Dear Carlos and Robert,
thanks a lot for your adivce! In fact, the time I posted was resulted from 32-CPU running. As you pointed out, the data distribution and I/O were the bottle-neck because our cluster architecture, although our computational nodes were interconnected by Myrinet2, the computational nodes and storage server were switched by Gigabit ethernet, moreover many users use the cluster at the same time, so the I/O are usually walled up. It looks like that 8 or 16 cpus are more preferable in my case. thanks again!
sincerely yours, Zhihong
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed May 31 2006 - 00:12:00 PDT
