Report on IEPM PPDG efforts for PPDG for the quarter July-September 2003

Report prepared by Les Cottrell, October 7, 2003

Collaboration with IEPM, Network Performance Monitoring

Web/Grid Services

To enable easier ways to search through the IEPM-BW data we built several tools to push the data into an Oracle back-end database. We have been working with the UCL developers (Yee-Ting Lee anmd Paul Meallor) to discuss and understand the OGSI schema. Following this Warren worked on porting the MAGGIE Web service to an OGSI type grid service using the perl module.

We co-authored and had accepted a paper from the GGF/NMWG on Enabling Network Measurement Portability through a Hierarchy of Characteristics.

Data Presentation and Visualization

We have developed some new web based methods to visualize the traceroutes that we run every 10 minutes to all the remote IEMP-BW hosts. This provides a table of time of day versus remote host showing the route numbers and identifying changes. It also provides the ability to select hosts and times and show graphical topologies of the routes. See for example Traceroute Analysis for 10/07/2003.

PingER

There is increasing interest from the HENP and other scientific communities to understand and do something about the Digital Divide, i.e. the difference in Internet performance to developing and developed countries. Since 10-20% of HEP collaborators on the major experiments come from countries in developing nations, this is very important to HENP.

In order to provide a more balanced view seen from Europe, following a recommendation from the ICFA/SCIC we added ICTP/eJDS sites to CERN's PingER monitoring.

In preparation for a series of presentations, talks and papers this winter at the WSIS, RSIS, ICTP/eJDS etc. we energetically worked on cleaning up and replacing broken links, adding new countries and more nodes in existing countries. New countries include: Phillippines, Cuba, Tajikistan, Turkmenistan, Khirgizstan, Namibia. Hosts have also been clean out and added in Macedonia, Serbia/Montenegro, Belarus, Turkey, Armenia, Mexico, Azerbaijan, South Africa, Saudi Arabia. We are also working on getting a monitoring host in Iran.

Les gave a presentation at the ICFA/SCIC meeting on The Digital Divide and PingER
Warren gave a presentation on Quantifying the Global Digital Divide at the Internet2 Joint Techs meeting in Kansas.
Warren gave a presentation at the SciDAC project review on Internet end-to-end Performance Monitoring (IEPM) and the PingER project.
Les gave a presentation to the RIPE 46 meeting in Amsterdam on PingER: a lightweight active end-to-end network monitoring tool/project.
For the Abdus Salam International Center for Theoretical Physics (ICTP) Electronic Journals Distribution System (eJDS) Second Open Round Table on Developing Countries Access to Scientific Knowledge: Quantifying the Digital Divide meeting, we submitted 3 papers:
- Pinger History and Methodology
- Measuring the Digital Divide with PingER
- Internet Performance to Africa

Les is on the program committee for the eJDS/ICTP sponsored Open Round Table on Quantifying the Digital Divide in Trieste in October 2003.

High speed networking

The Internet2 Land Speed Record is now in the Guinness Book of records.

We also submitted and had publised an article on High Speeds are Good for Guinness for the DoE Pulse magazine.

We co-authored a paper with LANL and Caltech on Optimizing 10-Gigabit Ethernet in Networks of Workstations, Clusters, and Grids: A Case Study.

For the SC2003 Bandwidth Challenge we submitted a proposal on "Bandwidth Lust": Distributed Particle Physics Analysis using Ultra high speed TCP on the GriD with Caltech. We are now working with Cisco, Level(3), Stanford, QWest and CENIC to get a 10Gbits/s from SLAC to the SC2003 show in Phoenix.

We have been evaluating and providing feedback to an interesting commercial passive monitoring tool from Network Physics. We hope to demonstrate it at SC2003.

We agreed to co-chair the 2nd Protocols for Fast Long Distance Networks to be held in Chicago in February 2004.

Advanced TCP STack Evaluation

With the emergence of many new advanced TCP (FAST, HS, Scalable, LP, H, Bic, Westwood+ ...) stacks that are trying to provide high-, fair-, responsive-performance on fast long-distance paths without needing to resort to using multiple parallel streams, it is important to evaluate and compare these stacks on high-speed production links to understand their domains of applicability etc. We (Les and Hadrien Bullot a summer intern) have put together tools to enable evaluating and comparing the performance of new TCP stacks. The tools included two new fast GE connected dual 3GHz cpu Dell 2650s, a modification to iperf to provide sinusoidally varying UDP streams, plus automated measurement scheduling scripts. We also worked with teh stack developers to install, configure FAST-TCP, HS-TCP, HSTCP-LP, Bic-TCP, H-TCP and Westwood+. With these stacks we could saturate networks, so we worked carefully with administrators at several sites (e.g. CERN, UFL, Manchester, UIUC, Caltech and UMich) to get access to 2 fast hosts (the second to use for cross-traffic generation) at each site and to carefully schedule our tests.

To facilitate the installation and testing of HSTCP-LP (a fusion of HS-TCP and TCP-LP) Aleksander Kuzmanovic, a student from Rice University, spent 3 fruitful weeks at SLAC assisting with the installation and providing fixes, enhancements as we tested and found out more about the stack on our fast, production links.

Proposals

Warren gave a presentation on MAGGIE at the Measurement SIG at the ESCC/Internet2 Joint TEchs meeting.

Les submitted a White Paper on MAGGIE to the DoE/MICS office.

IPv6

Following renewed interest in IPv6 to support large numbers of IP addresses that will be needed as more and more objects (phones, soldiers, mines, personal devices etc.) are networked, we revived our IPv6 monitoring effort. Paola prepared a presentation on IPv6 for the Internet2 Joint Techs Workshop in August at Lawrence, Kansas.

Trouble Shooting

We added two Network Problem Case studies. Both had to do woth bad routes resulting in factors of 2 to 4 reductions in throughput. One was to Southern Crossing in Atlanta, the other to Caltech.

Warren worked with the Internet2 Backbone Measurement Infrastructure (BMI) group to serve the one-way OWAMP data to the Advisor tool.