1 / 30

One More Bit Is Enough

One More Bit Is Enough. Yong Xia, RPI Lakshminarayanan Subramanian, UCB Ion Stoica, UCB Shivkumar Kalyanaraman, RPI SIGCOMM’05, August 22-26, 2005, Philadelphia, Pennsylvania, USA. Goal.

adah
Download Presentation

One More Bit Is Enough

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. One More Bit Is Enough Yong Xia, RPI Lakshminarayanan Subramanian, UCB Ion Stoica, UCB Shivkumar Kalyanaraman, RPI SIGCOMM’05, August 22-26, 2005, Philadelphia, Pennsylvania, USA

  2. Goal • Achieve fair bandwidth allocation and high utilization, and minimize packet loss in high bandwidth-delay product (BDP) network • TCP ? • TCP + AQM / ECN? • XCP ? VCP 2

  3. congestion congestion avoidance slow! slow start Additive Increase (AI) Why TCP does not scale? • TCP uses duplicate ACK or timeout: no degree of congestion and instability congestion window Multiplicative Decrease (MD) time • AI with a fixed step-size can be very slow for large bandwidth VCP 3

  4. bottleneck utilization bottleneck utilization 100% 100% TCP TCP 70% 50% bandwidth (Mbps) round-trip delay (ms) TCPdoes not perform well in high b/w VCP 4

  5. XCP scales C spare bandwidth DATA x ACK rate sender router receiver • But, XCP needs multiple bits (128 bits in its current IETF draft) to carry the congestion-related information from/to network VCP 5

  6. Goal Design a TCP-like scheme that: • requires a small amount of congestion information (e.g., 2 bits) • scales across a wide range of network scenarios VCP 6

  7. General idea degree of explicit information XCP: multiple bits, report spare bandwidth VCP: two bits, low-load, high-load and overload ECN: one bit, underload and overload TCP: no explicit feedback, relay on duplicate ACKand timeout

  8. range of interest scale-free region code control load factor overload (11) Multiplicative Decrease (MD) 1 high-load (10) Additive Increase (AI) (01) low-load Multiplicative Increase (MI) 0 traffic rate link capacity x 2-bit ECN 2-bit ECN sender router receiver ACK Variable-structure congestion Control Protocol (VCP) • Routerssignalthelevel of congestion • End-hosts adapt the control algorithm accordingly VCP 8

  9. ECN doesn’t differentiate between low-load and high-load regions TCP VCP AQM/ECN load factor (1) (11) 1 high-load (10) Additive Increase (AI) Additive Increase (AI) underload (0) (01) low-load Multiplicative Increase (MI) 0 x 2-bit ECN 1-bit ECN sender router receiver ACK ACK VCP vs. ECN region code control overload Multiplicative Decrease (MD) VCP 9

  10. TCP+RED/ECN utilization cwnd time (sec) VCP vs. ECN VCP utilization cwnd time (sec) VCP 10

  11. router end-host overload high-load fairness control AIMD low-load efficiency control MI VCP key ideas and properties • Use network link load factor as the congestion signal • Decouple efficiencyandfairness controls in different load regions • Achieve high efficiency, low loss, and small queue • Fairness model is similar to TCP: • Long flows get lower bandwidth than in XCP (proportional vs. max-min fairness) • Fairness convergence much slower than XCP VCP 11

  12. control load region code overload (11) Multiplicative Decrease (MD) high-load Additive Increase (AI) (10) (01) low-load Multiplicative Increase (MI) Major design issues • At the router • How to measure and encode the load factor? • At the end-host • When to switch from MI to AI? • What MI/AI/MD parameters to use? • How to handle heterogeneous RTTs? VCP 12

  13. t= 200ms most rtt demand load_factor = capacity arrival_traffic + queue_size = Design issue #1: measuring and encoding load factor • Calculate the link load factor  link_bandwidth * t • The load factor is quantized and encoded into the two ECN bits VCP 13

  14. overload high-load AIMD low-load MI Design issue #2: setting MI/AI/MD parameters (, , ) VCP 14

  15.  =  = 0.875 *= safety margin 80% 20%  = 80% Design issue #2: setting MI/AI/MD parameters (, , ) load factor MD 100% AI 1.0 TCP:  = 1.0 VCP:  = 0.06 STCP:  = 0.01 k (1  *) / * where k = 0.25 (for stability) MI 0% VCP 15

  16. Design issue #3: Handling RTT heterogeneity for MI/AI VCP 16

  17. 150Mbps, 80ms, 50 forward flows and 50 reverse flows VCP scales across b/w, rtt, num flows • Evaluation using extensive ns2 simulations VCP 17

  18. Impact of bottleneck capacity vary link capacity from 100Kbps to 5Gbps high utilization, gap comparing toXCP at most 7% at extremely low capacities alpha = 1 is too large for such low capacity

  19. Impact of feedback delay average queue < 5%maximal queue < 15% fix bottleneck capacity at 150Mbps, vary round-trip propagation delay from 1ms to 1500ms utilization > 90% no packet loss lower utilization due to comparable smallmeasurement interval, i.e. tp = 200ms << RTT

  20. Impact of number of long-lived flows small queue length, even outperform XCP increase the # forward FTP flows high utilization zero packet drop

  21. Impact of short-lived traffic small queue length arrive according to Poisson process, avg. arrival rate varying from 1/s to 1k/s,transfer size obeys Pareto distribution with avg. 30 packets high utilization zero packet drop

  22. Multiple bottlenecks < 0.2% buffer size average queue length typical parking-lot topology as good as single-bottleneck scenarios zero packet drop

  23. Fairness • same RTT • small RTT difference • huge RTT difference even distribution among all flows in case (a) and (b) fairness discrepancy due to high value of MI and AI, whose value is bound in implementation to prevent burst

  24. Convergence behavior high utilization during the whole period introduce 5 flows one after another use router buffer size to scale queue length axis; low queue length during the whole period

  25. Sudden demand change high utilization 150 flows join at 80s and leave at 160s remain much lower than full size

  26. Conclusions • With a few minor changes over TCP+AQM/ECN, VCP is able to approximate the performance of XCP • High efficiency • Low persistent bottleneck queue • Negligible congestion-caused packet loss • Reasonable (i.e., TCP-like) fairness VCP 26

  27. VCP comparisons • Compared to TCP+AQM/ECN • Same architecture (end-hosts control, routers signal) • Router congestion detection: queue-based  load-based • Router congestion signaling: 1-bit  2-bit ECN • End-host adapts (MI/AI/MD) according to the ECN feedback • End-host scales its MI/AI parameters with its RTT • Compared to XCP • Decouple efficiency/fairness control across load regions • Functionality primarily placed at end-hosts, not in routers VCP 27

  28. THE END VCP 28

  29. VCP Parameter Setting

  30. Remark Modified from www.ecse.rpi.edu/Homepages/ shivkuma/research/papers/vcp05.ppt

More Related