INCITE meeting at SC2001 between SLAC & LANL, 11/14/01

Authors: Les Cottrell and Wu-chun Feng. Created: November 16, 2001, last update November 18, 2001

Attendees: Wu-chun Feng and Les Cottrell

XIWT | IEPM | PingER | Tutorial

Introduction

Originally a meeting between Rolf Riedi, Robert Nowag of Rice, Wu-chun Feng of LANL, Les Cottrell of SLAC and Thomas Ndousse of DoE was scheduled to coincide with SC2001 in Denver. Thomas was unable to attend so the meeting was downsized to a metting between Les and Wu. These meetings took place on Wednesday evening November 14 and Thursday morning November 14.

Discussions

Wu described in detail the TICKET project to provide a kernelized versioin of tcpdump that is able to monitor streams running at Gbits/s rates and runs on a low cost commodity PC running Linux. Wu has an instance of this working and it is able to record the data to a RAID TByte data store. he has enough storage in the store to hold about 1 week's worth of data (the data contains the headers of the packets). This will be very useful to Rice to analyze and use with their simulations. The main challenge now is to anonymize the data so its distribution is acceptable to the LANL security folks. This could be done by someone from LANL (I understood Wu is rather limited in the resources he has to apply to this task) or someone from outside (e.g. Rice) who spends time at LANL to anonymize the data.

The LANL MAGNeT project enables them to capture traffic at the application interface to the TCP stack. This is interesting since most captured data today in use for modelling is from below the TCP stack, and the stack modulates behavior quite dramatically. Thus erroneous conclusions can result from replaying the exixting data into the TCP stack simulations. Again the data would be valuable to the Rice modelling efforts, but anonymization is the next step.

Les described the IEPM/PingER project to Wu, in particular the infrastructure that is already in place for making lightweight active end to end measurements of the Internet. He went on to describe the next generation PingER NG infrastructure (a prototype of which was used for the SLAC led Bandwidth Challenge: Bandwidth to the World and will be extended after SC2001 in particular to improve reliability and manageability) which will be used for more intensive monitoring of the Internet, in particular for Grid, HENP and other high performance sites. We will build the chirp and big boy sandwich measurement techniques from Rice into these infrastructures, and look at incrorporating other bandwidth estimation techniques such as pipechar. The next steps will be to analyze and cross-validate/compare the various measurements and determine regions of applicability. Following this we expect to apply some of the techniques to application steering, in particular for the data grid and the SLAC developed file copy application bbcp.

Summary

The project will be an iterative one, e.g., use MAGNeT and/or TICKET to collect traces (LANL), analyze them via multi-fractal modeling (Rice), propose new algorithms (Rice), implement in PingER (Rice & SLAC), evaluate algorithms (Rice & SLAC), and then re-iterate (e.g., use MAGNeT and/or TICKET to analyze performance of new algorithms, etc.) Another perspective ... MAGNeT & TICKET are passively monitoring traffic whereas PingER is a lightweight active monitoring system, so they will complement one another.
[ Feedback ]