Internet Engineering Task Force (IETF)                         M. Allman
Request for Comments: 5827                                          ICSI
Category: Experimental                                    K. Avrachenkov
ISSN: 2070-1721                                                    INRIA
                                                              U. Ayesta
                                          BCAM-IKERBASQUE and LAAS-CNRS
                                                             J. Blanton
                                                        Ohio University
                                                              P. Hurtig
                                                    Karlstad University
                                                             April 2010


                       Early Retransmit for TCP
           and Stream Control Transmission Protocol (SCTP)

Abstract

  This document proposes a new mechanism for TCP and Stream Control
  Transmission Protocol (SCTP) that can be used to recover lost
  segments when a connection's congestion window is small.  The "Early
  Retransmit" mechanism allows the transport to reduce, in certain
  special circumstances, the number of duplicate acknowledgments
  required to trigger a fast retransmission.  This allows the transport
  to use fast retransmit to recover segment losses that would otherwise
  require a lengthy retransmission timeout.

Status of This Memo

  This document is not an Internet Standards Track specification; it is
  published for examination, experimental implementation, and
  evaluation.

  This document defines an Experimental Protocol for the Internet
  community.  This document is a product of the Internet Engineering
  Task Force (IETF).  It represents the consensus of the IETF
  community.  It has received public review and has been approved for
  publication by the Internet Engineering Steering Group (IESG).  Not
  all documents approved by the IESG are a candidate for any level of
  Internet Standard; see Section 2 of RFC 5741.

  Information about the current status of this document, any errata,
  and how to provide feedback on it may be obtained at
  http://www.rfc-editor.org/info/rfc5827.







Allman, et al.                Experimental                      [Page 1]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


Copyright Notice

  Copyright (c) 2010 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.

  This document may contain material from IETF Documents or IETF
  Contributions published or made publicly available before November
  10, 2008.  The person(s) controlling the copyright in some of this
  material may not have granted the IETF Trust the right to allow
  modifications of such material outside the IETF Standards Process.
  Without obtaining an adequate license from the person(s) controlling
  the copyright in such materials, this document may not be modified
  outside the IETF Standards Process, and derivative works of it may
  not be created outside the IETF Standards Process, except to format
  it for publication as an RFC or to translate it into languages other
  than English.

1.  Introduction

  Many researchers have studied the problems with TCP's loss recovery
  [RFC793, RFC5681] when the congestion window is small, and they have
  outlined possible mechanisms to mitigate these problems
  [Mor97, BPS+98, Bal98, LK98, RFC3150, AA02].  SCTP's [RFC4960] loss
  recovery and congestion control mechanisms are based on TCP, and
  therefore the same problems impact the performance of SCTP
  connections.  When the transport detects a missing segment, the
  connection enters a loss recovery phase.  There are several variants
  of the loss recovery phase depending on the TCP implementation.  TCP
  can use slow-start-based recovery or fast recovery [RFC5681], NewReno
  [RFC3782], and loss recovery, based on selective acknowledgments
  (SACKs) [RFC2018, FF96, RFC3517].  SCTP's loss recovery is not as
  varied due to the built-in selective acknowledgments.

  All of the above variants have two methods for invoking loss
  recovery.  First, if an acknowledgment (ACK) for a given segment is
  not received in a certain amount of time, a retransmission timer
  fires, and the segment is resent [RFC2988, RFC4960].  Second, the
  "fast retransmit" algorithm resends a segment when three duplicate



Allman, et al.                Experimental                      [Page 2]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


  ACKs arrive at the sender [Jac88, RFC5681].  Duplicate ACKs are
  triggered by out-of-order arrivals at the receiver.  However, because
  duplicate ACKs from the receiver are triggered by both segment loss
  and segment reordering in the network path, the sender waits for
  three duplicate ACKs in an attempt to disambiguate segment loss from
  segment reordering.  When the congestion window is small, it may not
  be possible to generate the required number of duplicate ACKs to
  trigger fast retransmit when a loss does happen.

  Small congestion windows can occur in a number of situations, such
  as:

  (1) The connection is constrained by end-to-end congestion control
      when the connection's share of the path is small, the path has a
      small bandwidth-delay product, or the transport is ascertaining
      the available bandwidth in the first few round-trip times of slow
      start.

  (2) The connection is "application limited" and has only a limited
      amount of data to send.  This can happen any time the application
      does not produce enough data to fill the congestion window.  A
      particular case when all connections become application limited
      is as the connection ends.

  (3) The connection is limited by the receiver's advertised window.

  The transport's retransmission timeout (RTO) is based on measured
  round-trip times (RTT) between the sender and receiver, as specified
  in [RFC2988] (for TCP) and [RFC4960] (for SCTP).  To prevent spurious
  retransmissions of segments that are only delayed and not lost, the
  minimum RTO is conservatively chosen to be 1 second.  Therefore, it
  behooves TCP senders to detect and recover from as many losses as
  possible without incurring a lengthy timeout during which the
  connection remains idle.  However, if not enough duplicate ACKs
  arrive from the receiver, the fast retransmit algorithm is never
  triggered -- this situation occurs when the congestion window is
  small, if a large number of segments in a window are lost, or at the
  end of a transfer as data drains from the network.  For instance,
  consider a congestion window of three segments' worth of data.  If
  one segment is dropped by the network, then at most two duplicate
  ACKs will arrive at the sender.  Since three duplicate ACKs are
  required to trigger fast retransmit, a timeout will be required to
  resend the dropped segment.  Note that delayed ACKs [RFC5681] may
  further reduce the number of duplicate ACKs a receiver sends.
  However, we assume that receivers send immediate ACKs when there is a
  gap in the received sequence space per [RFC5681].





Allman, et al.                Experimental                      [Page 3]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


  [BPS+98] shows that roughly 56% of retransmissions sent by a busy Web
  server are sent after the RTO timer expires, while only 44% are
  handled by fast retransmit.  In addition, only 4% of the RTO timer-
  based retransmissions could have been avoided with SACK, which has to
  continue to disambiguate reordering from genuine loss.  Furthermore,
  [All00] shows that for one particular Web server, the median number
  of bytes carried by a connection is less than four segments,
  indicating that more than half of the connections will be forced to
  rely on the RTO timer to recover from any losses that occur.  Thus,
  loss recovery that does not rely on the conservative RTO is likely to
  be beneficial for short TCP transfers.

  The limited transmit mechanism introduced in [RFC3042] and currently
  codified in [RFC5681] allows a TCP sender to transmit previously
  unsent data upon receipt of each of the two duplicate ACKs that
  precede a fast retransmit.  SCTP [RFC4960] uses SACK information to
  calculate the number of outstanding segments in the network.  Hence,
  when the first two duplicate ACKs arrive at the sender, they will
  indicate that data has left the network, and they will allow the
  sender to transmit new data (if available), similar to TCP's limited
  transmit algorithm.  In the remainder of this document, we use
  "limited transmit" to include both TCP and SCTP mechanisms for
  sending in response to the first two duplicate ACKs.  By sending
  these two new segments, the sender is attempting to induce additional
  duplicate ACKs (if appropriate), so that fast retransmit will be
  triggered before the retransmission timeout expires.  The sender-side
  "Early Retransmit" mechanism outlined in this document covers the
  case when previously unsent data is not available for transmission
  (case (2) above) or cannot be transmitted due to an advertised window
  limitation (case (3) above).

  Note: This document is being published as an experimental RFC, as
  part of the process for the TCPM working group and the IETF to assess
  whether the proposed change is useful and safe in the heterogeneous
  environments, including which variants of the mechanism are the most
  effective.  In the future, this specification may be updated and put
  on the standards track if its safeness and efficacy can be
  demonstrated.

2.  Terminology

  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 RFC 2119 [RFC2119].

  The reader is expected to be familiar with the definitions given in
  [RFC5681].




Allman, et al.                Experimental                      [Page 4]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


3.  Early Retransmit Algorithm

  The Early Retransmit algorithm calls for lowering the threshold for
  triggering fast retransmit when the amount of outstanding data is
  small and when no previously unsent data can be transmitted (such
  that limited transmit could be used).  Duplicate ACKs are triggered
  by each arriving out-of-order segment.  Therefore, fast retransmit
  will not be invoked when there are less than four outstanding
  segments (assuming only one segment loss in the window).  However,
  TCP and SCTP are not required to track the number of outstanding
  segments, but rather the number of outstanding bytes or messages.
  (Note that SCTP's message boundaries do not necessarily correspond to
  segment boundaries.)  Therefore, applying the intuitive notion of a
  transport with less than four segments outstanding is more
  complicated than it first appears.  In Section 3.1, we describe a
  "byte-based" variant of Early Retransmit that attempts to roughly map
  the number of outstanding bytes to a number of outstanding segments
  that is then used when deciding whether to trigger Early Retransmit.
  In Section 3.2, we describe a "segment-based" variant that represents
  a more precise algorithm for triggering Early Retransmit.  This
  precision comes at the cost of requiring additional state to be kept
  by the TCP sender.  In both cases, we describe SACK-based and non-
  SACK-based versions of the scheme (of course, the non-SACK version
  will not apply to SCTP).  This document explicitly does not prefer
  one variant over the other, but leaves the choice to the implementer.

3.1.  Byte-Based Early Retransmit

  A TCP or SCTP sender MAY use byte-based Early Retransmit.

  Upon the arrival of an ACK, a sender employing byte-based Early
  Retransmit MUST use the following two conditions to determine when an
  Early Retransmit is sent:

  (2.a) The amount of outstanding data (ownd) -- data sent but not yet
        acknowledged -- is less than 4*SMSS bytes (as defined in
        [RFC5681]).

        Note that in the byte-based variant of Early Retransmit, "ownd"
        is equivalent to "FlightSize" (defined in [RFC5681]).  We use
        different notation, because "ownd" is not consistent with
        FlightSize throughout this document.

        Also note that in SCTP, messages will have to be converted to
        bytes to make this variant of Early Retransmit work.






Allman, et al.                Experimental                      [Page 5]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


  (2.b) There is either no unsent data ready for transmission at the
        sender, or the advertised receive window does not permit new
        segments to be transmitted.

  When the above two conditions hold and a TCP connection does not
  support SACK, the duplicate ACK threshold used to trigger a
  retransmission MUST be reduced to:

               ER_thresh = ceiling (ownd/SMSS) - 1                 (1)

  duplicate ACKs, where ownd is expressed in terms of bytes.  We call
  this reduced ACK threshold enabling "Early Retransmission".

  When conditions (2.a) and (2.b) hold and a TCP connection does
  support SACK or SCTP is in use, Early Retransmit MUST be used only
  when "ownd - SMSS" bytes have been SACKed.

  If either (or both) condition (2.a) and/or (2.b) does not hold, the
  transport MUST NOT use Early Retransmit, but rather prefer the
  standard mechanisms, including fast retransmit and limited transmit.

  As noted above, the drawback of this byte-based variant is precision
  [HB08].  We illustrate this with two examples:

     + Consider a non-SACK TCP sender that uses an SMSS of 1460 bytes
       and transmits three segments, each with 400 bytes of payload.
       This is a case where Early Retransmit could aid loss recovery if
       one segment is lost.  However, in this case, ER_thresh will
       become zero, per Equation (1), because the number of outstanding
       bytes is a poor estimate of the number of outstanding segments.
       A similar problem occurs for senders that employ SACK, as the
       expression "ownd - SMSS" will become negative.

     + Next, consider a non-SACK TCP sender that uses an SMSS of
       1460 bytes and transmits 10 segments, each with 400 bytes of
       payload.  In this case, ER_thresh will be 2 per Equation (1).
       Thus, even though there are enough segments outstanding to
       trigger fast retransmit with the standard duplicate ACK
       threshold, Early Retransmit will be triggered.  This could cause
       or exacerbate performance problems caused by segment reordering
       in the network.










Allman, et al.                Experimental                      [Page 6]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


3.2.  Segment-Based Early Retransmit

  A TCP or SCTP sender MAY use segment-based Early Retransmit.

  Upon the arrival of an ACK, a sender employing segment-based Early
  Retransmit MUST use the following two conditions to determine when an
  Early Retransmit is sent:

  (3.a) The number of outstanding segments (oseg) -- segments sent but
        not yet acknowledged -- is less than four.

  (3.b) There is either no unsent data ready for transmission at the
        sender, or the advertised receive window does not permit new
        segments to be transmitted.

  When the above two conditions hold and a TCP connection does not
  support SACK, the duplicate ACK threshold used to trigger a
  retransmission MUST be reduced to:

                 ER_thresh = oseg - 1                              (2)

  duplicate ACKs, where oseg represents the number of outstanding
  segments.  (We discuss tracking the number of outstanding segments
  below.)  We call this reduced ACK threshold enabling "Early
  Retransmission".

  When conditions (3.a) and (3.b) hold and a TCP connection does
  support SACK or SCTP is in use, Early Retransmit MUST be used only
  when "oseg - 1" segments have been SACKed.  A segment is considered
  to be SACKed when all of its data bytes (TCP) or data chunks (SCTP)
  have been indicated as arrived by the receiver.

  If either (or both) condition (3.a) and/or (3.b) does not hold, the
  transport MUST NOT use Early Retransmit, but rather prefer the
  standard mechanisms, including fast retransmit and limited transmit.

  This version of Early Retransmit solves the precision issues
  discussed in the previous section.  As noted previously, the cost is
  that the implementation will have to track segment boundaries to form
  an understanding as to how many actual segments have been
  transmitted, but not acknowledged.  This can be done by the sender
  tracking the boundaries of the three segments on the right side of
  the current window (which involves tracking four sequence numbers in
  TCP).  This could be done by keeping a circular list of the segment
  boundaries, for instance.  Cumulative ACKs that do not fall within
  this region indicate that at least four segments are outstanding, and
  therefore Early Retransmit MUST NOT be used.  When the outstanding
  window becomes small enough that Early Retransmit can be invoked, a



Allman, et al.                Experimental                      [Page 7]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


  full understanding of the number of outstanding segments will be
  available from the four sequence numbers retained.  (Note: the
  implicit sequence number consumed by the TCP FIN bit can also be
  included in the tracking of segment boundaries.)

4.  Discussion

  In this section, we discuss a number of issues surrounding the Early
  Retransmit algorithm.

4.1.  SACK vs. Non-SACK

  The SACK variant of the Early Retransmit algorithm is preferred to
  the non-SACK variant in TCP due to its robustness in the face of ACK
  loss (since SACKs are sent redundantly), and due to interactions with
  the delayed ACK timer (SCTP does not have a non-SACK mode and
  therefore naturally supports SACK-based Early Retransmit).  Consider
  a flight of three segments, S1...S3, with S2 being dropped by the
  network.  When S1 arrives, it is in order, and so the receiver may or
  may not delay the ACK, leading to two scenarios:

  (A) The ACK for S1 is delayed: In this case, the arrival of S3 will
      trigger an ACK to be transmitted, covering S1 (which was
      previously unacknowledged).  In this case, Early Retransmit
      without SACK will not prevent an RTO because no duplicate ACKs
      will arrive.  However, with SACK, the ACK for S1 will also
      include SACK information indicating that S3 has arrived at the
      receiver.  The sender can then invoke Early Retransmit on this
      ACK because only one segment remains outstanding.

  (B) The ACK for S1 is not delayed: In this case, the arrival of S1
      triggers an ACK of previously unacknowledged data.  The arrival
      of S3 triggers a duplicate ACK (because it is out of order).
      Both ACKs will cover the same segment (S1).  Therefore,
      regardless of whether SACK is used, Early Retransmit can be
      performed by the sender (assuming no ACK loss).

4.2.  Segment Reordering

  Early Retransmit is less robust in the face of reordered segments
  than when using the standard fast retransmit threshold.  Research
  shows that a general reduction in the number of duplicate ACKs
  required to trigger fast retransmit to two (rather than three) leads
  to a reduction in the ratio of good to bad retransmits by a factor of
  three [Pax97].  However, this analysis did not include the additional
  conditioning on the event that the ownd was smaller than four
  segments and that no new data was available for transmission.




Allman, et al.                Experimental                      [Page 8]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


  A number of studies have shown that network reordering is not a rare
  event across some network paths.  Various measurement studies have
  shown that reordering along most paths is negligible, but along
  certain paths can be quite prevalent [Pax97, BPS99, BS02, Pir05].
  Evaluating Early Retransmit in the face of real segment reordering is
  part of the experiment we hope to instigate with this document.

4.3.  Worst Case

  Next, we note two "worst case" scenarios for Early Retransmit:

  (1) Persistent reordering of segments coupled with an application
      that does not constantly send data can result in large numbers of
      needless retransmissions when using Early Retransmit.  For
      instance, consider an application that sends data two segments at
      a time, followed by an idle period when no data is queued for
      delivery.  If the network consistently reorders the two segments,
      the sender will needlessly retransmit one out of every two unique
      segments transmitted when using the above algorithm (meaning that
      one-third of all segments sent are needless retransmissions).
      However, this would only be a problem for long-lived connections
      from applications that transmit in spurts.

  (2) Similar to the above, consider the case of that consist of two
      segment each and always experience reordering.  Just as in (1)
      above, one out of every two unique data segments will be
      retransmitted needlessly; therefore, one-third of the traffic
      will be spurious.

  Currently, this document offers no suggestion on how to mitigate the
  above problems.  However, the worst cases are likely pathological.
  Part of the experiments that this document hopes to trigger would
  involve better understanding of whether such theoretical worst-case
  scenarios are prevalent in the network, and in general, to explore
  the trade-off between spurious fast retransmits and the delay imposed
  by the RTO.  Appendix A does offer a survey of possible mitigations
  that call for curtailing the use of Early Retransmit when it is
  making poor retransmission decisions.

5.  Related Work

  There are a number of similar proposals in the literature that
  attempt to mitigate the same problem that Early Retransmit addresses.

  Deployment of Explicit Congestion Notification (ECN) [Flo94, RFC3168]
  may benefit connections with small congestion window sizes [RFC2884].
  ECN provides a method for indicating congestion to the end-host
  without dropping segments.  While some segment drops may still occur,



Allman, et al.                Experimental                      [Page 9]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


  ECN may allow a transport to perform better with small congestion
  window sizes because the sender will be required to detect less
  segment loss [RFC2884].

  [Bal98] outlines another solution to the problem of having no new
  segments to transmit into the network when the first two duplicate
  ACKs arrive.  In response to these duplicate ACKs, a TCP sender
  transmits zero-byte segments to induce additional duplicate ACKs.
  This method preserves the robustness of the standard fast retransmit
  algorithm at the cost of injecting segments into the network that do
  not deliver any data, and therefore are potentially wasting network
  resources (at a time when there is a reasonable chance that the
  resources are scarce).

  [RFC4653] also defines an orthogonal method for altering the
  duplicate ACK threshold.  The mechanisms proposed in this document
  decrease the duplicate ACK threshold when a small amount of data is
  outstanding.  Meanwhile, the mechanisms in [RFC4653] increase the
  duplicate ACK threshold (over the standard of 3) when the congestion
  window is large in an effort to increase robustness to segment
  reordering.

6.  Security Considerations

  The security considerations found in [RFC5681] apply to this
  document.  No additional security problems have been identified with
  Early Retransmit at this time.

7.  Acknowledgments

  We thank Sally Floyd for her feedback in discussions about Early
  Retransmit.  The notion of Early Retransmit was originally sketched
  in an Internet-Draft co-authored by Sally Floyd and Hari
  Balakrishnan.  Armando Caro, Joe Touch, Alexander Zimmermann, and
  many members of the TSVWG and TCPM working groups provided good
  discussions that helped shape this document.  Our thanks to all!

8.  References

8.1.  Normative References

  [RFC793]    Postel, J., "Transmission Control Protocol", STD 7,
              RFC 793, September 1981.

  [RFC2018]   Mathis, M., Mahdavi, J., Floyd, S., and A. Romanow, "TCP
              Selective Acknowledgment Options", RFC 2018,
              October 1996.




Allman, et al.                Experimental                     [Page 10]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


  [RFC2119]   Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119, March 1997.

  [RFC2883]   Floyd, S., Mahdavi, J., Mathis, M., and M. Podolsky, "An
              Extension to the Selective Acknowledgement (SACK) Option
              for TCP", RFC 2883, July 2000.

  [RFC2988]   Paxson, V. and M. Allman, "Computing TCP's Retransmission
              Timer", RFC 2988, November 2000.

  [RFC3042]   Allman, M., Balakrishnan, H., and S. Floyd, "Enhancing
              TCP's Loss Recovery Using Limited Transmit", RFC 3042,
              January 2001.

  [RFC4960]   Stewart, R., Ed., "Stream Control Transmission Protocol",
              RFC 4960, September 2007.

  [RFC5681]   Allman, M., Paxson, V., and E. Blanton, "TCP Congestion
              Control", RFC 5681, September 2009.

8.2.  Informative References

  [AA02]      Urtzi Ayesta, Konstantin Avrachenkov, "The Effect of the
              Initial Window Size and Limited Transmit Algorithm on the
              Transient Behavior of TCP Transfers", In Proc. of the
              15th ITC Internet Specialist Seminar, Wurzburg,
              July 2002.

  [All00]     Mark Allman.  A Web Server's View of the Transport Layer.
              ACM Computer Communication Review, October 2000.

  [Bal98]     Hari Balakrishnan.  Challenges to Reliable Data Transport
              over Heterogeneous Wireless Networks.  Ph.D. Thesis,
              University of California at Berkeley, August 1998.

  [BPS+98]    Hari Balakrishnan, Venkata Padmanabhan,
              Srinivasan Seshan, Mark Stemm, and Randy Katz.  TCP
              Behavior of a Busy Web Server: Analysis and Improvements.
              Proc. IEEE INFOCOM Conf., San Francisco, CA, March 1998.

  [BPS99]     Jon Bennett, Craig Partridge, Nicholas Shectman.  Packet
              Reordering is Not Pathological Network Behavior.
              IEEE/ACM Transactions on Networking, December 1999.

  [BS02]      John Bellardo, Stefan Savage.  Measuring Packet
              Reordering, ACM/USENIX Internet Measurement Workshop,
              November 2002.




Allman, et al.                Experimental                     [Page 11]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


  [FF96]      Kevin Fall, Sally Floyd.  Simulation-based Comparisons of
              Tahoe, Reno, and SACK TCP.  ACM Computer Communication
              Review, July 1996.

  [Flo94]     Sally Floyd.  TCP and Explicit Congestion Notification.
              ACM Computer Communication Review, October 1994.

  [HB08]      Per Hurtig, Anna Brunstrom.  Enhancing SCTP Loss
              Recovery: An Experimental Evaluation of Early Retransmit.
              Elsevier Computer Communications, Vol. 31(16),
              October 2008, pp. 3778-3788.

  [Jac88]     Van Jacobson.  Congestion Avoidance and Control.  ACM
              SIGCOMM 1988.

  [LK98]      Dong Lin, H.T. Kung.  TCP Fast Recovery Strategies:
              Analysis and Improvements.  Proc. IEEE INFOCOM Conf.,
              San Francisco, CA, March 1998.

  [Mor97]     Robert Morris.  TCP Behavior with Many Flows.  Proc.
              Fifth IEEE International Conference on Network Protocols,
              October 1997.

  [Pax97]     Vern Paxson.  End-to-End Internet Packet Dynamics.  ACM
              SIGCOMM, September 1997.

  [Pir05]     N. M. Piratla, "A Theoretical Foundation, Metrics and
              Modeling of Packet Reordering and Methodology of Delay
              Modeling using Inter-packet Gaps," Ph.D. Dissertation,
              Department of Electrical and Computer Engineering,
              Colorado State University, Fort Collins, CO, Fall 2005.

  [RFC2884]   Hadi Salim, J. and U. Ahmed, "Performance Evaluation of
              Explicit Congestion Notification (ECN) in IP Networks",
              RFC 2884, July 2000.

  [RFC3150]   Dawkins, S., Montenegro, G., Kojo, M., and V. Magret,
              "End-to-end Performance Implications of Slow Links",
              BCP 48, RFC 3150, July 2001.

  [RFC3168]   Ramakrishnan, K., Floyd, S., and D. Black, "The Addition
              of Explicit Congestion Notification (ECN) to IP",
              RFC 3168, September 2001.

  [RFC3517]   Blanton, E., Allman, M., Fall, K., and L. Wang, "A
              Conservative Selective Acknowledgment (SACK)-based Loss
              Recovery Algorithm for TCP", RFC 3517, April 2003.




Allman, et al.                Experimental                     [Page 12]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


  [RFC3522]   Ludwig, R. and M. Meyer, "The Eifel Detection Algorithm
              for TCP", RFC 3522, April 2003.

  [RFC3782]   Floyd, S., Henderson, T., and A. Gurtov, "The NewReno
              Modification to TCP's Fast Recovery Algorithm", RFC 3782,
              April 2004.

  [RFC4653]   Bhandarkar, S., Reddy, A., Allman, M., and E. Blanton,
              "Improving the Robustness of TCP to Non-Congestion
              Events", RFC 4653, August 2006.









































Allman, et al.                Experimental                     [Page 13]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


Appendix A.  Research Issues in Adjusting the Duplicate ACK Threshold

  Decreasing the number of duplicate ACKs required to trigger fast
  retransmit, as suggested in Section 3, has the drawback of making
  fast retransmit less robust in the face of minor network reordering.
  Two egregious examples of problems caused by reordering are given in
  Section 4.  This appendix outlines several schemes that have been
  suggested to mitigate the problems caused by Early Retransmit in the
  face of segment reordering.  These methods need further research
  before they are suggested for general use (and current consensus is
  that the cases that make Early Retransmit unnecessarily retransmit a
  large amount of data are pathological, and therefore, these
  mitigations are not generally required).

  MITIGATION A.1: Allow a connection to use Early Retransmit as long as
     the algorithm is not injecting "too much" spurious data into the
     network.  For instance, using the information provided by TCP's
     D-SACK option [RFC2883] or SCTP's Duplicate Transmission Sequence
     Number (Duplicate-TSN) notification, a sender can determine when
     segments sent via Early Retransmit are needless.  Likewise, using
     Eifel [RFC3522], the sender can detect spurious Early Retransmits.
     Once spurious Early Retransmits are detected, the sender can
     either eliminate the use of Early Retransmit, or limit the use of
     the algorithm to ensure that an acceptably small fraction of the
     connection's transmissions are not spurious.  For example, a
     connection could stop using Early Retransmit after the first
     spurious retransmit is detected.

  MITIGATION A.2: If a sender cannot reliably determine whether an
     Early-Retransmitted segment is spurious or not, the sender could
     simply limit Early Retransmits, either to some fixed number per
     connection (e.g., Early Retransmit is allowed only once per
     connection), or to some small percentage of the total traffic
     being transmitted.

  MITIGATION A.3: Allow a connection to trigger Early Retransmit using
     the criteria given in Section 3, in addition to a "small" timeout
     [Pax97].  For instance, a sender may have to wait for two
     duplicate ACKs and then T msec before Early Retransmit is invoked.
     The added time gives reordered acknowledgments time to arrive at
     the sender and avoid a needless retransmit.  Designing a method
     for choosing an appropriate timeout is part of the research that
     would need to be involved in this scheme.








Allman, et al.                Experimental                     [Page 14]

RFC 5827            Early Retransmit for TCP and SCTP         April 2010


Authors' Addresses

  Mark Allman
  International Computer Science Institute
  1947 Center Street, Suite 600
  Berkeley, CA 94704-1198
  USA
  Phone: 440-235-1792
  EMail: [email protected]
  http://www.icir.org/mallman/


  Konstantin Avrachenkov
  INRIA
  2004 route des Lucioles, B.P.93
  06902, Sophia Antipolis
  France
  Phone: 00 33 492 38 7751
  EMail: [email protected]
  http://www-sop.inria.fr/members/Konstantin.Avratchenkov/me.html


  Urtzi Ayesta
  BCAM-IKERBASQUE                         LAAS-CNRS
  Bizkaia Technology Park, Building 500   7 Avenue Colonel Roche
  48160 Derio                             31077, Toulouse
  Spain                                   France
                                          EMail: [email protected]
                                          http://www.laas.fr/~urtzi


  Josh Blanton
  Ohio University
  301 Stocker Center
  Athens, OH  45701
  USA
  EMail: [email protected]


  Per Hurtig
  Karlstad University
  Department of Computer Science
  Universitetsgatan 2 651 88
  Karlstad
  Sweden
  EMail: [email protected]





Allman, et al.                Experimental                     [Page 15]


You are viewing proxied material from gopher.fnord.one. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.