NLANR/DAST Monthly Update, September 2002

(Unless otherwise noted, all persons mentioned in this report get part or all of their funding from the NLANR/DAST cooperative agreement.) a. Activities related to network tool development for end users

• Research assitant Ajay Tirumala has left the group to work with his faculty advisor. Research assistant Kevin Gibbs has taken on the main responsibility of the Iperf development, guided by Jim Ferguson and Jon Dugan of NCSA's Network Engineering and Research group. Ferguson, Dugan, and Gibbs met to map out the near future of upgrades to the Iperf code, based on feedback from recent Joint Techs meeting and requests from users of the tool. A list of requested features was produced, prioritized, and assigned. An updated version of the Windows version (bringing it even with the various Unix versions) of Iperf was generated and put up for download.
• Steve Engelhardt, Tanya Brethour, Richa Hingle and Feng Qin continue to develop various aspects of the Network Performance Advisor, formerly known informally as the "Framework". Engelhardt and Qin have been focusing on data collection, authorization and security, while Brethour and Hingle are focused on the user interface and analysis engine. Jim Ferguson is providing overall coordination of the Framework effort, as well as being chief liaison with Internet2's testing project.
• Research assistant Yanli Tong has made excellent progress (with the shepherding of John Estabrook and Ferguson) on her project to adapt the NLANR/DAST auto-tuning FTP code for further impact on the community. She has developed a "plug-in" module for the community GridFTP project that incorporates the technologies developed by the NLANR/DAST Autobuf FTP client and server. Tong has worked out issues dealing with data transfer between two remote hosts, which GridFTP supports. Tong and Estabrook are now cleaning the code of the plug-in and are developing a daemon using the ICMP version of the code.

• Mitch Kutzko and Tony Rimovsky have collected comments and suggestions at several venues from users of the newly deployed version of the Multicast Beacon servers. Kutzko and Rimovsky are working on the design of the upcoming new version of the Beacon System, which Kutzko will be doing the coding on in the near future. Talks with a group led by Deb Agarwal (LBL) at the Lawrence Berkeley Laboratory to share in the development of this next generation, more useful and reliable Beacon, are underway. Rimovsky also had discussions at iGrid2002 with a group from Poland which is interested in collaborating on beacon work. A general meeting for interested collaborators will be held late this year on the Access Grid to discuss future directions and the planned roadmap for the beacon system.
• See GridFTP work above also

• Steve Engelhardt has successfully implemented a scheduler for the BIMA project's multi-layered handling of data that arrives via the radio telescopes it employs. During September, Steve continued to transfer knowledge along of utilizing existing middleware infrastructure to this group, who will advance this total project on their own. He is done with any more day-to-day programming, and total handoff of the project will occur when the BIMA group brings in a newly hired staff member to take over from Engelhardt's start. He will, of course, be providing advice and support on a continuing basis.

• The BIMA suite of codes (see above) astronomy community applications.
• No other similar codes in September.

• The initial design for the NextInet web site vehicle for the current awareness service has been completed by Natasha Bulashova and Greg Cole. A prototype has been done and it being reviewed by the team for organizational layout and ease of use. The site will not be rolled out until more content is added to the database, currently being done by Anita Colliate-Howard and Lyn Prowse-Bishop on a temporary basis. All database maintenance applications are complete and in place for distributed maintenance of the web site. The search/match engine for the current awareness service is now in place. An initial thesaurus for describing database content and subscriber interests is in place.
• The NLANR/DAST staff were the official "end-user services" organization of the Demonstrations iGrid2002. DAST was recruited for this effort by Maxine Brown of STARTAP and STARLIGHT. The NLANR/DAST staff started by directly contacting iGrid2002 demo teams, multiple times, in order to offer support. Each demo team was assigned to an NLANR/DAST staff member, and each was contacted a minimum of twice by email and twice by direct phone to offer assistance. Steve Engelhardt lead the effort to inform sites on the required NMI rev.1 software if they wished to use SARA resources in their demo. Direct, active involvement in three demos was identified--two immersive applications, and one high-speed data transfer. New hardware (courtesy of the NSF-funded Web100 project) and GigE connectivity was a feature of the latter application, headed by Les Cottrell at SLAC. Tony Rimovsky, John Towns, and Steve Engelhardt traveled to Amsterdam in support and to participate in iGrid2002. John Estabrook participated in 2 demos (which ran in three different time-slots each) in the NCSA CAVE, and NLANR/DAST and Web100 provided equipment and support here for the third demo. Rimovsky went early to Amsterdam and was on the main networking team at the conference. Towns delivered a tutorial a the conference entitled "Grid Services: Middleware Infrastructure for use of Distributed Resources", and also hosted a Birds of a Feather session on the Global Grid Forum. Engelhardt consulted with several end users on the floor at the conference, and worked (along with many other people) on a single-stream TCP problem that seemed to be endemic at the conference. Through the help of Web100 and John Heffner (Pittsburgh Supercomputing Center and Web100), Engelhardt and Rimovsky were able to achieve 450Mbps on a single TCP stream from Amsterdam to UIUC, but it took 20 minutes plus to ramp up to that speed. Before the Web100 tuning users were getting stuck at around 170Mbps. The problem turned out to be a limitation on the size of the congestion window on the sender side. The 450Mbps figure would likely be unreachable in current Linux without Web100. Rimovsky worked with Net100 tools provided by Jason Lee (LBNL) and was able to sustain 700 Mbit/s between Amsterdam and NCSA in tests. The results of these tests and their impact continue to be examined. Engelhardt also assisted Antony Antony in logging a receiver-side tcpdump of Mac OS X 10.2 (Jaguar) on Engelhardt's PowerBook G4 800 running a single-stream TCP Iperf test from Amsterdam to Chicago. Mac OS X peaked at ~330Mbps, but it reached the figure very quickly.
• The NLANR/DAST web site finished its overhaul and put up the revised site. Mitch Kutzko is led the effort to provide a clearer message and easier to access information for users who visit the site. One new feature on the top page is a directly searchable Advanced Application Database area.

• Ronn Ritke (SDSC, NLANR/MNA), John Towns and Jim Ferguson held a phone call to discuss cooperation issues in the NLANR program. Both sites are eager to cooperate on projects. Initial targets agreed to for NLANR/DAST and NLANR/MNA teamwork are the Gemini project and the Observatory Project, as well as possibly an information collection project, starting with a set number of common network problems, worked out in a systematic fashion. Follow-up phone calls on these are planned.