorithm for determining how much credit to signal (with a separate approach used for Slow Start). However, experimentation in credit setting algorithms is expected and encouraged. The wider goal of ConEx is to reflect the 'cost' of the risk of causing congestion on those that contribute most to it. Thus, experimentation is encouraged to improve or maintain performance while reducing the risk of causing congestion, and therefore potentially reducing the need to signal so much credit.

5. Loss of ConEx information

Packets carrying ConEx signals could be discarded themselves. This will be a second order problem (e.g. if the loss probability is 0.1%, the probability of losing a ConEx L signal will be 0.1% of 0.1% = 0.01%). Further, the penality an audit induces should be propotional to the mismatch of expected ConEx marks and observed congestion, therefore the audit might only slightly increase the loss level of this flow. Therefore, an implementer MAY choose to ignore this problem, accepting instead the risk that an audit function might wrongly penalize a flow.

Nonetheless, a ConEx sender is responsible to always signal sufficient congestion feedback and therefore SHOULD remember which packet was marked with either the L, the E or the C flag. If one of these packets is detected as lost, the sender SHOULD increase the respective gauge(s), LEG or CEG, by the number of lost payload bytes in addition to increasing LEG for the loss.

6. Timeliness of the ConEx Signals

ConEx signals will only be useful to a network node within a time delay of about one RTT after the congestion occurred. To avoid further delays, a ConEx sender SHOULD send the ConEx signaling on the next available packet.

Any or all of the ConEx flags can be used in the same packet, which allows delay to be minimised when multiple signals are pending. The need to set multiple ConEx flags at the same time, can occur if e.g an ACK is received by the sender that simultaneously indicates that at least one ECN mark was received, and that one or more segements were lost. This may e.g. happen during excessive congestion, where the queues overflow even though ECN was used and currently all forwarded packets are marked, while others have to be dropped nevertheless. Another case when this might happen is when ACKs are lost, so that a subsequent ACK carries summary information not previously available to the sender.

If a flow becomes application-limited, there could be insufficient bytes to send to reduce the gauges to zero or below. In such cases, the sender cannot help but delay ConEx signals. Nonetheless, as long as the sender is marking all outgoing packets, an audit function is unlikely to penalize ConEx-marked packets. Therefore, no matter how long a gauge has been positive, a sender MUST NOT reduce the gauge by more than the ConEx marked bytes it has sent.

If the CEG or LEG counter is negative, the respective counter MAY be reset to zero within one RTT after it was decreased the last time or one RTT after recovery if no further congestion occurred.

7. Open Areas for Experimentation

All proposed mechanisms in this document are experimental, and therefore further large-scale experimentation in the Internet is required to evaluate if the signaling provided by these mechanisms is accurate and timely enough to produce value for ConEx-based (traffic management or other) mechanisms.

The current ConEx specifications assume that congestion is counted in number of bytes (including the IP header that directly encapsulates the CDO and everything that IP header encapsulates) [draft-ietf-conex-destopt]. This decision was taken because most network devices today expirience byte-congestion where the memory is filled exactly with the number of bytes a packet carries. However, there are also devices that may allocate a certain amount of memory per packet, no matter how larger a packet is. These devices get congested based on the number of packets in their memory and therefore in this case congestion is determined by the number of packets that have been lost or marked. Furthermore, a transport layer endpoint, such as a TCP sender or receiver, might not know the exact number of bytes that a lower layer was carrying. Therefore a TCP endpoint may only be able to estimate the exact number of congested bytes (assuming that all lower layer header have the same length). If this estimation is sufficient to work with the ConEx signal needs to be further evaluated in tests in the Internet together with different auditor implementations.

Further, the proposed marking schemes in this document are designed under the assumption that all TCP packets of a ConEx-capable flow are of equal size or that flows have a constant mean packet size over a rather small time frame, like one RTT or less. In most implementations this assumption might be taken as well and probably is true for most of the traffic flows. However, it should be evaluted how much the accuracy degrades if this precondition is not fulfilled, while the proposed scheme is used. Especially evaluating this with real traffic from different application is important to make a decision if the proposed schemes are sufficient or a more complexe scheme is needed.

In this context the proposed scheme to set credit markings in Slow Start runs a risk to provide an insufficient number of markings which can cause an audit function to penalize this flow. Both the proposed credit scheme for Slow Start as well as the scheme in Congestion Avoidance must be evaluated together with one or more specific implementations of an ConEx auditor to ensure that both algorithms, in the sender and in the auditor, work propoerly together with a low risk of false positives (which would lead to penalization of an honest sender). However, if a sender is wrongly assumed to cheat, the penalization of the audit should be adequate and should allow an honest sender using a congestion control scheme that is commonly used today to recover quickly.

Another open issue is the accuracy of the ECN feedback signal. At time of publication of this document there is no AccECN mechanism s pecified yet, and further AccECN will also take some time to be widely deployed. This document proposes an advanced compatibility mode for Classic ECN. The proposed mechanism can provide more accurate feedback by utilizing the way Classic ECN is speficed but has a higher risk of loosing information. To figure out how high this risk is in a real deployment scenario, further experimental evaluation is needed. The following argument is intended to prove that suppressing repetitions of ECE, however, is still safe against possible congestion collapse due to lost congestion feedback and should be further proven in experimentation:

Repetition of ECE in classic ECN is intended to ensure reliable delivery of congestion feedback. However, with advanced compatibility mode, it is possible to miss congestion notifications. This can happen in some implementations if delayed acknowledgements are used. Further an ACK containing ECE can simply get lost. If only a few CE marks are received within one congestion event (e.g., only one), the loss of one acknowledgements due to (heavy) congestion on the reverse path can prevent that any congestion notification is received by the sender.

However, if loss of feedback exacerbates congestion on the forward path, more forward packets will be CE marked, increasing the likelihood that feedback from at least one CE will get through per RTT. As long as one ECE reaches the sender per RTT, the sender's congestion response will be the same as if CWR were not continuous. The only way that heavy congestion on the forward path could be completely hidden would be if all ACKs on the reverse path were lost. If total ACK loss persisted, the sender would time out and do a congestion response anyway. Therefore, the problem seems confined to potential suppression of a congestion response during light congestion.

Anyway, even if loss of all ECN feedback leads to no congestion response, the worst that could happen would be loss instead of ECN-signaled congestion on the forward path. Given compatibility mode does not affect loss feedback, there would be no risk of congestion collapse.

8. Acknowledgements

The authors would like to thank Bob Briscoe who contributed with this initial ideas [I-D.briscoe-conex-re-ecn-tcp] and valuable feedback. Moreover, thanks to Jana Iyengar who also provided valuable feedback.

9. IANA Considerations

This document does not have any requests to IANA.

10. Security Considerations

General ConEx security considerations are covered extensively in the ConEx abstract mechanism [draft-ietf-conex-abstract-mech]. This section covers TCP-specific concerns.

The ConEx modifications to TCP provide no mechanism for a receiver to force a sender not to use ConEx. A receiver can degrade the accuracy of ConEx by claiming that it does not support SACK, AccECN or ECN, but the sender will never have to turn ConEx off. The receiver cannot force the sender to have to mark ConEx more conservatively, in order to cover the risk of any inaccuracy. Instead the sender can choose to mark inaccurately, which will only increase the likelihood of loss at an audit function. Thus the receiver will only harm itself.

Assuming the sender is limited in some way by a congestion allowance or quota, a receiver could spoof more loss or ECN congestion feedback than it actually experiences, in an attempt to make the sender draw down its allowance faster than necessary. However, over-declaring congestion simply makes the sender slow down. If the receiver is interested in the content it will not want to harm its own performance.

However, if the receiver is solely interested in making the sender draw down its allowance, the net effect will depend on the sender's congestion control algorithm as permanetly adding more and more additional congestion would cause the sender to more and more reduce its sending rate. Therefore a receiver can only maintain a certain congestion level that is corresponding to a certain sending rate. With New Reno [RFC5681], doubling congestion feedback causes the sender to reduce its sending rate such that it would only to consume sqrt(2) = 1.4 times more congestion allowance. However, to improve scaling, congestion control algorithms are tending towards less responsive algorithms like Cubic or Compound TCP, and ultimately to linear algorithms like DCTCP [DCTCP] that aim to maintain the same congestion level independent of the current sending rate and always reduce its sending window if the signaled congestion feedback is higher. In each case, if the receiver doubles congestion feedback, it causes the sender to respectively consume more allowance by a factor of 1.2, 1.15 or 1, where 1 implies the attack has become completely ineffective as no further congestion allowance is consumed but the flow will decrease its sending rate to a minimum instead.

11. References

11.1. Normative References

[draft-ietf-conex-abstract-mech] Mathis, M. and B. Briscoe, "Congestion Exposure (ConEx) Concepts and Abstract Mechanism", Internet-Draft draft-ietf-conex-abstract-mech-06, October 2012.
[draft-ietf-conex-destopt] Krishnan, S., Kuehlewind, M. and C. Ucendo, "IPv6 Destination Option for ConEx", Internet-Draft draft-ietf-conex-destopt-04, March 2013.
[RFC2018] Mathis, M., Mahdavi, J., Floyd, S. and A. Romanow, "TCP Selective Acknowledgment Options", RFC 2018, DOI 10.17487/RFC2018, October 1996.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997.
[RFC3168] Ramakrishnan, K., Floyd, S. and D. Black, "The Addition of Explicit Congestion Notification (ECN) to IP", RFC 3168, DOI 10.17487/RFC3168, September 2001.
[RFC5681] Allman, M., Paxson, V. and E. Blanton, "TCP Congestion Control", RFC 5681, DOI 10.17487/RFC5681, September 2009.

11.2. Informative References

[DCTCP] Alizadeh, M., Greenberg, A., Maltz, D., Padhye, J., Patel, P., Prabhakar, B., Sengupta, S. and M. Sridharan, "DCTCP: Efficient Packet Transport for the Commoditized Data Center", Jan 2010.
[draft-kuehlewind-tcpm-accurate-ecn] Kuehlewind, M. and R. Scheffenegger, "More Accurate ECN Feedback in TCP", Internet-Draft draft-kuehlewind-tcpm-accurate-ecn-02, Jun 2013.
[I-D.briscoe-conex-re-ecn-tcp] Briscoe, B., Jacquet, A., Moncaster, T. and A. Smith, "Re-ECN: Adding Accountability for Causing Congestion to TCP/IP", Internet-Draft draft-briscoe-conex-re-ecn-tcp-04, July 2014.
[RFC3522] Ludwig, R. and M. Meyer, "The Eifel Detection Algorithm for TCP", RFC 3522, DOI 10.17487/RFC3522, April 2003.
[RFC3708] Blanton, E. and M. Allman, "Using TCP Duplicate Selective Acknowledgement (DSACKs) and Stream Control Transmission Protocol (SCTP) Duplicate Transmission Sequence Numbers (TSNs) to Detect Spurious Retransmissions", RFC 3708, DOI 10.17487/RFC3708, February 2004.
[RFC4015] Ludwig, R. and A. Gurtov, "The Eifel Response Algorithm for TCP", RFC 4015, DOI 10.17487/RFC4015, February 2005.
[RFC5682] Sarolahti, P., Kojo, M., Yamamoto, K. and M. Hata, "Forward RTO-Recovery (F-RTO): An Algorithm for Detecting Spurious Retransmission Timeouts with TCP", RFC 5682, DOI 10.17487/RFC5682, September 2009.
[RFC6789] Briscoe, B., Woundy, R. and A. Cooper, "Congestion Exposure (ConEx) Concepts and Use Cases", RFC 6789, DOI 10.17487/RFC6789, December 2012.
[RFC7141] Briscoe, B. and J. Manner, "Byte and Packet Congestion Notification", BCP 41, RFC 7141, DOI 10.17487/RFC7141, February 2014.

Appendix A. Revision history

RFC Editor: This section is to be removed before RFC publication.

00 ... initial draft, early submission to meet deadline.

01 ... refined draft, updated LEG "drain" from per-packet to RTT-based.

02 ... added Section 5 and expanded discussion about ECN interaction.

03 ... expanded the discussion around credit bits.

04 ... review comments of Jana addressed. (Change in full compliance mode.)

05 ... changes on Loss Detection without SACK, support of classic ECN and credit handling.

07 ... review feedback provided by Nandita

08 ... based on Bob's feedback: Wording edits and structuring of a few paragraphs; change of SHOULD to MAY for resetting negative LEG/CEG; additional security considerations provided by Bob (thanks!).

Authors' Addresses

Mirja Kuehlewind (editor) ETH Zurich Switzerland EMail: mirja.kuehlewind@tik.ee.ethz.ch
Richard Scheffenegger NetApp, Inc. Am Euro Platz 2 Vienna, 1120 Austria Phone: +43 1 3676811 3146 EMail: rs@netapp.com