SLAC / Internet2 E2Epi meeting

Held at SLAC January 13, 2003

Rough Notes by Les Cottrell

Attendees: Eric Boyd, Russ Hobby (Internet 2), Les Cottrell, Connie Logg, Jiri Navratil, Warren Matthews, Fabrizio Coccetti, Jerrod Williams  (SLAC)

E2E piPES - Eric Boyd

Want to provide tools to address E2E performance problems, to aid the user in providing input to the network adminstrators to fix/by-pass probems. Want to provide glue to bring together a measurement infrastructure. New Abilene will have measurement devices as part of its infrastructure (11 nodes full mesh). Want to extend to campus borders. Will perform regular tests. Want to enable users to look at regular test results and also to be able to make tests. Can extend to GigaPoPsd and regional networks.

PiPES will provide who the contract is, and provide information to be given to the contact. New Abilene is running, need to get software developed and deployed. Do not expect a uniform set of boxes, will provide a template etc. to enable people to build their own platforms. Concerned about quality if have multiple heterogeneous. May need to strictly limit  what tools, what OS'. Showed prototype NLANR form to select source/destination and test to be performed or if tests are available then access them. The architecture allows for existing results and on demand measurements, it covers the whole gamut of trouble-shooting (i.e. a complete vision). For version 1 (by summer Joint Techs) want to have a: development environment, collaboration forums, deploy a core tool set of OWAMP, Iperf, traceroute, determine schema for measurements, build a distributed database of measurements, visualize and use data from database, deploy scheduler, extend to Abilene and 2 initial campuses. Version 2 will have an end user GUI and end user PMP, a testing and analyzing engine, a contact database, tool abstraction/grid service, authentication/authorization, multiple PMP templates, right way to do v 1.0, wider deployment (gigapops, 2 campuses). End of this month want OWAMP deployed on Abilene nodes and collected and made available. Next step would be iperf. Then will need a scheduler.

R&D Issues:

Will have a Lab environment for software development, then will roll out to other sites as it becomes production.

Eight I2 people working on it including software developers, net engineers, documentation person. Peter Clarke embraced and has 2-3 people who will work on it from UCL (want to make it a Grid Services effort a la GGF).  We need a road-map of how SLAC will be involved, what can be done and what time frames. Also NLANR/DAST folks are interested.

Questions/Answers:

IEPM Overview and Motivation - Les Cottrell

IEPM-BW Architecture - Connie Logg

Redesigning to increase modularity, to simplify adding new tests, provide more customization.

Big challenge is the failure modes of the application and the remote hosts (e.g. insufficient disk space, port blocked, ssh keys don't work), hung processes.

 Test engine done, next performance analysis, then get the analysis, reporting etc. working. If no major interruption may be ready mid February.

IEPM-BW & PIPES analysis & publishing - Warren Matthews

Detecting problems by rolling averages and looking for below 1 or 2 standard deviations below rolling averages.

Guthrie provides web server access to a distributed database. 

Want to get the NLANR/DAST people (Kevin Walsh) to be a repository for the iepm-bw data. If the PiPES OWAMP data was archived there then we (SLAC) could start to analyze the data.

Jalaam start-up in Washington/Vancouver, requires a single end only, sends pattern of packets, diagnoses problems. Want to provide a deal for universities, can buy GUI and sequencers separately from analyzers. I2 set up sequencers so can run tests, San Diego bought sequencer too. Interim approach before PiPES. Share an analyzer.

Another interesting tool (free) is Rich. Carlson's Network Configuration tester at http://miranda.ctd.anl.gov:7123/

It would be good to get a specific project to work together on, it would act as a catalyst for further development etc.

Warren will work with the PiPES folks to look at the OWAMP data and tie it into his trouble-shooting.

Quick Links