January 27, 2004 from 11:20 - 11:50 am

SPEAKERS
Eric Boyd, Internet2 [PPT]
Jim Ferguson, NLANR

Jim Ferguson talked about Tthe Advisor/E2E piPEs interoperability demo on Sunday. Jim provided background on The Advisor project. NLANR is funded, by the NSF, to help end-users (not necessarily "savvy" network engineers) to use high-end networks in an easier way, preferably to solve E2E problems. They work on the multicast beacon and Iperf. This spring, they expect to push Iperf out as an open-source project. The goal of The Advisor is to put something in the hands of end-users to help them diagnose some of the problems they are having with the network. It will be easy for Network Engineers to add in their own favorite tests but, at the end, it will analysize network tests on-demand or bring in data from other network tests to analyze and give clear text advice to non-savvy users on what they can do. The Advisor must be easily downloaded and run; it needs to be portable and extensible.

Jim emphasized that the interoperability work with Internet2 had begun more than a month before The Advisor was funded and work started. Also, that both groups are using the GGF's NM-WG schemas and that it would be great if others used these schemas when developing their projects to make interoperability easier.

Eric Boyd presented an overview of the E2E piPEs project - he emphasized that this is a heavily collaborative project: NLANR/DAST, UCL, GEANT, and HENP-funded research projects. The overarching goal for the project is to help an end-user detect a problem ("is this typical behavior?") but also to find and deliver supporting data when a problem is identified. Short term goals are to allow authorized users ("savvy" folks) to schedule on-demand tests under a scheduler that prevents duplicate tests hitting a measurement node at any given time.

As part of the Abilene Observatory, there are regularly scheudled OWAMP and BWCTL (Iperf) tests available at http://abilene.internet2.edu/observatory. All this data can be grabbed off the Abilene backbone and you can pull it into your own tool.

We are running UDP and TCP tests for v4 and v6 for all 11 Abilene nodes every 30 minutes. The Top 10 Worst for the past 8 hours are available so it makes it easier to determine if the network was the problem or not (generally not). The same data range is available for latency via OWAMP in v4 and v6.

Establishing a performance measurement mesh identified a number of problems - how do you do scheduling in the presence of scarce resources? Do you want to be able to schedule bi-directionally? Do you have any security for these tests? Also you want to be able to ensure the correct source/target pairs are availalble. When writing the Iperf wrapper for this mesh, Jeff Boote ran into these problems and created the BWCTL tool (see talk at the BoF) to resolve it.

Based on a Performance Measurement Architecture Workshop (Matt Zekauskas NSF grant-funded) identified that most tools can't fit into any one overaraching framework. What is needed is interoperability so that a variety of tools/frameworks can be create that can meet the end-users needs. Some tools will continue and others will be phased out. However, by using a common schema of for test/data request and response so that tools & frameworks can be interoperable. A number of issues (discovery, authentication, authorization, etc.) remain but by working in concert, the tools can eventaully work together and can work together incrementally as well.

The demo was to demonstrate partial path analysis, interoperable measreument domains and frameworks, and create a kernel for a federation of interoperable tools. The partial path ran from University of Hawaii to the Hawaii PoP, to Seattle, to Denver, to Kansas city, to NOC to UIUC desktop adn back.

Next steps: build a federation of measurement infrastructures, deploy across network backbones (ESnet, Ultralight) and research communities. Share best practices among several infrastructures, focusing on filling in missing pieces. Also working on a proposal (with several DoE labs and the University of Delaware) to continue this measurement effort.