Authors: Les Cottrell. Created: November 6, 2001
Attendees: Rolf Riedi, Rob Nowag, Warren Matthews, Les Cottrell
Set up Wednesday meeting with Thomas. Les has an Internet 2 E2Epi from 11am probably to 12:30pm. Will try and get a room later. Need to think about time required. We deferred this until we pinned down what we wanted to use the time for. Robert Nowak, Rolf Riedi, Les Cottrell, Wu Fang and Thomas are expected to attend. Warren will be at SLAC, and Richard Baraniuk is on Sabatical in Lyon France. Thomas has presentations on Thursday.
Today's meeting agenda:
Goal of chirping is to learn about losses, jitter, RTT, cross traffic (i.e. additional delay and jitter), bandwidth capacity etc.. Thomas wants to build this into a PingER like infrastructure. Could cross-validate bandwidth estimators with Chirp analysis, and with iperf. SLAC has excellent relations with the data grid people, HENP, ESnet and high performance sites, which may help deployment and gather interest.
We promised to deliver a set of tools to exploit the understanding of networks. We will develop E2E tools for high speed networks.
The tools mentioned in the proposal and funded include:
there is interest in deploying chirp and the sandwich (see below) at PingER sites. Also in providing chirp feedback to an application such as bbcp to enable the application to pace itself. Bbcp is a file copy utility developed at SLAC that supports multiple streams, big windows, sequential output (can read and write from tapes, /dev/zero, /dev/null), supports time limits, checkpointing, as well as being able to set the bandwidth it uses, & providing progress reports.
Tomography using chirp and "fat boy sandwiches" (two thin packets around one or more big packets) and see the effect of extra delays on 2nd thin packet due to fat boys. It also provides tomography information by showing branch points from say A to multiple other places. We could deploy this at PingER NG sites and measure from SLAC.
Validation of tools for high performance, e.g. by comparing the various measurements, chirp, ping, iperf, bbcp ...
Goal achieving better performance throughput on high speed networks. There are today a limited number of high performance sites that need measuring today. For example HENP, Grid sites, etc. Need to understand the networks, e.g. tomography, detect and localize faults (loss rates, utilization, queuing delays). This leads to the need for measurements. The measurements include existing (e.g. delays and losses, bandwidth measurements such as pipechar), and new techniques such as chirp, fat boy sandwiches, and applications such as iperf, bbcp. The measurements need validation between one another to assess the confidences (e.g. variability).
What are the questions we will answer:
Send email to Thomas asking what he wants at Denver, could suggest we make presentations, or do free form discussions. However, Thomas said it should be an INCITE kickoff meeting so we should drive the agenda.
Present high level transparencies of highlights (total 25 minutes without interruption, but interruptions are welcome and necessary):
Rolf & Rob will arrive Denver around 11am Wednesday 14th, meet around 1pm for 1.5 to 2 hours, think about early dinner together. Wu, Les & Thomas will be there already for SC2001.