In my insitute, I have several jobs each of which has 21s' computation time and 1M input/output data. Now I submit the jobs through multiple clusters(each cluster has 32 jobs) to Windows based dual core condor machines. the speedup does not increase after the number of condor machines reaches 12 or 16. Do you think the submit machine is a choke point in this case? We tried to use two submit machines, but notice that Condor always schedule jobs from one submitter first, then another, so cannot schedule from two submit machines fairly. Any idea?
Thanks for any suggestion here.
Insert movie times and more without leaving Hotmail®. See how.