Wow, according to dave jackson anything > 32 should bail. But I get 2x people using ~64.
64 would be a good # for us.. Also, can anyone recommend the most efficient method to poll for job completion (and hence qstat -f). Sorry Garrick, your perl pbs didn't work right off and I got lazy =) So what we wrote is a phat wrapper around qsub, qstat -f that basically does: Qstat -f id1 id2 id3 ... But somewhere, shell-dependent I *think*, you hit max arg length (someone remind me, is ie argc or strlen(args). So we know we can ask 512 jobs at min (with ids of form 13415153.host.something.yahoo.com) And it works fine. Dave said to turn on some var that made it ok for torque to 'ask' job status if it was stale via heartbeat piggybacks instead of waiting for pbs_mom to push info back (special case pull model). Anyone know the param? I didn't see it in the docs? And in my 4 notepads it's somewhere...something simple like poll_node_ok = true in qmgr? Help? Garrick? Sam Rash [EMAIL PROTECTED] 408-349-7312 vertigosr37 -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Garrick Staples Sent: Wednesday, February 28, 2007 5:48 PM To: [EMAIL PROTECTED]; [email protected] Subject: Re: [torqueusers] max queues On Wed, Feb 28, 2007 at 04:25:01PM -0800, Sam Rash alleged: > This sounds like d?j? vu: we want maui to respect more than the 16 (well 13 > user) queues. It seems that it takes the top (or first defined) 13 queues > and lumps the rest into class ALL. > > (like not only this topic has come up, but _I_ brought it up, ack) > > > > > > I changed this by rebuilding maui with the simple change: > > > > msched-common.h:478:#define MAX_MCLASS 16 > > msched-common.h:479:#define MMAX_CLASS 16 > > > > I changed 16 to 256. I suspect 256 is probably going to be too high. This is all I change in my own maui rpm. perl -pi -e 's/^#define MMAX_JOB .*/#define MMAX_JOB 8192/' include/msched.h perl -pi -e 's/^#define MAX_MJOB .*/#define MAX_MJOB 8192/' include/msched.h perl -pi -e 's/^#define MAX_MCLASS .*/#define MAX_MCLASS 64/' include/msched-common.h _______________________________________________ torqueusers mailing list [EMAIL PROTECTED] http://www.supercluster.org/mailman/listinfo/torqueusers _______________________________________________ mauiusers mailing list [email protected] http://www.supercluster.org/mailman/listinfo/mauiusers
