Building the HKU GD300 Cluster: Linpack Test (Oct 17, 2002)
Date Descriptions
Oct 7, 2002 Sit has run the Linpack using 256 PCs of the Gideon 300 Cluster. The 256 PCs can achieve 290 Gflops, which can be ranked 169th according to the old TOP500 list (Jun 2002).
Oct 17, 2002 We can achieve Linpack performance 304 Gflops using 256 nodes. 304 is about 29.6% efficiency for 256 nodes. In this test, Intel compiler and LAM/MPI were used. Each test runs for slightly more than a hour, with a problem size of 119808. Anthony reported that when he left HW305, the measured room temperature was 36 degree celsius (only one air con was working).
Oct 19, 2002 HKU School Open Day. The PC Cluster Laboratory was opened for the public. Parallel raytracing and JESSICA2 were shown.
Oct 21, 2002 Linpack benchmark on the whole Gideon cluster were started.
Oct 26, 2002 N1/2 measurement: GD193 went dead completely. It can't even boot up the BIOS. Besides the broken machine, we found that one machine has the memory error. When we perform the Linpack test, the benchmark result showed that there are errors in the calculated results. After investigation, Sit identified the cause was coming from the memory of a particular machine.
Oct 28, 2002 (11:19) Sit has identified the problem that broke the machine. It is due to the memory failure. There are total 2 machines have been identified with memory problem.
Oct 28, 2002 (13:30) One more machine was dead. Again, due to the memory failure problem. So altogether we have TRHEE broken machines with memory error. Memory bars were replaced.
Oct 28, 2002 (18:00) Final Linpack result: 355.5 GFLOPS, N=129600 (Full report)
Photos
BACK >> Photo Collection
Systems Research Group Department of Computer Science The University of Hong Kong