1 / 10

OptIPuter Networking and Middleware

OptIPuter Networking and Middleware. Joe Bannister Aaron Falk Carl Kesselman Marcus Thiébaux USC Information Sciences Institute February 7, 2003. The Need for High-Performance Transport. TCP performance suffers in some environments High bandwidth-delay networks

tgreco
Télécharger la présentation

OptIPuter Networking and Middleware

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. OptIPuter Networking and Middleware Joe Bannister Aaron Falk Carl Kesselman Marcus Thiébaux USC Information Sciences Institute February 7, 2003

  2. The Need for High-Performance Transport • TCP performance suffers in some environments • High bandwidth-delay networks • Cautious probing of bandwidth • Aggressive backoff after presumed packet loss • Designed for under-engineered networks • Stability and farness for all • Information comes from the edges of the network • Crude congestion signals from the core OptIPuter Project Meeting

  3. Explicit Control Protocol (XCP) • XCP is a newly proposed congestion-control scheme • Decouples fairness control from congestion control • Use aggressive window adjustments (MIMD) for congestion control which can grab/release large bandwidth quickly • More conservative window adjustments (AIMD) provide fairness control which converges to fair bandwidth allocation • Bottleneck router explicitly tells sender about bandwidth allocation • Applications may request explicit send rate • XCP requires modified router behavior but no per-flow state in routers • State carried in congestion header in packets Dina Katabi, Mark Handley, and Charles Rohrs, “Internet Congestion Control for Future High Bandwidth-Delay Product Environments,” Proc. ACM Sigcomm 2002, Aug. 2002. http://ana.lcs.mit.edu/dina/XCP/ OptIPuter Project Meeting

  4. Congestion Window Set by Bottleneck Router OptIPuter Project Meeting

  5. XCP Performance • Superior to TCP • Almost never drops packets • Converges to available bandwidth very quickly, ~1 RTT • Fair over large variations in bandwidth and RTT • Supports existing TCP semantics • Replaces only congestion control, reliability unchanged • No change to application/network API • Status • Analysis based entirely on ns-2 simulations • Software design and coding begun (multiple implementations) • Experimentation and interoperability testing planned • Researcher and vendor collaborations OptIPuter Project Meeting

  6. Comparison of XCP and TCP Avg. Utilization Avg. Utilization Bottleneck Bandwidth (Mb/s) Round Trip Delay (sec) From http://www.ana.lcs.mit.edu/dina/XCP/Sigcomm2002.ppt OptIPuter Project Meeting

  7. Credits and Collaboration • Dina Katabi – proposed XCP • USC ISI • Aaron Falk • Ted Faber • Bob Braden • MIT LCS • John Wroclawski • Dina Katabi • UCB ICIR • Sally Floyd • Mark Handley (till May 2003) • UCL • Mark Handley (after May 2003) • PSC • Matt Mathis OptIPuter Project Meeting

  8. Progress and Plans • Wide ongoing collaboration effort • Mailing list • Web site • Design, implementation, and interoperability • Protocol design • ns-2 simulations • Coding • Linux/FreeBSD hosts and routers • IPX-1200 • Globus Toolkit integration • Application-level performance • IETF and standardization • A. Falk, Transport Area Directorate, PILC/DCCP WG Chair • Vendor adoption • Collaboration and prototyping • Market plans OptIPuter Project Meeting

  9. ISI Personnel for XCP Development Task • Aaron Falk: software development leader, standardization • Ted Faber: software development, congestion management • Venkata Pingali (GRA): simulation, performance analysis • Bob Braden: collaboration liaison, senior advisor • Joe Bannister: task manager OptIPuter Project Meeting

  10. Why Use XCP? • XCP is a serious candidate to replace TCP congestion control in the Internet • No change to bulk of TCP kernel code or API semantics • Excellent performance, link utilization, and packet loss across wide ranges of RTT, link bandwidth, flow size • Capable of fair coexistence with other Internet protocols • Gradual Internet deployment possible • But … requires changes to routers OptIPuter Project Meeting

More Related