Internet Performance and IP Provider Metrics

Papers about of Internet performance and its measurement:

• Diagnosing Internet Congestion with a Transport Layer Performance Tool by Matt Mathis and Jamshid Mahdavi, presented at INET'96 in Montreal, Quebec (Session D3)

  In this paper we present an overview of one particular metric, bulk transfer capacity, and a tool which could be a basis for a formal bulk transfer metric. Bulk transfer capacity requires a very clean network and has other properties that make it an important gauge of network performance. We introduce our tool, TReno, and describe its relation to current TCP implementations. Our research with TReno led us to key observations about TCP dymanics and influenced details of the new specifications for the TCP Selective Acknowledgment (SACK) option. TReno does not precisely estimate current TCP implementations' performance over the Internet. As SACK is deployed, and the research community reaches consensus regarding its proper behavior, TReno and state-of-the-art TCP will converge to identical performance and behavior.

  The slides from the prblic presentation are online. (Note that the CD-ROM version of the proccedings have a bug in the index. You must go to an adjacent session and use the "next" or "previous" buttons to reach session D3).

• The TReno tool itself is available as source or you can try it yourself from PSC via a form based TReno test server.

• Towards a Framework for Defining Internet Performance Metrics by Vern Paxson (also INET'96 session D3) presents an initial framework for defining formal metrics.

• The Internet-Draft draft-almes-ippm-framework-00.txt is the start at a formal (standard?) framework for provider metrics. Once this document becomes sufficiently mature we hope to move it into the IETF standards track.

• The revised "TCP Selective Acknowledgement Options" (superseeding part of RFC1072) is currently in IETF last-call for proposed standards status. The current version is draft-ietf-tcplw-sack-02.txt. SACK will have a substantial impact on TCP's efficiency under a number of adverse conditions.

• Tracie Monk (DynCorp) and Kimberly Claffy (NLANR) have compiled A Survey of Internet Statistics / Metrics Activities.

• The agenda and minutes from the first IPPM BOF (at the 32nd IETF in Danvers MA) have the best introductory discussion of IP provider metrics, although some of the details have already changed.

The Official Working Group

• The official IPPM pages are (will be?) under http://www.advanced.org/IPPM/index.html and ftp://ftp.advanced.org/pub/IPPM/.

• To request a subscription to the IPPM mailing list, send a note to ippm-request@advanced.org.

• Use ippm@advanced.org to send mail to the entire group. DO NOT send subscription messages to this address.

Minutes and other records from past IPPM meetings:

• Minutes and archive Montreal QE, June 1996, (under BMWG at the 36th IETF)
  (Not online yet. Check the IETF pages for updates.)

• Minutes and archive Los Angeles CA, March 1996 (under BMWG at the 35th IETF)
  (Not online yet. Check the IETF pages for updates.)

• Minutes and archive Dallas TX, December 1995 (under BMWG at the 34th IETF)

• Minutes and archive (More) Stockholm Sweden, July 1995 (BMWG at the 33rd IETF), to discuss the formal relationship between IPPM and BMWG.

• Minutes and archive Pittsburgh PA, September 1995, (At the Pittsburgh Supercomputing Center, following a NANOG meeting. At that meeting Matt Mathis presented a TReno update.

• Danvers MA, April 1995 (IPPM BOF the 32nd IETF). The Agenda was to frame the task of developing IP provider metrics and to consider the formation of an IETF working group.

You may be interested in additional information about Pittsburgh Supercomputing Center, Network research at PSC, or Matt Mathis. Please send comments and suggestions to mathis@psc.edu.