<DIV>Thanx Mark & chris for flourshing my knowledge, What i want to do is that if some one executes a parallel program of 2.5 hrs execution time while this is executing(on all nodes), if i want to run a 2 minute parallel program on the same machines how will cluster (made with MPICH as an MPI) manage this.</DIV>
<DIV>Actually i have created a GUI environment for the execution of these parallel programs now what provision to make in that environment for the above problem............</DIV>
<DIV>Akhtar<BR><BR><B><I>ark Hahn <firstname.lastname@example.org></I></B> wrote:</DIV>
<BLOCKQUOTE class=replbq style="PADDING-LEFT: 5px; MARGIN-LEFT: 5px; BORDER-LEFT: #1010ff 2px solid">> >>i have made a 4 node cluster now how to restrict CPU usage of a particular<BR>> >>node i.e. if i want node 2's CPU not to exceed 70% usage......... <BR><BR>Chris is right. I read this very strictly (CPU meaning a particular <BR>physical processor). but the author probably just meant "how do I configure<BR>my queueing system to keep the load at under .7". the answer is that you <BR>read the fine documentation of your queueing system, of course!<BR><BR>this is actually somewhat relevant, since anyone who talks about<BR>loadaverage is not doing parallel-type HPC clustering. any parallel <BR>program which has any collective operations is quite sensitive to <BR>preemption, and so cannot tolerate competition for access to CPUs.<BR>this sensitivity an very steep function of the size of jobs (ncpus)<BR>and the frequency of collectives.<BR><BR>people who are mostly used
to serial farms find this a hard concept<BR>to grasp, somehow. we've actually thought of dynamically designating<BR>some nodes to be parallel-only, since serial jobs can tolerate CPU<BR>overcommittment fairly nicely.<BR><BR>interestingly, there are three main categories of jobs I see that would <BR>sometimes benefit from overcommittment:<BR>- serial jobs. they don't actually benefit, but can tolerate <BR>overcommittment without serious degredation. and serial people<BR>sometimes like to see that they've got jobs at least started, <BR>even if they're not getting 100% CPU.<BR><BR>- gaussian jobs, even running in sysv-marallel mode. here the <BR>issue is that gaussian is often waiting for IO to its obscenely<BR>large scratch files. it's not unusual to see a 4-way gaussian<BR>job *average* around 26% utilization of a 4-way node.<BR><BR>- interactive or other IO-intensive jobs. we have some people <BR>who want to run interactive matlab sessions, or other programs<BR>that have guis.
mostly, they're OK if they just do this on a<BR>head node, but that sometimes causes trouble, too.<BR><BR>regards, mark hahn.<BR><BR>_______________________________________________<BR>Beowulf mailing list, Beowulf@beowulf.org<BR>To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf<BR></BLOCKQUOTE><p>
<hr size=1>Do you Yahoo!?<br>
<a href="http://us.rd.yahoo.com/mail_us/taglines/10/*http://promotions.yahoo.com/new_mail/static/efficiency.html">New and Improved Yahoo! Mail</a> - Send 10MB messages!