Joint CDF/D0/CD Run II Computing Meeting of May 12, 1998

ZOOM mini-workshop report -- Mark Fischler summarized his view of the workshop

HPSS Tests -- Krzysztof Genser was joined by Jeff Deutsch on the speaker phone. Together they discussed a series of Tests they were performing, the configuration and some of the results.

• The Nameserver tests involved doing an "ls" of the whole catalog/namespace. This consisted of 300,000 files. However, the number of 'queries' - i.e. the lookup of one file or one directory in the namespace - needed to traverse the entire catalog is considerably more than 300,000. Each node of the directory tree may be looked up several times in the traversal. The resulting times presesnted are for one 'query' in this process.
• Stress tests consisted of writing to tapes, using a disk buffer, from 5 client nodes, all of which were fiber-channel connected to the system. The tests were actually performed with real users still on the system. The clients included 2 F50 machines, 1 570 machine and 1 4 processor challenge.
• The next step is to replace the mover/tape and disk 580 machines with the more powerful F50s which we have ready. These also have a newer FDDI card. There was considerable discussion and speculation about what results this would produce, where the bottleneck would be, and what the 'scaleable' unit in the system really is. Some thought the tape I/O would be the bottleneck, unable to exceed more than about 6 MB/sec. Others speculated that the disks would become the bottleneck with their limits being approx 5MB/sec for write and 7 MB/sec for read. Another concern was that the CPU on the movers would actually max out, even with the more powerful processors. The real limitation of the fiber-channel network (assumed about 20MB/sec for quarter speed fiber channel) was also mentioned. The HPPC people went off to delve deeper into these numbers and to formulate a better parameterized model of how the whole system works. Estimated that it will take about a week to replace the 580s with F50s.

PASFRG -- Irwin Gaines reported that there is a new , almost final report posted on the PASFRG Home page . The group is still arguing about which requirements are mandatory and which desirable. A joint meeting with the PASSUMA group is planned for next Monday, after which the documents will be closed out. Comments are welcomed.

PASSUMA -- Jeff Kallenbach (Jeffk@fnal.gov) who has replaced Gene Oleynik as leader of the PASSUMA working group - presented the final report.

Config Management tools - UPS/UPD status - Eileen's transparencies