how to get the performance by using linpack in scyld beowulf?
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Ferdinand Mahr MAHRF at de.ibm.comMon Jul 2 04:11:01 PDT 2001
- Previous message: how to get the performance by using linpack in scyld beowulf?
- Next message: ATIP report on CCGrid 2001. + cfp for next meeting in Berlin
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Hi! I experimented with Linpack this weekend and although I don't understand all parameters, this is what I found out so far: since you have so much memory, increase the N value (the docs tell you how), so all your RAM is used (but avoid swapping!!!). Did you run <mpirun -np 4 ...>? It seemes Linpack only used four of your eight nodes (p=1;q=8 should be OK, but the result only lists p=1 q=4), so try mpirun -np 8! Also check your machines file of MPICH so no node runs more than one Linpack task and another one is idle. Varying NB might help too, it affects the network load. If this works, telling me your results would be interesting. Best regards Ferdinand now,i am using scyld beowulf.when i use linpack to benchmark the performace of my cluster(PIII 933 ,1G Memory,41 G Harddisk,8 nodes),i find that the linpack use the parameter :p=1;q=8 and i get the result: N=3000 NB=60 p=1 q=4 time =14.56 Gflops=1.23 how can i change the parameter?such as p=2;q=4 or p=4;q=2? is the result correct? can somebody give me some advice? thank you!:)
- Previous message: how to get the performance by using linpack in scyld beowulf?
- Next message: ATIP report on CCGrid 2001. + cfp for next meeting in Berlin
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
