mpich and cpi

[andreas at lindblom.as]

Hello!
I've just set up my first 8 node AMD cluster using Red Hat 7.2 and mpich. When I
run the sample program cpi the cluster scales fine up to a certain number of
intervals around 2.14*10e9, which takes rougly 8 secs for the entire cluster to
calculate. If I raise the number of intervals to 2.2*10e9 it suddenly takes
"forever" to calculate. I havent checked how long it would actually take but I
aborted it after 70 minutes. 
I've checked the memory (vmstat and top) for possible leaks or a lot of swapping
but that doesn't occur at all. What's going on?
Does anyone know of a good sample program to test scaling on (as cpi doesn't
involve much communication)?

/Andreas


-- 
Get your firstname at lastname email at http://Nameplanet.com/?su