[Mauiusers] Maui double dipping to the AM

Garrick Staples garrick at usc.edu
Wed Sep 1 19:13:06 MDT 2004


On Mon, Aug 30, 2004 at 11:36:12AM -0600, jacksond at supercluster.org alleged:
> Garrick,
> 
>   With the help of CHPC we were able to track and correct what we believe 
> is the source of the problem.  This solution is in testing now and should 
> be rolled into to the latest Maui release this week.  Would you like us to 
> directly email you when this release is available?

The new snapshot doesn't seem to be working correctly.  No matter how many
nodes I request, I only get 1.

Given:
$ qsub -I -l nodes=4
qsub: waiting for job 3913.hpc-master.usc.edu to start

Here's some log snippets:
09/01 17:56:00 MPBSJobLoad(3913,3913.hpc-master.usc.edu,J,TaskList,0)
09/01 17:56:00 MReqCreate(3913,SrcRQ,DstRQ,DoCreate)
09/01 17:56:00 INFO:     processing node request line '4'
09/01 17:56:00 INFO:     188 feasible tasks found for job 3913:0 in partition DEFAULT (1 Needed)
09/01 17:56:00 INFO:     located job '3913' in MBFBestFit (size: 1 duration: 1800)
09/01 17:56:00 INFO:     188 feasible tasks found for job 3913:0 in partition DEFAULT (1 Needed)
09/01 17:56:00 INFO:     tasks located for job 3913:  4 of 1 required (184 feasible)
09/01 17:56:00 MAMQBDoCommand(hpc,0,COMMAND=make_reservation AUTH=maui MACHINE=hpc ACCOUNT=hpccadm USER=garrick W CLIMIT=1800 PROCCOUNT=1 QOS=DEFAULT CLASS=[DEFAULT] NODETYPE=DEFAULT TYPE=maui JOBID=3913 JOBTYPE=job NODES=1,E,SC,Response)


> 
> Dave
> 
> On Sat, 28 Aug 2004, Garrick Staples wrote:
> 
> >On Fri, Aug 27, 2004 at 04:30:39PM -0700, Garrick Staples alleged:
> >>maui-3.2.6-p6.1079990700
> >>torque-1.0.1-0.p6
> >>qbank-2.11.0
> >>
> >>It seems that Maui is overcharging some users.  I haven't figured out the
> >>
> >>From qbank's bnklog:
> >>REQUEST=COMMAND=make_reservation AUTH=maui MACHINE=hpc ACCOUNT=lc_seb 
> >>USER=pap WCLIMIT=36000 PROCCOUNT=10 QOS=DEFAULT CLASS=[DEFAULT] 
> >>NODETYPE=DEFAULT TYPE=maui JOBID=2548 JOBTYPE=job NODES=10
> >>REQUEST=COMMAND=remove_reservation AUTH=maui ACCOUNT=lc_seb JOBID=2548
> >>REQUEST=COMMAND=make_withdrawal AUTH=maui MACHINE=hpc ACCOUNT=lc_seb 
> >>USER=pap WCTIME=852 PROCCOUNT=20 PROCCRATE=1.00 QOS=DEFAULT 
> >>CLASS=[DEFAULT] NODETYPE= JOBID=2548 JOBTYPE=job NODES=10
> >>
> >>See how all of the withdrawal's PROCCOUNT are doubled?
> >>
> >>I don't have maui's logs since they are currently rotating too fast.
> >>
> >
> >Now that I've slept on it, I guess the doubling PROCCOUNT might make sense 
> >if
> >Maui is accounting for the second processor on each node that can't be 
> >assigned
> >to another job (these are all dual proc nodes, and we dedicate nodes to 
> >jobs).
> >
> >So I guess the question is... why is Maui withdrawing so often?
> >
> >

-- 
Garrick Staples, Linux/HPCC Administrator
University of Southern California
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/mauiusers/attachments/20040901/8275f0f2/attachment.bin


More information about the mauiusers mailing list