Internet Engineering Task Force S. Floyd INTERNET DRAFT ICSI draft-ietf-tsvwg-newreno-00.txt T. Henderson Boeing June 2003 The NewReno Modification to TCP's Fast Recovery Algorithm Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. 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." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. Abstract RFC 2581 [RFC2581] documents the following four intertwined TCP congestion control algorithms: Slow Start, Congestion Avoidance, Fast Retransmit, and Fast Recovery. RFC 2581 [RFC2581] explicitly allows certain modifications of these algorithms, including modifications that use the TCP Selective Acknowledgement (SACK) option [RFC2018], and modifications that respond to "partial acknowledgments" (ACKs which cover new data, but not all the data outstanding when loss was detected) in the absence of SACK. The NewReno mechanism described in this document describes a specific algorithm for responding to partial acknowledgments, referred to as NewReno. This response to partial acknowledgments was first proposed by Janey Hoe in [Hoe95]. RFC 2582 [RFC2582] specified the NewReno mechanisms as Experimental Floyd & Henderson [Page 1] draft-ietf-tsvwg-newreno June 2003 in 1999. This document is a small revision of RFC 2582 intended to advance the NewReno mechanisms to Proposed Standard. RFC 2581 notes that the Fast Retransmit/Fast Recovery algorithm specified in that document does not recover very efficiently from multiple losses in a single flight of packets, and that RFC 2582 contains one set of modifications to address this problem. NOTE TO THE RFC EDITOR: PLEASE REMOVE THIS SECTION UPON PUBLICATION. Changes from draft-floyd-newreno-00.txt: * In Section 8 on "Implementation issues for the data sender", mentioned alternate methods for limiting bursts when exiting Fast Recovery. * Changed draft from draft-floyd-newreno to draft-ietf-tsvwg-newreno Changes from RFC 2582: * Rephrasing and rearrangements of the text. * RFC 2582 described the Careful and Less Careful variants of NewReno, along with a default version that was neither Careful nor Less Careful, and recommended the Careful variant. This document only specifies the Careful version. * RFC 2582 used two separate variables, "send_high" and "recover", and this document has merged them into a single variable "recover". * Added sections on "Comparisons between Reno and NewReno TCP", and on "Changes relative to RFC 2582". The section on "Comparisons between Reno and NewReno TCP" includes a discussion of the one area where NewReno is known to perform worse than Reno or SACK, and that is in the response to reordering. * Moved all of the discussions of the Impatient and Slow-but-Steady variants to one place, and specified the Impatient variant (as in the default version in RFC 2582). * Added a section on Implementation issues for the data sender, mentioning maxburst_. * Added a paragraph about differences between RFC 2582 and [FF96]. END OF NOTE TO RFC EDITOR Floyd & Henderson [Page 2] draft-ietf-tsvwg-newreno June 2003 1. Introduction For the typical implementation of the TCP Fast Recovery algorithm described in [RFC2581] (first implemented in the 1990 BSD Reno release, and referred to as the Reno algorithm in [FF96]), the TCP data sender only retransmits a packet after a retransmit timeout has occurred, or after three duplicate acknowledgements have arrived triggering the Fast Retransmit algorithm. A single retransmit timeout might result in the retransmission of several data packets, but each invocation of the Fast Retransmit algorithm in RFC 2581 leads to the retransmission of only a single data packet. Problems can arise, therefore, when multiple packets have been dropped from a single window of data and the Fast Retransmit and Fast Recovery algorithms are invoked. In this case, if the SACK option is available, the TCP sender has the information to make intelligent decisions about which packets to retransmit and which packets not to retransmit during Fast Recovery. This document applies only for TCP connections that are unable to use the TCP Selective Acknowledgement (SACK) option, either because the option is not locally supported or because the TCP peer did not indicate a willingness to use SACK. In the absence of SACK, there is little information available to the TCP sender in making retransmission decisions during Fast Recovery. From the three duplicate acknowledgements, the sender infers a packet loss, and retransmits the indicated packet. After this, the data sender could receive additional duplicate acknowledgements, as the data receiver acknowledges additional data packets that were already in flight when the sender entered Fast Retransmit. In the case of multiple packets dropped from a single window of data, the first new information available to the sender comes when the sender receives an acknowledgement for the retransmitted packet (that is, the packet retransmitted when Fast Retransmit was first entered). If there had been a single packet drop and no reordering, then the acknowledgement for this packet will acknowledge all of the packets transmitted before Fast Retransmit was entered. However, when there were multiple packet drops, then the acknowledgement for the retransmitted packet will acknowledge some but not all of the packets transmitted before the Fast Retransmit. We call this acknowledgement a partial acknowledgment. Along with several other suggestions, [Hoe95] suggested that during Fast Recovery the TCP data sender respond to a partial acknowledgment by inferring that the next in-sequence packet has been lost, and retransmitting that packet. This document describes a modification to the Fast Recovery algorithm in RFC 2581 that incorporates a response to partial acknowledgements received during Fast Recovery. Floyd & Henderson [Page 3] draft-ietf-tsvwg-newreno June 2003 We call this modified Fast Recovery algorithm NewReno, because it is a slight but significant variation of the basic Reno algorithm in RFC 2581. This document does not discuss the other suggestions in [Hoe95] and [Hoe96], such as a change to the ssthresh parameter during Slow-Start, or the proposal to send a new packet for every two duplicate acknowledgements during Fast Recovery. The version of NewReno in this document also draws on other discussions of NewReno in the literature [LM97]. We do not claim that the NewReno version of Fast Recovery described here is an optimal modification of Fast Recovery for responding to partial acknowledgements, for TCP connections that are unable to use SACK. Based on our experiences with the NewReno modification in the NS simulator [NS] and with numerous implementations of NewReno, we believe that this modification improves the performance of the Fast Retransmit and Fast Recovery algorithms in a wide variety of scenarios. 2. Terminology and Definitions In this document, the key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" are to be interpreted as described in BCP 14, RFC 2119 and indicate requirement levels for compliant TCP implementations implementing the NewReno Fast Retransmit and Fast Recovery algorithms described in this document. This document assumes that the reader is familiar with the terms SENDER MAXIMUM SEGMENT SIZE (SMSS), CONGESTION WINDOW (cwnd), and FLIGHT SIZE (FlightSize) defined in [RFC2581]. FLIGHT SIZE is defined as in [RFC2581] as follows: FLIGHT SIZE: The amount of data that has been sent but not yet acknowledged. 3. The Fast Retransmit and Fast Recovery algorithms in NewReno The standard implementation of the Fast Retransmit and Fast Recovery algorithms is given in [RFC2581]. The NewReno modification of these algorithms is given below. The NewReno modification concerns the Fast Recovery procedure that begins when three duplicate ACKs are received and ends when either a retransmission timeout occurs or an ACK arrives that acknowledges all of the data up to and including the data that was outstanding when the Fast Recovery procedure began. The NewReno algorithm specified in this document differs from the implementation in [RFC2581] in the introduction of the variable "recover" in step 1, in the response to a partial or new Floyd & Henderson [Page 4] draft-ietf-tsvwg-newreno June 2003 acknowledgement in step 5, and in modifications to step 1 and the addition of step 6 for avoiding multiple Fast Retransmits caused by the retransmission of packets already received by the receiver. The algorithm specified in this document uses a variable "recover", whose initial value is the initial send sequence number. 1) When the third duplicate ACK is received and the sender is not already in the Fast Recovery procedure, check to see if the Cumulative Acknowledgement field covers more than "recover". If so, then set ssthresh to no more than the value given in equation 1 below. (This is equation 3 from [RFC2581]). ssthresh = max (FlightSize / 2, 2*SMSS) (1) In addition, record the highest sequence number transmitted in the variable "recover", and go to Step 2. If the Cumulative Acknowledgement field didn't cover more than "recover", then do not enter the Fast Retransmit and Fast Recovery procedure. In particular, do not change ssthresh, do not go to Step 2 to retransmit the "lost" segment, and do not execute Step 3 upon subsequent duplicate ACKs. 2) Retransmit the lost segment and set cwnd to ssthresh plus 3*SMSS. This artificially "inflates" the congestion window by the number of segments (three) that have left the network and which the receiver has buffered. 3) For each additional duplicate ACK received, increment cwnd by SMSS. This artificially inflates the congestion window in order to reflect the additional segment that has left the network. 4) Transmit a segment, if allowed by the new value of cwnd and the receiver's advertised window. 5) When an ACK arrives that acknowledges new data, this ACK could be the acknowledgment elicited by the retransmission from step 2, or elicited by a later retransmission. If this ACK acknowledges all of the data up to and including "recover", then the ACK acknowledges all the intermediate segments sent between the original transmission of the lost segment and the receipt of the third duplicate ACK. Set cwnd to either (1) min (ssthresh, FlightSize + SMSS); or (2) ssthresh, where ssthresh is the value set in step 1; this is termed "deflating" the window. (We note that "FlightSize" in step 1 Floyd & Henderson [Page 5] draft-ietf-tsvwg-newreno June 2003 referred to the amount of data outstanding in step 1, when Fast Recovery was entered, while "FlightSize" in step 5 refers to the amount of data outstanding in step 5, when Fast Recovery is exited.) If the second option is selected, the implementation should take measures to avoid a possible burst of data, in case the amount of data outstanding in the network was much less than the new congestion window allows. A simple mechanism is to limit the number of data packets that can be sent in response to a single acknowledgement. (This is known as "maxburst_" in the NS simulator). Exit the Fast Recovery procedure. If this ACK does *not* acknowledge all of the data up to and including "recover", then this is a partial ACK. In this case, retransmit the first unacknowledged segment. Deflate the congestion window by the amount of new data acknowledged, then add back one SMSS (if the partial ACK acknowledges at least one SMSS of new data) and send a new segment if permitted by the new value of cwnd. This "partial window deflation" attempts to ensure that, when Fast Recovery eventually ends, approximately ssthresh amount of data will be outstanding in the network. Do not exit the Fast Recovery procedure (i.e., if any duplicate ACKs subsequently arrive, execute Steps 3 and 4 above). For the first partial ACK that arrives during Fast Recovery, also reset the retransmit timer. 6) After a retransmit timeout, record the highest sequence number transmitted in the variable "recover" and exit the Fast Recovery procedure if applicable. Step 1 specifies a check that the Cumulative Acknowledgement field covers more than "recover". Because the acknowledgement field contains the sequence number that the sender next expects to receive, the acknowledgement "ack_number" covers more than "recover" when: ack_number - one > recover. Note that in Step 5, the congestion window is deflated after a partial acknowledgement is received. The congestion window was likely to have been inflated considerably when the partial acknowledgement was received. In addition, depending on the original pattern of packet losses, the partial acknowledgement might acknowledge nearly a window of data. In this case, if the congestion window was not deflated, the data sender might be able to send nearly a window of data back-to-back. This document does not specify the sender's response to duplicate ACKs when the Fast Retransmit/Fast Recovery algorithm is not invoked. Floyd & Henderson [Page 6] draft-ietf-tsvwg-newreno June 2003 This is addressed in other documents, such as those describing the Limited Transmit procedure [RFC3042]. This document also does not address issues of adjusting the duplicate acknowledgement threshold, but assumes the threshold of three duplicate acknowledgements currently specified in RFC 2581. As a final note, we would observe that in the absence of the SACK option, the data sender is working from limited information. When the issue of recovery from multiple dropped packets from a single window of data is of particular importance, the best alternative would be to use the SACK option. 4. Resetting the retransmit timer in response to partial acknowledgements. One possible variant to the response to partial acknowledgements specified in Section 3 concerns when to reset the retransmit timer after a partial acknowledgement. The algorithm in Section 3, Step 5, resets the retransmit timer only after the first partial ACK. In this case, if a large number of packets were dropped from a window of data, the TCP data sender's retransmit timer will ultimately expire, and the TCP data sender will invoke Slow-Start. (This is illustrated on page 12 of [F98].) We call this the Impatient variant of NewReno. In contrast, the NewReno simulations in [FF96] illustrate the algorithm described above with the modification that the retransmit timer is reset after each partial acknowledgement. We call this the Slow-but-Steady variant of NewReno. In this case, for a window with a large number of packet drops, the TCP data sender retransmits at most one packet per roundtrip time. (This behavior is illustrated in the New-Reno TCP simulation of Figure 5 in [FF96], and on page 11 of [F98]. The tests "../../ns test-suite-newreno.tcl newreno1_B0" and "../../ns test-suite-newreno.tcl newreno1_B" in the NS simulator also illustrate the Slow-but-Steady and the Impatient variants of NewReno, respectively.) When N packets have been dropped from a window of data for a large value of N, the Slow-but-Steady variant can remain in Fast Recovery for N round-trip times, retransmitting one more dropped packet each round-trip time; for these scenarios, the Impatient variant gives a faster recovery and better performance. One can also construct scenarios where the Slow-but-Steady variant would give better performance, where only a small number of packets are dropped, the RTO is sufficiently small that the retransmit timer expires, and performance would have been better without a retransmit timeout. Thus, neither of these variants are optimal; our recommendation is for the Impatient variant, as specified in Section 3 of this document. Floyd & Henderson [Page 7] draft-ietf-tsvwg-newreno June 2003 One possibility for a more optimal algorithm would be one that recovered from multiple packet drops as quickly as does slow-start, while resetting the retransmit timers after each partial acknowledgement, as described in the section below. We note, however, that there is a limitation to the potential performance in this case in the absence of the SACK option. 5. Retransmissions after a partial acknowledgement. One possible variant to the response to partial acknowledgements specified in Section 3 would be to retransmit more than one packet after each partial acknowledgement, and to reset the retransmit timer after each retransmission. The algorithm specified in Section 3 retransmits a single packet after each partial acknowledgement. This is the most conservative alternative, in that it is the least likely to result in an unnecessarily-retransmitted packet. A variant that would recover faster from a window with many packet drops would be to effectively Slow-Start, retransmitting two packets after each partial acknowledgement. Such an approach would take less than N roundtrip times to recover from N losses [Hoe96]. However, in the absence of SACK, recovering as quickly as slow-start introduces the likelihood of unnecessarily retransmitting packets, and this could significantly complicate the recovery mechanisms. We note that the response to partial acknowledgements specified in Section 3 of this document and in RFC 2582 differs from the response in [FF96], even though both approaches only retransmit one packet in response to a partial acknowledgement. Step 5 of Section 3 specifies that the TCP sender responds to a partial ACK by deflating the congestion window by the amount of new data acknowledged, then adding back one SMSS if the partial ACK acknowledges at least one SMSS of new data, and sending a new segment if permitted by the new value of cwnd. Thus, only one previously-sent packet is retransmitted in response to each partial acknowledgement, but additional new packets might be transmitted as well, depending on the amount of new data acknowledged by the partial acknowledgement. In contrast, the variant of NewReno illustrated in [FF96] simply set the congestion window to ssthresh when a partial acknowledgement was received. The approach in [FF96] is more conservative, and does not attempt to accurately track the actual number of outstanding packets after a partial acknowledgement is received. While either of these approaches gives acceptable performance, the variant specified in Section 3 recovers more smoothly when multiple packets are dropped from a window of data. (The [FF96] behavior can be seen in the NS simulator by setting the variable "partial_window_deflation_" for "Agent/TCP/Newreno" to 0, and the behavior specified in Section 3 is achieved by setting "partial_window_deflation_" to 1.) Floyd & Henderson [Page 8] draft-ietf-tsvwg-newreno June 2003 6. Avoiding Multiple Fast Retransmits This section describes the motivation for the sender's state variable "recover". In the absence of the SACK option, a duplicate acknowledgement carries no information to identify the data packet or packets at the TCP data receiver that triggered that duplicate acknowledgement. The TCP data sender is unable to distinguish between a duplicate acknowledgement that results from a lost or delayed data packet, and a duplicate acknowledgement that results from the sender's retransmission of a data packet that had already been received at the TCP data receiver. Because of this, multiple segment losses from a single window of data can sometimes result in unnecessary multiple Fast Retransmits (and multiple reductions of the congestion window) [F94]. With the Fast Retransmit and Fast Recovery algorithms in Reno TCP, the performance problems caused by multiple Fast Retransmits are relatively minor compared to the potential problems with Tahoe TCP, which does not implement Fast Recovery. Nevertheless, unnecessary Fast Retransmits can occur with Reno TCP unless some explicit mechanism is added to avoid this, such as the use of the "recover" variable. (This modification is called "bugfix" in [F98], and is illustrated on pages 7 and 9. Unnecessary Fast Retransmits for Reno without "bugfix" is illustrated on page 6 of [F98].) Section 3 of RFC 2582 defined a default variant of NewReno TCP that did not use the variable "recover", and did not check if duplicate ACKs cover the variable "recover" before invoking Fast Retransmit. With this default variant from RFC 2582, the problem of multiple Fast Retransmits from a single window of data can occur after a Retransmit Timeout (as in page 8 of [F98]) or in scenarios with reordering (as in the validation test "./test-all-newreno newreno5_noBF" in directory "tcl/test" of the NS simulator. This gives performance similar to that on page 8 of [F03].) RFC 2582 also defined Careful and Less Careful variants of the NewReno algorithm, and recommended the Careful variant. The algorithm specified in Section 3 of this document corresponds to the Careful variant of NewReno TCP from RFC 2582, and eliminates the problem of multiple Fast Retransmits. This algorithm uses the variable "recover", whose initial value is the initial send sequence number. After each retransmit timeout, the highest sequence number transmitted so far is recorded in the variable "recover". If, after a retransmit timeout, the TCP data sender retransmits three consecutive packets that have already been received by the data Floyd & Henderson [Page 9] draft-ietf-tsvwg-newreno June 2003 receiver, then the TCP data sender will receive three duplicate acknowledgements that do not cover more than "recover". In this case, the duplicate acknowledgements are not an indication of a new instance of congestion. They are simply an indication that the sender has unnecessarily retransmitted at least three packets. We note that if the TCP data sender receives three duplicate acknowledgements that do not cover more than "recover", the sender does not know whether these duplicate acknowledgements resulted from a new packet drop or not. For a TCP that implements the algorithm specified in Section 3 of this document, the sender does not infer a packet drop from duplicate acknowledgements in these circumstances. As always, the retransmit timer is the backup mechanism for inferring packet loss in this case. 7. Implementation issues for the data receiver. [RFC2581] specifies that "Out-of-order data segments SHOULD be acknowledged immediately, in order to accelerate loss recovery." Neal Cardwell has noted that some data receivers do not send an immediate acknowledgement when they send a partial acknowledgment, but instead wait first for their delayed acknowledgement timer to expire [C98]. As [C98] notes, this severely limits the potential benefit from NewReno by delaying the receipt of the partial acknowledgement at the data sender. Our recommendation is that the data receiver send an immediate acknowledgement for an out-of-order segment, even when that out-of-order segment fills a hole in the buffer. 8. Implementation issues for the data sender. In Section 3, Step 5 above, it is noted that implementations should take measures to avoid a possible burst of data when leaving Fast Recovery, in case the amount of new data that the sender is eligible to send due to the new value of the congestion window is large. This can arise during NewReno when ACKs are lost or treated as pure window updates, thereby causing the sender to underestimate the number of new segments that can be sent during the recovery procedure. Specifically, bursts can occur when the FlightSize is much less than the new congestion window when exiting from Fast Recovery. One simple mechanism to avoid a burst of data when leaving Fast Recovery is to limit the number of data packets that can be sent in response to a single acknowledgment. (This is known as "maxburst_" in the ns simulator.) Other possible mechanisms for avoiding bursts include rate-based pacing, or setting the slow-start threshold to the resultant congestion window and then resetting the congestion window to FlightSize. A recommendation on the general mechanism to avoid excessively bursty sending patterns is outside the scope of this Floyd & Henderson [Page 10] draft-ietf-tsvwg-newreno June 2003 document. 9. Simulations Simulations with NewReno are illustrated with the validation test "tcl/test/test-all-newreno" in the NS simulator. The command "../../ns test-suite-newreno.tcl reno" shows a simulation with Reno TCP, illustrating the data sender's lack of response to a partial acknowledgement. In contrast, the command "../../ns test-suite- newreno.tcl newreno_B" shows a simulation with the same scenario using the NewReno algorithms described in this paper. 10. Comparisons between Reno and NewReno TCP. As we stated in the introduction, we believe that the NewReno modification described in this document improves the performance of the Fast Retransmit and Fast Recovery algorithms of Reno TCP in a wide variety of scenarios. This has been discussed in some depth in [FF96], which illustrates Reno TCP's poor performance when multiple packets are dropped from a window of data and also illustrates NewReno TCP's good performance in that scenario. We do, however, know of one scenario where Reno TCP gives better performance than NewReno TCP, that we are describe here for the sake of completeness. Consider a scenario with no packet loss, but with sufficient reordering that the TCP sender receives three duplicate acknowledgements. This will trigger the Fast Retransmit and Fast Recovery algorithms. With Reno TCP or with Sack TCP, this will result in the unnecessary retransmission of a single packet, combined with a halving of the congestion window (shown on pages 4 and 6 of [F03]). With NewReno TCP, however, this reordering will also result in the unnecessary retransmission of an entire window of data (shown on page 5 of [F03]). While Reno TCP performs better than NewReno TCP in the presence of reordering, NewReno's superior performance in the presence of multiple packet drops generally outweighs its less optimal performance in the presence of reordering. (Sack TCP is the preferred solution, with good performance in both scenarios.) This document recommends the Fast Retransmit and Fast Recovery algorithms of NewReno TCP instead of those of Reno TCP for those TCP connections that do not support SACK. We would also note that NewReno's Fast Retransmit and Fast Recovery mechanisms are widely deployed in TCP implementations in the Internet today, as documented in [PF01]. For example, tests of TCP implementations in several thousand web servers in 2001 showed that for those TCP connections where the web browser was not SACK-capable, more web servers used the Fast Retransmit and Fast Recovery algorithms of NewReno than those of Reno or Tahoe TCP Floyd & Henderson [Page 11] draft-ietf-tsvwg-newreno June 2003 [PF01]. 11. Changes relative to RFC 2582 The purpose of this document is to advance the NewReno's Fast Retransmit and Fast Recovery algorithms in RFC 2582 to Proposed Standard. The main change in this document relative to RFC 2582 is to specify the Careful variant of NewReno's Fast Retransmit and Fast Recovery algorithms. The base algorithm described in RFC 2582 did not attempt to avoid unnecessary multiple Fast Retransmits that can occur after a timeout (described in more detail in the section above). However, RFC 2582 also defined "Careful" and "Less Careful" variants that avoid these unnecessary Fast Retransmits, and recommended the Careful variant. This document specifies the previously-named "Careful" variant as the basic version of NewReno. As described below, this algorithm uses a variable "recover", whose initial value is the send sequence number. The algorithm specified in Section 3 checks whether the acknowledgement field of a partial acknowledgement covers *more* than "recover". Another possible variant would be to require simply that the acknowledgement field *cover* "recover" before initiating another Fast Retransmit. We called this the Less Careful variant in RFC 2582. There are two separate scenarios in which the TCP sender could receive three duplicate acknowledgements acknowledging "recover" but no more than "recover". One scenario would be that the data sender transmitted four packets with sequence numbers higher than "recover", that the first packet was dropped in the network, and the following three packets triggered three duplicate acknowledgements acknowledging "recover". The second scenario would be that the sender unnecessarily retransmitted three packets below "recover", and that these three packets triggered three duplicate acknowledgements acknowledging "recover". In the absence of SACK, the TCP sender in unable to distinguish between these two scenarios. For the Careful variant of Fast Retransmit, the data sender would have to wait for a retransmit timeout in the first scenario, but would not have an unnecessary Fast Retransmit in the second scenario. For the Less Careful variant to Fast Retransmit, the data sender would Fast Retransmit as desired in the first scenario, and would unnecessarily Fast Retransmit in the second scenario. This document only specifies the Careful variant in Section 3. Unnecessary Fast Retransmits with the Less Careful variant in scenarios with reordering are illustrated in page 8 of [F03]. Floyd & Henderson [Page 12] draft-ietf-tsvwg-newreno June 2003 12. Conclusions This document specifies the NewReno Fast Retransmit and Fast Recovery algorithms for TCP. This NewReno modification to TCP can be important even for TCP implementations that support the SACK option, because the SACK option can only be used for TCP connections when both TCP end-nodes support the SACK option. NewReno performs better than Reno (RFC 2581) in a number of scenarios discussed herein. A number of options to the basic algorithm presented in Section 3 are also described. These include the handling of the retransmission timer (Section 4), the response to partial acknowledgments (Section 5), and the value of the congestion window when leaving Fast Recovery (section 3, step 5). Our belief is that the differences between these variants of NewReno are small compared to the differences between Reno and NewReno. That is, the important thing is to implement NewReno instead of Reno, for a TCP connection without SACK; it is less important exactly which of the variants of NewReno is implemented. 13. Acknowledgements Many thanks to Anil Agarwal, Mark Allman, Armando Caro, Vern Paxson, Kacheong Poon, Keyur Shah, and Bernie Volz for detailed feedback on this document or on its precursor RFC 2582. Floyd & Henderson [Page 13] draft-ietf-tsvwg-newreno June 2003 14. References Normative References [RFC2018] M. Mathis, J. Mahdavi, S. Floyd, A. Romanow, "TCP Selective Acknowledgement Options", RFC 2018, October 1996. [RFC2581] W. Stevens, M. Allman, and V. Paxson, "TCP Congestion Control", RFC 2581, April 1999. [RFC2582] S. Floyd and T. Henderson, The NewReno Modification to TCP's Fast Recovery Algorithm, RFC 2582, April 1999. [RFC3042] M. Allman, H. Balakrishnan, and S. Floyd, Enhancing TCP's Loss Recovery Using Limited Transmit, RFC 3042, January 2001. Informative References [C98] Neal Cardwell, "delayed ACKs for retransmitted packets: ouch!". November 1998. Email to the tcpimpl mailing list, Message-ID "Pine.LNX.4.02A.9811021421340.26785-100000@sake.cs.washington.edu", archived at "http://tcp-impl.lerc.nasa.gov/tcp-impl". [F98] Sally Floyd. Revisions to RFC 2001. Presentation to the TCPIMPL Working Group, August 1998. URLs "ftp://ftp.ee.lbl.gov/talks/sf-tcpimpl-aug98.ps" and "ftp://ftp.ee.lbl.gov/talks/sf-tcpimpl-aug98.pdf". [F03] Sally Floyd. Moving NewReno from Experimental to Proposed Standard? Presentation to the TSVWG Working Group, March 2003. URLs " "http://www.icir.org/floyd/talks/newreno-Mar03.ps" and "http://www.icir.org/floyd/talks/newreno-Mar03.pdf". [FF96] Kevin Fall and Sally Floyd. Simulation-based Comparisons of Tahoe, Reno and SACK TCP. Computer Communication Review, July 1996. URL "ftp://ftp.ee.lbl.gov/papers/sacks.ps.Z". [F94] S. Floyd, TCP and Successive Fast Retransmits. Technical report, October 1994. URL "ftp://ftp.ee.lbl.gov/papers/fastretrans.ps". [Hen98] Tom Henderson, Re: NewReno and the 2001 Revision. September 1998. Email to the tcpimpl mailing list, Message ID "Pine.BSI.3.95.980923224136.26134A-100000@raptor.CS.Berkeley.EDU", archived at "http://tcp-impl.lerc.nasa.gov/tcp-impl". [Hoe95] J. Hoe, Startup Dynamics of TCP's Congestion Control and Avoidance Schemes. Master's Thesis, MIT, 1995. URL "http://ana- Floyd & Henderson [Page 14] draft-ietf-tsvwg-newreno June 2003 www.lcs.mit.edu/anaweb/ps-papers/hoe-thesis.ps". [Hoe96] J. Hoe, Improving the Start-up Behavior of a Congestion Control Scheme for TCP. In ACM SIGCOMM, August 1996. URL "http://www.acm.org/sigcomm/sigcomm96/program.html". [LM97] Dong Lin and Robert Morris, "Dynamics of Random Early Detection", SIGCOMM 97, September 1997. URL "http://www.acm.org/sigcomm/sigcomm97/program.html". [NS] The Network Simulator (NS). URL "http://www.isi.edu/nsnam/ns/". [PF01] J. Padhye and S. Floyd, Identifying the TCP Behavior of Web Servers. June 2001, SIGCOMM 2001. 15. Security Considerations RFC 2581 discusses general security considerations concerning TCP congestion control. This document describes a specific algorithm that conforms with the congestion control requirements of RFC 2581, and so those considerations apply to this algorithm, too. There are no known additional security concerns for this specific algorithm. AUTHORS' ADDRESSES Sally Floyd International Computer Science Institute Phone: +1 (510) 666-2989 Email: floyd@acm.org URL: http://www.icir.org/floyd/ Tom Henderson The Boeing Company Email: thomas.r.henderson@boeing.com Floyd & Henderson [Page 15]