News

These days, no computer is an island. High-performance networks enable computer centers to share resources and communicate with users. Grid computing would be impossible without them. Even PC users feel lost without an Internet connection. But high-performance networks can be temperamental creatures, prone to such maladies as traffic jams, routing inefficiencies, and outright link failures. They need constant attention from network system engineers to run at peak performance. But how can you find the source of a problem in a networked system with dozens of sites when all of their administrators are convinced that the problem must be someone else's fault?

The Measurement and Network Analysis group of the National Laboratory for Applied Network Research (NLANR/MNA) assesses the performance of next-generation computer networks - measuring the flow of message traffic, analyzing performance issues, and making all of the data, analyses and tools available to the community - so they can be tuned for maximum end-to-end performance. In particular, network measurements are essential for identifying and locating problems (malfunctions, bottlenecks, inefficiencies, incompatibilities, etc.) in ultrafast research networks and in high-speed international links.

"Although NLANR's measurement and analysis efforts may not be visible to casual users of the Internet, all users ultimately benefit," said Tom Greene, senior program director for the National Science Foundation's Division of Advanced Networking Infrastructure and Research. "Measurement and analysis tools are essential for solving performance issues, and we all need solutions there."

Based at the San Diego Supercomputer Center (SDSC), the NSF-funded NLANR/MNA group has created a network analysis infrastructure to conduct their own research and to support the efforts of outside researchers, systems administrators, and students. With few exceptions, the monitoring devices have been located on high-performance networks operating in the United States. International expansion began cautiously three years ago; now the research infrastructure extends to other continents, and network researchers and system administrators around the world are enthusiastically collaborating with NLANR/MNA researchers, setting up measurement and analysis research projects of their own.

Gathering Data

NLANR/MNA research has two main components. The Passive Measurement and Analysis (PMA) project uses information gleaned from observing network traffic, without interacting with the networks themselves. A low-technology analogy for this project would count envelopes, weigh them, and sort them by postmark to determine how efficiently the U.S Postal Service is handling the mail. The PMA project uses network message packet header traces instead of envelopes, but the principle is the same.

The Active Measurement Project (AMP) performs site-to-site active measurements and analyses, which enable network researchers and engineers to track problems and changes in network performance, by inserting test messages into the networks it studies and observing their progress through the systems. A Postal Service analog of the AMP project would send test letters between many locations, noting their travel times and postmarks. More than 130 AMP monitors are deployed on high-speed research networks in the United States and in other countries, most of them in a "full mesh" configuration - each monitor exchanging messages with every other monitor - to measure message path, round-trip time, and packet loss. In addition to gathering data, the AMP project also analyzes long-term, large-scale trends in message flow patterns.

The monitoring devices for both projects are rack-mounted PCs with high-speed network cards, installed in the network equipment racks of universities and research institutions that participate in the two projects. All participation is voluntary, and both NLANR/MNA and participating sites are meticulous about maintaining the privacy of network users. To use another Postal Service analogy, the monitors count envelopes and examine postmarks and Zip codes, but they do not look at the full addresses or open any of the envelopes or read the letters.

"We go to a lot of trouble to anonymize any address information before we ever see it," said Hans-Werner Braun, founder of NLANR and principal investigator of the Measurement and Network Analysis group. "It's like encrypting the Zip codes with a key that we don't have - it's non-reversible, so we can't tell what the addresses really were, but we use the anonymized versions of the addresses for statistical purposes."

International Networks

Four years ago, researchers in Norway, New Zealand, and Israel borrowed single network monitor units from NLANR/MNA on an experimental basis. International interest in cooperative network research was low-key, until one small event started a landslide.

The NSF created the High Performance International Internet Services (HPIIS) Program in 1997 to provide international researchers access to high-performance networks in the United States and to give U.S. educators access to researchers and international high-performance networks around the globe. HPIIS project participants include AMPATH (which links most of the countries in Central and South America to the U.S.), Euro-Link (between the U.S. and Europe), the TransPAC link (between the U.S. and Japan), NaukaNet (formerly called MIRnet, between the U.S. and Russia), and NSF's STAR TAP and StarLight international access points.

As a result of hosting an NSF-funded HPIIS workshop at SDSC in 2001, NLANR/MNA is now conducting joint research with all four HPIIS grant projects. The SDSC researchers also are collaborating with three additional countries not in the HPIIS program - Australia, Korea, and Canada - that sent representatives to the workshop.

"The entire conference cost less than $6,000," said Ronn Ritke of SDSC, the co-principal investigator and manager of the NLANR/MNA program. "But the return on investment was enormous. Thirty researchers, representing organizations that included about two dozen different countries, met for face-to-face discussions of issues. Suddenly everyone saw the advantages of international cooperation and coordinated research. The enthusiasm was contagious."

New AMP monitors were placed in Australia, Canada, and Korea soon after the conference. The NLANR/MNA group now has AMP machines in Australia, Brazil, Canada, China, Finland, Germany, Holland, Hungary, Israel, Japan, Korea, Mexico, New Zealand, Norway, and Thailand; Russia, Taiwan, and China soon will host monitors as well. They already had placed a PMA monitor in Israel, and soon there will be PMA monitors on networks in Thailand, Germany, and Korea. The STAR TAP and StarLight international links include both AMP and PMA monitors, and many measurement machines are located at individual U.S. sites that serve as international entry/exit points on high-speed networks. And as part of the cooperative effort, all of these projects are sharing their data with other researchers around the world.

The result is that NLANR/MNA is well on the way to establishing a world-wide infrastructure for gathering performance data from high-speed next-generation research networks, vitally important in an age of global communications.

But of equal importance is the intangible network of person-to-person contacts among scientists and engineers in dozens of countries that have been engendered by the program. Researchers and administrators are engaging in joint efforts, exchanging information, and helping to solve one another's problems using a range of different collaboration models, all started as experimental efforts facilitated by NLANR/MNA's activities.

The NLANR/MNA participates in cooperative network measurement research efforts with Australia, Brazil, Canada, China, Finland, Germany, Holland, Hungary, Israel, Japan, Korea, Mexico, New Zealand, Norway, Russia, Taiwan, and Thailand. Click here for a full-size map.

The success of the effort can be judged by another workshop that NLANR/MNA hosted at SDSC less than two years after the HPIIS meeting. PAM2003, the international passive and active network measurement conference, was a three-day event attended by more than 80 researchers and students from Australia, Belgium, Canada, Croatia, France, Germany, Italy, Japan, the Netherlands, Norway, the United Kingdom, and Sweden as well as from the United States. A record number of papers was submitted to the workshop - of 92 submissions, 20 were accepted as full papers and five were accepted as posters.

Exactly 40 of the attendees were graduate students. "One of our major goals was to promote student participation," Ritke said. "It's important to cultivate the next generation of network researchers. A mixed workshop like this one enables us to encourage the best and the brightest students to go into the field and gives them a head start on working with well-established researchers. The students not only attended, they were full participants - 12 of the 20 accepted papers listed students as first authors."

Technology transfer is an important aspect of the collaborations. AMP technology is now being used in parallel national programs to instrument and characterize the performance of the domestic research networks of Australia, Korea, and Taiwan; interest in building similar national infrastructures has been expressed by research groups in Thailand, Brazil, Russia, and China.

And the transfer flows in more than one direction - these cooperating national efforts are being integrated into the infrastructure established and run by NLANR/MNA, with new ideas shared among all participants. For example, presentation topics by international participants at PAM2003 included leading-edge research in active measurement (sending test packets and observing their progress through the network), passive measurement (observing actual traffic on "live" networks), performance metrics (developing indicators to characterize traffic behavior), traffic statistics (to develop models of "real" network traffic), and visualization (finding effective ways to display what is happening in a network).

"The continuous growth of the NLANR Measurement and Network Analysis effort has fostered the growth of wider understanding of the importance of network measurement and the international demand for collaborations with the NLANR team," said Tom Greene of NSF.

-Mike Gannis and Ronn Ritke