RFC 3042






Network Working Group                                          M. Allman
Request for Comments: 3042                                  NASA GRC/BBN
Category: Standards Track                                H. Balakrishnan
                                                                     MIT
                                                                S. Floyd
                                                                   ACIRI
                                                            January 2001


          Enhancing TCP's Loss Recovery Using Limited Transmit

Status of this Memo



   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

Copyright Notice



   Copyright (C) The Internet Society (2001).  All Rights Reserved.

Abstract



   This document proposes a new Transmission Control Protocol (TCP)
   mechanism that can be used to more effectively recover lost segments
   when a connection's congestion window is small, or when a large
   number of segments are lost in a single transmission window.  The
   "Limited Transmit" algorithm calls for sending a new data segment in
   response to each of the first two duplicate acknowledgments that
   arrive at the sender.  Transmitting these segments increases the
   probability that TCP can recover from a single lost segment using the
   fast retransmit algorithm, rather than using a costly retransmission
   timeout.  Limited Transmit can be used both in conjunction with, and
   in the absence of, the TCP selective acknowledgment (SACK) mechanism.

1   Introduction



   A number of researchers have observed that TCP's loss recovery
   strategies do not work well when the congestion window at a TCP
   sender is small.  This can happen, for instance, because there is
   only a limited amount of data to send, or because of the limit
   imposed by the receiver-advertised window, or because of the
   constraints imposed by end-to-end congestion control over a
   connection with a small bandwidth-delay product
   [Riz96,Mor97,BPS+98,Bal98,LK98].  When a TCP detects a missing
   segment, it enters a loss recovery phase using one of two methods.



Allman, et al.              Standards Track                     [Page 1]

RFC 3042              Enhancing TCP Loss Recovery           January 2001


   First, if an acknowledgment (ACK) for a given segment is not received
   in a certain amount of time a retransmission timeout occurs and the
   segment is resent [RFC793,PA00].  Second, the "Fast Retransmit"
   algorithm resends a segment when three duplicate ACKs arrive at the
   sender [Jac88,RFC2581].  However, because duplicate ACKs from the
   receiver are also triggered by packet reordering in the Internet, the
   TCP sender waits for three duplicate ACKs in an attempt to
   disambiguate segment loss from packet reordering.  Once in a loss
   recovery phase, a number of techniques can be used to retransmit lost
   segments, including slow start-based recovery or Fast Recovery
   [RFC2581], NewReno [RFC2582], and loss recovery based on selective
   acknowledgments (SACKs) [RFC2018,FF96].

   TCP's retransmission timeout (RTO) is based on measured round-trip
   times (RTT) between the sender and receiver, as specified in [PA00].
   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
   when 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
   or if a large number of segments in a window are lost.  For instance,
   consider a congestion window (cwnd) of three segments.  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 packet.

   [BPS+97] found that roughly 56% of retransmissions sent by a busy web
   server were sent after the RTO expires, while only 44% were handled
   by Fast Retransmit.  In addition, only 4% of the RTO-based
   retransmissions could have been avoided with SACK, which of course
   has to continue to disambiguate reordering from genuine loss.  In
   contrast, using the technique outlined in this document and in
   [Bal98], 25% of the RTO-based retransmissions in that dataset would
   have likely been avoided.

   The next section of this document outlines small changes to TCP
   senders that will decrease the reliance on the retransmission timer,
   and thereby improve TCP performance when Fast Retransmit is not
   triggered.  These changes do not adversely affect the performance of
   TCP nor interact adversely with other connections, in other
   circumstances.







Allman, et al.              Standards Track                     [Page 2]

RFC 3042              Enhancing TCP Loss Recovery           January 2001


1.1 Terminology



   In this document, he key words "MUST", "MUST NOT", "REQUIRED",
   "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY",
   AND "OPTIONAL" are to be interpreted as described in RFC 2119 [1] and
   indicate requirement levels for protocols.

2   The Limited Transmit Algorithm



   When a TCP sender has previously unsent data queued for transmission
   it SHOULD use the Limited Transmit algorithm, which calls for a TCP
   sender to transmit new data upon the arrival of the first two
   consecutive duplicate ACKs when the following conditions are
   satisfied:

     * The receiver's advertised window allows the transmission of the
       segment.

     * The amount of outstanding data would remain less than or equal
       to the congestion window plus 2 segments.  In other words, the
       sender can only send two segments beyond the congestion window
       (cwnd).

   The congestion window (cwnd) MUST NOT be changed when these new
   segments are transmitted.  Assuming that these new segments and the
   corresponding ACKs are not dropped, this procedure allows the sender
   to infer loss using the standard Fast Retransmit threshold of three
   duplicate ACKs [RFC2581].  This is more robust to reordered packets
   than if an old packet were retransmitted on the first or second
   duplicate ACK.

   Note: If the connection is using selective acknowledgments [RFC2018],
   the data sender MUST NOT send new segments in response to duplicate
   ACKs that contain no new SACK information, as a misbehaving receiver
   can generate such ACKs to trigger inappropriate transmission of data
   segments.  See [SCWA99] for a discussion of attacks by misbehaving
   receivers.

   Limited Transmit follows the "conservation of packets" congestion
   control principle [Jac88].  Each of the first two duplicate ACKs
   indicate that a segment has left the network.  Furthermore, the
   sender has not yet decided that a segment has been dropped and
   therefore has no reason to assume that the current congestion control
   state is inaccurate.  Therefore, transmitting segments does not
   deviate from the spirit of TCP's congestion control principles.






Allman, et al.              Standards Track                     [Page 3]

RFC 3042              Enhancing TCP Loss Recovery           January 2001


   [BPS99] shows that packet reordering is not a rare network event.
   [RFC2581] does not provide for sending of data on the first two
   duplicate ACKs that arrive at the sender.  This causes a burst of
   segments to be sent when an ACK for new data does arrive following
   packet reordering.  Using Limited Transmit, data packets will be
   clocked out by incoming ACKs and therefore transmission will not be
   as bursty.

   Note: Limited Transmit is implemented in the ns simulator [NS].
   Researchers wishing to investigate this mechanism further can do so
   by enabling "singledup_" for the given TCP connection.

3   Related Work



   Deployment of Explicit Congestion Notification (ECN) [Flo94,RFC2481]
   may benefit connections with small congestion window sizes [SA00].
   ECN provides a method for indicating congestion to the end-host
   without dropping segments.  While some segment drops may still occur,
   ECN may allow TCP to perform better with small congestion window
   sizes because the sender can avoid many of the Fast Retransmits and
   Retransmit Timeouts that would otherwise have been needed to detect
   dropped segments [SA00].

   When ECN-enabled TCP traffic competes with non-ECN-enabled TCP
   traffic, ECN-enabled traffic can receive up to 30% higher goodput.
   For bulk transfers, the relative performance benefit of ECN is
   greatest when on average each flow has 3-4 outstanding packets during
   each round-trip time [ZQ00].  This should be a good estimate for the
   performance impact of a flow using Limited Transmit, since both ECN
   and Limited Transmit reduce the reliance on the retransmission timer
   for signaling congestion.

   The Rate-Halving congestion control algorithm [MSML99] uses a form of
   limited transmit, as it calls for transmitting a data segment on
   every second duplicate ACK that arrives at the sender.  The algorithm
   decouples the decision of what to send from the decision of when to
   send.  However, similar to Limited Transmit the algorithm will always
   send a new data segment on the second duplicate ACK that arrives at
   the sender.

4   Security Considerations



   The additional security implications of the changes proposed in this
   document, compared to TCP's current vulnerabilities, are minimal.
   The potential security issues come from the subversion of end-to-end
   congestion control from "false" duplicate ACKs, where a "false"
   duplicate ACK is a duplicate ACK that does not actually acknowledge
   new data received at the TCP receiver.  False duplicate ACKs could



Allman, et al.              Standards Track                     [Page 4]

RFC 3042              Enhancing TCP Loss Recovery           January 2001


   result from duplicate ACKs that are themselves duplicated in the
   network, or from misbehaving TCP receivers that send false duplicate
   ACKs to subvert end-to-end congestion control [SCWA99,RFC2581].

   When the TCP data receiver has agreed to use the SACK option, the TCP
   data sender has fairly strong protection against false duplicate
   ACKs.  In particular, with SACK, a duplicate ACK that acknowledges
   new data arriving at the receiver reports the sequence numbers of
   that new data.  Thus, with SACK, the TCP sender can verify that an
   arriving duplicate ACK acknowledges data that the TCP sender has
   actually sent, and for which no previous acknowledgment has been
   received, before sending new data as a result of that acknowledgment.
   For further protection, the TCP sender could keep a record of packet
   boundaries for transmitted data packets, and recognize at most one
   valid acknowledgment for each packet (e.g., the first acknowledgment
   acknowledging the receipt of all of the sequence numbers in that
   packet).

   One could imagine some limited protection against false duplicate
   ACKs for a non-SACK TCP connection, where the TCP sender keeps a
   record of the number of packets transmitted, and recognizes at most
   one acknowledgment per packet to be used for triggering the sending
   of new data.  However, this accounting of packets transmitted and
   acknowledged would require additional state and extra complexity at
   the TCP sender, and does not seem necessary.

   The most important protection against false duplicate ACKs comes from
   the limited potential of duplicate ACKs in subverting end-to-end
   congestion control.  There are two separate cases to consider: when
   the TCP sender receives less than a threshold number of duplicate
   ACKs, and when the TCP sender receives at least a threshold number of
   duplicate ACKs.  In the latter case a TCP with Limited Transmit will
   behave essentially the same as a TCP without Limited Transmit in that
   the congestion window will be halved and a loss recovery period will
   be initiated.

   When a TCP sender receives less than a threshold number of duplicate
   ACKs a misbehaving receiver could send two duplicate ACKs after each
   regular ACK.  One might imagine that the TCP sender would send at
   three times its allowed sending rate.  However, using Limited
   Transmit as outlined in section 2 the sender is only allowed to
   exceed the congestion window by less than the duplicate ACK threshold
   (of three segments), and thus would not send a new packet for each
   duplicate ACK received.







Allman, et al.              Standards Track                     [Page 5]

RFC 3042              Enhancing TCP Loss Recovery           January 2001


Acknowledgments

   Bill Fenner, Jamshid Mahdavi and the Transport Area Working Group
   provided valuable feedback on an early version of this document.

References

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

   [BPS+97]  Hari Balakrishnan, Venkata Padmanabhan, Srinivasan Seshan,
             Mark Stemm, and Randy Katz.  TCP Behavior of a Busy Web
             Server:  Analysis and Improvements.  Technical Report
             UCB/CSD-97-966, August 1997.  Available from
             http://nms.lcs.mit.edu/~hari/papers/csd-97-966.ps.  (Also
             in 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.

   [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.

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

   [LK98]    Dong Lin, H.T. Kung.  TCP Fast Recovery Strategies:
             Analysis and Improvements.  Proceedings of InfoCom, March
             1998.

   [MSML99]  Matt Mathis, Jeff Semke, Jamshid Mahdavi, Kevin Lahey.  The
             Rate Halving Algorithm, 1999. URL:
             http://www.psc.edu/networking/rate_halving.html.

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

   [NS]      Ns network simulator.  URL: http://www.isi.edu/nsnam/.





Allman, et al.              Standards Track                     [Page 6]

RFC 3042              Enhancing TCP Loss Recovery           January 2001


   [PA00]    Paxson, V. and M. Allman, "Computing TCP's Retransmission
             Timer", RFC 2988, November 2000.

   [Riz96]   Luigi Rizzo.  Issues in the Implementation of Selective
             Acknowledgments for TCP.  January, 1996.  URL:
             http://www.iet.unipi.it/~luigi/selack.ps

   [SA00]    Hadi Salim, J. and U. Ahmed, "Performance Evaluation of
             Explicit Congestion Notification (ECN) in IP Networks", RFC
             2884, July 2000.

   [SCWA99]  Stefan Savage, Neal Cardwell, David Wetherall, Tom
             Anderson.  TCP Congestion Control with a Misbehaving
             Receiver.  ACM Computer Communications Review, October
             1999.

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

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

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

   [RFC2481] Ramakrishnan, K. and S. Floyd, "A Proposal to Add Explicit
             Congestion Notification (ECN) to IP", RFC 2481, January
             1999.

   [RFC2581] Allman, M., Paxson, V. and W. Stevens, "TCP Congestion
             Control", RFC 2581, April 1999.

   [RFC2582] Floyd, S. and T. Henderson, "The NewReno Modification to
             TCP's Fast Recovery Algorithm", RFC 2582, April 1999.

   [ZQ00]    Yin Zhang and Lili Qiu, Understanding the End-to-End
             Performance Impact of RED in a Heterogeneous Environment,
             Cornell CS Technical Report 2000-1802, July 2000.  URL
             http://www.cs.cornell.edu/yzhang/papers.htm.












Allman, et al.              Standards Track                     [Page 7]

RFC 3042              Enhancing TCP Loss Recovery           January 2001


Authors' Addresses



   Mark Allman
   NASA Glenn Research Center/BBN Technologies
   Lewis Field
   21000 Brookpark Rd.  MS 54-5
   Cleveland, OH  44135

   Phone: +1-216-433-6586
   Fax:   +1-216-433-8705
   EMail: mallman@grc.nasa.gov
   http://roland.grc.nasa.gov/~mallman


   Hari Balakrishnan
   Laboratory for Computer Science
   545 Technology Square
   Massachusetts Institute of Technology
   Cambridge, MA 02139

   EMail: hari@lcs.mit.edu
   http://nms.lcs.mit.edu/~hari/


   Sally Floyd
   AT&T Center for Internet Research at ICSI (ACIRI)
   1947 Center St, Suite 600
   Berkeley, CA 94704

   Phone: +1-510-666-2989
   EMail: floyd@aciri.org
   http://www.aciri.org/floyd/



















Allman, et al.              Standards Track                     [Page 8]

RFC 3042              Enhancing TCP Loss Recovery           January 2001


Full Copyright Statement



   Copyright (C) The Internet Society (2001).  All Rights Reserved.

   This document and translations of it may be copied and furnished to
   others, and derivative works that comment on or otherwise explain it
   or assist in its implementation may be prepared, copied, published
   and distributed, in whole or in part, without restriction of any
   kind, provided that the above copyright notice and this paragraph are
   included on all such copies and derivative works.  However, this
   document itself may not be modified in any way, such as by removing
   the copyright notice or references to the Internet Society or other
   Internet organizations, except as needed for the purpose of
   developing Internet standards in which case the procedures for
   copyrights defined in the Internet Standards process must be
   followed, or as required to translate it into languages other than
   English.

   The limited permissions granted above are perpetual and will not be
   revoked by the Internet Society or its successors or assigns.

   This document and the information contained herein is provided on an
   "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
   TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
   BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
   HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Acknowledgement



   Funding for the RFC Editor function is currently provided by the
   Internet Society.



















Allman, et al.              Standards Track                     [Page 9]