<HTML dir=ltr><HEAD><TITLE>Re: [Mauiusers] mpi job on multi-core nodes, fails to run on multiplenodes</TITLE>
<META http-equiv=Content-Type content="text/html; charset=unicode">
<META content="MSHTML 6.00.2900.3429" name=GENERATOR></HEAD>
<BODY>
<DIV id=idOWAReplyText84342 dir=ltr>
<DIV dir=ltr><FONT face=Arial color=#000000 size=2>do the mpds start and exit properly when you do it this way? i've always started it from within my job file -- i do something like:</FONT></DIV>
<DIV dir=ltr><FONT face=Arial size=2></FONT> </DIV>
<DIV dir=ltr><FONT face=Arial size=2>#PBS -l nodes=4:ppn=4</FONT></DIV>
<DIV dir=ltr><FONT face=Arial size=2>...</FONT></DIV>
<DIV dir=ltr><FONT face=Arial size=2>mpdboot -n 4 -f $PBS_NODEFILE</FONT></DIV>
<DIV dir=ltr><FONT face=Arial size=2>mpiexec ...</FONT></DIV>
<DIV dir=ltr><FONT face=Arial size=2>mpdallexit</FONT></DIV>
<DIV dir=ltr><FONT face=Arial size=2></FONT> </DIV>
<DIV dir=ltr><FONT face=Arial size=2>it's been a while since i've used an MPI with mpds, but i thought it just needed one mpd per host (not one per processor), right? that's why i start 4 here...</FONT></DIV>
<DIV dir=ltr><FONT face=Arial size=2></FONT> </DIV>
<DIV dir=ltr><FONT face=Arial size=2>--Joe</FONT></DIV></DIV>
<DIV dir=ltr><BR>
<HR tabIndex=-1>
<FONT face=Tahoma size=2><B>From:</B> mauiusers-bounces@supercluster.org on behalf of Mary Ellen Fitzpatrick<BR><B>Sent:</B> Mon 11/3/2008 9:43 AM<BR><B>To:</B> Joseph Hargitai; mauiusers@supercluster.org; Mary Ellen Fitzpatrick<BR><B>Subject:</B> Re: [Mauiusers] mpi job on multi-core nodes, fails to run on multiplenodes<BR></FONT><BR></DIV>
<DIV>
<P><FONT size=2>My pbs script<BR>-snippet<BR># Request 4 processor/node<BR>#PBS -l nodes=4:ppn=4<BR><BR># How many procs do I have?<BR>NP=$(wc -l $PBS_NODEFILE | awk '{print $1}')<BR>echo Number of processors is $NP<BR><BR>mpiexec -n $NP dock6.mpi -i dock.in -o dock.out &> dock.log<BR><BR>My output file list "Number of processors is 16" which is what I request<BR><BR><BR>I start all of the mpd on all of the nodes from the head node with the<BR>following command:<BR>mpdboot -n 47 -f /etc/mpd.hosts<BR><BR>Should I be starting the mpd daemon from within my pbs script?<BR><BR>/etc/mpd.hosts is on every compute node and lists the following:<BR>node1045:4<BR>node1046:4<BR>node1047:4<BR>node1048:4<BR><BR>My $PBS_NODEFILE has the following:<BR>node1045 np=4 lomem spartans<BR>node1046 np=4 lomem spartans<BR>node1047 np=4 lomem spartans<BR>node1048 np=4 lomem spartans<BR><BR>Thanks<BR>Mary Ellen<BR><BR>Joseph Hargitai wrote:<BR>> What is in the pbs script? In most cases you need a -hostfile $PBS_NODEFILE entry, otherwise you get all processes piled on one node ie. the job does not know of other hosts than the one it landed on.<BR>><BR>><BR>> j<BR>><BR>> ----- Original Message -----<BR>> From: Mary Ellen Fitzpatrick <mfitzpat@bu.edu><BR>> Date: Friday, October 31, 2008 11:45 am<BR>> Subject: [Mauiusers] mpi job on multi-core nodes, fails to run on multiple nodes<BR>><BR>> <BR>>> Hi,<BR>>> Trying to figure out if this is an maui or mpi issue. I have 48<BR>>> (dual-dual core cpus) linux cluster. I have torque-2.3.3,<BR>>> maui-3.2.6p19, mpich2-1.07 installed. Not sure if I have maui<BR>>> configured correctly. What I want to do is submit an mpi job that<BR>>> runs<BR>>> one process/per node requests all 4 cores on the node and I want to<BR>>> submit this one process to 4 nodes.<BR>>><BR>>> If I request in my pbs script 1 node with 4 processors, then it works<BR>>><BR>>> fine: #PBS -l nodes=1:ppn=4, everything runs on one node 4 cpus, mpi<BR>>><BR>>> output says everything ran perfect.<BR>>><BR>>> If I request in my pbs script 4 nodes with 4 processors then it fails:<BR>>><BR>>> #PBS -l nodes=4:ppn=4, my epilogue/proloque output file say the job<BR>>> ran<BR>>> on 4 nodes and requests 16 processors.<BR>>><BR>>> But my mpi output file says it crashed:<BR>>> --snippet--<BR>>> Initializing MPI Routines...<BR>>> Initializing MPI Routines...<BR>>> Initializing MPI Routines...<BR>>> Initializing MPI Routines...<BR>>> rank 15 in job 29 node1047_40014 caused collective abort of all ranks<BR>>> exit status of rank 15: killed by signal 9<BR>>> rank 13 in job 29 node1047_40014 caused collective abort of all ranks<BR>>> exit status of rank 13: killed by signal 9<BR>>> rank 12 in job 29 node1047_40014 caused collective abort of all ranks<BR>>> exit status of rank 12: return code 0<BR>>> --snippet--<BR>>><BR>>> Maui.cfg pertinent info:<BR>>> JOBPRIOACCRUALPOLOCY ALWAYS # accrue priority as soon as job is submitted<BR>>> JOBNODEMATCHPOLICY EXACTNODE<BR>>> NODEALLOCATIONPOLICY MINRESOURCE<BR>>> NODEACCESSPOLICY SHARED<BR>>><BR>>> /var/spool/torque/server_priv/nodes file<BR>>> node1048 np=4<BR>>> etc<BR>>><BR>>> torque queue info:<BR>>> set queue spartans queue_type = Execution<BR>>> set queue spartans resources_default.neednodes = spartans<BR>>> set queue spartans resources_default.nodes = 1<BR>>> set queue spartans enabled = True<BR>>> set queue spartans started = True<BR>>><BR>>> Anyone know why my mpi job is crashing? Or if this is an maui/torque<BR>>> or<BR>>> mpi issue?<BR>>><BR>>> --<BR>>><BR>>> Thanks<BR>>> Mary Ellen<BR>>><BR>>> _______________________________________________<BR>>> mauiusers mailing list<BR>>> mauiusers@supercluster.org<BR>>> <A href="http://www.supercluster.org/mailman/listinfo/mauiusers">http://www.supercluster.org/mailman/listinfo/mauiusers</A><BR>>> <BR>><BR>> <BR><BR>--<BR>Thanks<BR>Mary Ellen<BR><BR>_______________________________________________<BR>mauiusers mailing list<BR>mauiusers@supercluster.org<BR><A href="http://www.supercluster.org/mailman/listinfo/mauiusers">http://www.supercluster.org/mailman/listinfo/mauiusers</A><BR></FONT></P></DIV></BODY></HTML>