70 likes | 164 Views
CC-J: Progress, Prospects and PBS. Shin’ya Sawada (KEK) For CCJ-WG. Current Configuration. SUN E450 2 x 400MHz CPUs 0.3GB work disk Linux farms (Alta Cluster) 32 CPUs (16 nodes) with memory of 128MB/cpu Pentium II 450MHz: 18.5 SPECint95/cpu HPSS 100TB tape robot 5 SP2 servers
E N D
CC-J: Progress, Prospects and PBS Shin’ya Sawada (KEK) For CCJ-WG PHENIX Comp. Mtg.
Current Configuration • SUN E450 • 2 x 400MHz CPUs • 0.3GB work disk • Linux farms (Alta Cluster) • 32 CPUs (16 nodes) with memory of 128MB/cpu • Pentium II 450MHz: 18.5 SPECint95/cpu • HPSS • 100TB tape robot • 5 SP2 servers • Network: Gigabit ethernet and HiPPI PHENIX Comp. Mtg.
Performance Test • PHENIX software • Pftp between Linux nodes and HPSS • ~50MB/s total with ~100% CPU usage of disk serers PHENIX Comp. Mtg.
AFS • Arla vs Transarc AFS client • Arla is still very unstable from test results at CC-J. • Transarc AFS 3.5 (patch2) client test • RH 5.2, kernel 2.2.10 SMP: OK • RH 5.2, kernel 2.2.10 SMP with NFSv3: NG with a CVS error • RH 6.0, kernel 2.2.10 SMP: OK • RH 6.0 kernel, 2.2.10 SMP with NFSv3: NG with a CVS error PHENIX Comp. Mtg.
Pbs_server On ccjsun Pbs_sched Pbs_mom On each node On each node PBS • http://pbs.mrj.com • Very flexible scheduling policies • Very flexible queue setting • Quick communication with the developing group • ‘Interactive’ batch job available • Current queues: see table • 1 job / cpu • Negative priority leads leads to jobs more than 1 job/cpu. PHENIX Comp. Mtg.
Prospects • Installation of new hardware • SUN E450 server with four 400MHz CPUs and 1GB memory by the end of October • Dedicated as a file server • Login by general users will be prohibited. • 1.6 TB disk by the end of October • Served as working space for users • Will have two 800GB partitions. • Alta Cluster boxes by the end of October • 16 nodes = 32 CPUs with RH5.2 • Pentium III 600MHz => 24 SPECint95/cpu • Total of 1360 SPECint95 will be available by the end of October. PHENIX Comp. Mtg.
Schedule • Hardware installation/tuning: - Nov 5 • Stress test (MDCJ3?): Nov 8 – Dec 10 • Hardware move: Dec 13 – Jan 14 • Final test/tuning: Jan 17 – Jan 31? • We are going to have a test period (MDCJ3?) in November and December for about one month. If the schedule of RCF/PHENIX MDC3(?) meets ours, CC-J may generate a part of simulation data for it. PHENIX Comp. Mtg.