HPC on a Linux cluster
 

Hi there,

Recently I set up a linux cluster consisted of 5 nodes 2 Opterons on each node. I did test run with a command like as follows:

"star 1", "star 2", "star 10"

The command "star 1" & "star 2" will run the job on the master node. My finding is that "star 2" execution time is around the half the "star 1". It makes sense. However, "star 10" execution time is nealy the same as the run time by "star 2". Can anybody explain what is happening in my trial run? One more thing. My cluster computer uses giga-bit switch and master and each slave node have enough memory larger than 24 giga bit in total.

Jihwan

Re: HPC on a Linux cluster
 

If the hostnames of your nodes are cluster1 cluster2 cluster3 cluster4 and cluster5

The correct command you should type is

star cluster1,2 cluster2,2 cluster3,2 cluster4,2 cluster5,2

This command can drive the solver run a 10 CPU job.

you can contact your local distritube for more information.

tomosada

Re: HPC on a Linux cluster
 

presumably star1 is your master node? How long is the run, what sort of load ballancing do you have and approximately how many cells are there per cpu?