[Mauiusers] nodes=x:ppn=1 is overridden by maui to nodes=x/2:ppn=2

Lech Nieroda lnieroda at gmail.com
Thu Feb 19 06:32:27 MST 2009


Dear list,

I'm having a strange problem when running mpi programs through the
maui/torque scheduler: as long as I set ppn to anything larger than 1,
the PBS_NODEFILE provides the correct hosts, i.e. nodes=4:ppn=2 works
exactly as it should - the file contains 4 hosts, each name printed
twice; however with nodes=4:ppn=1 it lists only two hosts with 2 cores
per host.
A quick grep on the logs has showed  that a modification takes place:

--snip--
02/19 13:29:52 MPBSJobModify(39775,Resource_List,Resource,compute-1-1.local:ppn=2+compute-1-2.local:ppn=2)
02/19 13:29:52 MPBSJobModify(39775,Resource_List,Resource,4:ppn=1)
--snap--

Any idea what is going on here?

As to our setup - we have 16 nodes with 8 cpus each, here's an excerpt
from our maui.cfg:

RMPOLLINTERVAL          00:00:30
SERVERMODE              NORMAL
RMCFG[base]             TYPE=PBS
LOGFILE               maui.log
LOGFILEMAXSIZE        10000000
LOGLEVEL              3
QUEUETIMEWEIGHT       1
FSPOLICY              DEDICATEDPS
FSDEPTH               7
FSINTERVAL            86400
FSDECAY               0.80
BACKFILLPOLICY        FIRSTFIT
RESERVATIONPOLICY     CURRENTHIGHEST
NODEALLOCATIONPOLICY  MINRESOURCE
USERCFG[DEFAULT]      FSTARGET=20.0+
FSWEIGHT 10
FSUSERWEIGHT 100
ENFORCERESOURCELIMITS ON
RESOURCELIMITPOLICY[0] MEM:ALWAYS:CANCEL
SRCFG[small] TASKCOUNT=1 RESOURCES=PROCS:4,MEM:16384
SRCFG[small] HOSTLIST=cluster1.local
SRCFG[small] PERIOD=INFINITY
SRCFG[small] TIMELIMIT=1:00:00
SRCFG[small] CLASSLIST=small

Regards,
Lech Nieroda


More information about the mauiusers mailing list