Allman M., Balakrishnan H., Floyd S., Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response, Internet Draft, June 2000.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....rate after packet loss. The proposal was further developed and implemented as the Rate Halving modi cation [124] which we discuss in more details in Section 7.5. A recent proposal, dubbed Limited Transmit , suggests sending a new data segment for each of the rst two duplicate ACKs received [14, 18]. This increases the chance of meeting the threshold condition for Fast Retransmit. When the duplicate ACKs actually result from packet reordering inside the network, the e ect of this scheme would have been to space the new segment transmissions, instead of sending a burst when the ACK for the ....

Allman M., Balakrishnan H., Floyd S., Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response, Internet Draft, June 2000.

....one half, while others [19] set the window to one packet and assume a linear increase from this low value. We believe that there will be no difference between the two approaches in the future given the expected low probability of timeout with the new enhancements proposed to TCP congestion control [20]. Our measurements showed that on the LD connection, most of the losses are detected with timeout; this is due to the small size of the TCP window. The same result is noticed in [8] On the SD and MD connections, the window is large and the timeout phenomenon is absent. In our work on TCP modeling ....

....Timeouts are assumed not to occur during the fast recovery phase. We believe that this will be the case with the new versions of TCP which use the SACK option [18] and have a robust and quick fast recovery phase. We also believe that with the modifications proposed to TCP error recovery in [20], failure to OE = n n N 2 1 , Figure 4. SDC: Window size vs. time. 1760 1755 Instantaneous TCP window (bytes) Time (s) Exact fluid model window Real window 10,000 0 20,000 30,000 40,000 50,000 60,000 70,000 1765 1770 1775 1780 Figure 5. The SD connection: expected RTT ....

M. Allman, H. Balakrishnan, and S. Floyd, "Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response," Internet draft, June 2000, work in progress.

....in section 5. Section 6 presents our experimental evaluation, and section 7 summarizes the paper. 2 Related Work For the last decade, researchers have studied and proposed improvements to TCP IP performance. These improvements include changes to TCP host implementations (e.g. Reno, SACK, etc [1, 7, 8, 9, 18, 25]) better active queue management algorithms (e.g. RED, FRED, etc [3, 10, 14, 15, 20, 21] or schemes which require both end system and bottleneck support (e.g. ECN, adaptive packet marking [5, 22, 24] Packeteer s rate control [13] is a buffer management scheme which manipulates TCP headers and ....

....cannot provide a high level of assurance that timeouts would be contained, especially during heavy loads and or during high degrees of TCP multiplexing. There are also end system proposals to reduce the TCP dupack threshold and make other TCP level implementation improvements to address this issue [1, 8, 9, 18]. This paper leverages the dimension of packet marking by extending our recent work [6] in the following ways: ffl Mark packets to protect small window ( 4 MSS) flows from packet loss; ffl Mark packets to maintain maximum spacing between packets marked as OUT (low priority) within limits of ....

M. Allman, H. Balakrishnan and S. Floyd, "Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response," IETF Internet RFC 3042, Janurary 2000.

.... in transmission media heterogeneous infrastructures, which exhibit characteristics different from wired ones, e.g. random segment corruption over wireless links (Lee, 1993; Tabbane, 2000) Different improvement proposals attempt to alleviate TCP s inefficiencies in a wired cum wireless environment (Allman et al. 2000; Bakre Badrinath, 1995; Balakrishnan et al. 1995; Cceres Iftode, 1995; Parsa Garcia Luna Aceves, 1999; Sinha et al. 1999) Testing the comparative efficiency of these proposals requires realistic error models (Zorzi Rao, 1999; Pentikousis, 2000) 2 TEMPORAL VS. RATE BASED ERROR MODELS ....

Allman, M., Balakrishnan, H., and Floyd, S., (2000). Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response, Internet Draft, June 2000.

....FRIO. Section 6 presents our experimental evaluation, and section 6.2 summarizes the paper. 2 Related Work For the last decade, researchers have studied and proposed improvements to TCP IP performance. These improvements include changes to TCP host implementations (eg: NewReno, SACK, FACK etc [1, 7, 8, 9, 18, 25]) better queue management algorithms (Eg: AQM, RED, SRED, FRED, FPQ etc [3, 10, 15, 20, 21, 14] or schemes which require both end system and bottleneck support (eg: ECN, adaptive packet marking [5, 22, 24] Packeteer s rate control [13] is a buffer management scheme which manipulates TCP ....

....cannot provide a high level of assurance that timeouts would be contained, especially during heavy loads and or during high degrees of TCP multiplexing. There are also end system proposals to reduce the TCP dupack threshold and make other TCP level implementation improvements to address this issue [1, 8, 9, 18]. We view our work as complementary to these efforts given the large deployed base of old TCP implementations. This paper leverages the dimension of packet marking (extending our recent work [6] in the following ways: ffl Mark packets to protect small window (window 4MSS) flows from packet ....

M. Allman, H. Balakrishnan and S. Floyd, "Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response," Internet draft, draft-allman-tcp-lossrec-00.txt, June 2000. http://www.aciri.org/floyd/papers.html

....for a Retransmit Timeout to learn of the lost packet. As discussed in Section 3.2, the use of Explicit Congestion Notification (ECN) would also avoid the unnecessary Retransmit Timeout in this case. The Limited Transmit mechanism has been submitted to the IETF and approved by the Working Group [2], and we hope that it will soon become an accepted part of the TCP specification. This should help in reducing unnecessary retransmit timeouts, while preserving the fundamental role of retransmit timers in congestion control for regimes where the available bandwidth is at most one packet per ....

M. Allman, H. Balakrishnan, and S. Floyd. Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response. Internet draft draft-allmantcp -lossrec-00.txt, work-in-progress, Jun. 2000.

....reaching 8 segments in the majority of the cases. This argues that default advertised windows should be increased. Rather than increasing the advertised window size, several researchers have suggested that TCP send new segments upon the reception of the first two duplicate ACKs [BPS 98, LK98, ABF00] This will trigger additional duplicate ACKs (if appropriate) and therefore fast retransmit will be invoked. However, the above algorithm does not aid short connections that have no new data to transmit in response to duplicate ACKs. ECN [Flo94, RF99] provides a possible mitigation to this ....

....that (i) one duplicate ACK has arrived, ii) there is no more data to send and (iii) based on the number of outstanding segments 3 duplicate ACKs cannot be expected. In this case, the TCP would trigger fast retransmit on a smaller number of duplicate ACKs. Such a scheme is discussed further in [ABF00] Semke [SMM98] argues for the use of an automatic socket buffer tuning algorithm and the notion that the network should dictate the performance of a TCP connection, rather than being limited by some arbitrary limit placed on the transfer by one of the endpoints. Note that the advertised window ....

[Article contains additional citation context not shown here]

Mark Allman, Hari Balakrishnan, and Sally Floyd. Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response, May 2000. Internet-Draft draft-allman-tcp-lossrec00. txt (work in progress).

....to recover from less that a full window of packet losses without a Retransmit Timeout. As discussed further in Section 3.2, the use of Explicit Congestion Notification (ECN) can also help to avoid unnecessary Retransmit Timeouts. The Limited Transmit mechanism has been submitted to the IETF [3], and we hope that it will soon become an accepted part of the TCP specification. This should reduce unnecessary retransmit timeouts, while preserving the fundamental role of retransmit timers in congestion control for regimes where the available bandwidth is at most one packet per round trip ....

M. Allman, H. Balakrishnan, and S. Floyd. Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response. Internet draft draft-allmantcp -lossrec-00.txt, work-in-progress, Jun. 2000.

No context found.

M. Allman, H. Balakrishnan, and S. Floyd. Enhancing TCP's Loss Recovery Using Early Duplicate Acknowledgment Response, Internet Draft, June 2000. Available at http://wwwsnmp.cs.utwente.nl/ietf/internetdrafts/complete/draft-allman-tcp-lossrec-00.txt

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC