AMBER Archive (2005)

Subject: Re: AMBER: problem with running AMBER with lam-mpi though PBS script

From: Ru-Zhen Li (r.li_at_qmul.ac.uk)
Date: Wed Mar 09 2005 - 19:48:56 CST

just for other people's future reference, the problem was solved by using
mpiexec -machinefile $PBS_NODEFILE -np 8 sander......
mpirun wouldn't work because seems the lam-mpirun haa different format of
the command, it doesn not have the -machinefile option

thank you very much for you help, Karol. ^_^
**********************************************************************
Ms Ru-Zhen Li

0044-020-78826327
r.li_at_qmul.ac.uk
http://www.freewebs.com/lrz/
Materials Department
Queen Mary
University of London
E1 4NS
**********************************************************************
----- Original Message -----
From: "Karol Miaskiewicz" <miaskiew_at_ncifcrf.gov>
To: <amber_at_scripps.edu>
Sent: Wednesday, March 09, 2005 1:57 PM
Subject: Re: AMBER: problem with running AMBER with lam-mpi though PBS
script

>
>> I am having strange errors when i tried to use mpirun (mpiexec) with PBS
>> script,
>>
>> there are 13 nodes on the linux cluster, lam-7.0.3 v8
>> 1)i used lamboot -v lamhosts, with the master and all the slave nodes
>> listed in lamhosts, and lam is runing successfuly
>> 2)i used mpirun -np 8 sander..... and it works ok
>> 3)i used the same command, but put it in a PBS script, and then it
>> complained that
>> it seems that there is no lamd running on the host node13
>> ..........
>>
>> i tried many work arounds including
>> mpirun n1-8 -np 8 sander
>> mpiexec -machinefile hostfile sander (with master and node1-8 listed in
>> hostfile)
>>
>> i also tried to use lower version of lam which is v7
>>
>> but no matter what i tried, the same error always came out, what i am
>> really confused about are:
>> I am not using node13 at all, why it complained about node13,
>> the lam seems to work on node13 since when i do lamboot, it shows node13
>> is successufully done
>> why i can use mpirun through interactive mode, but can't do it through
>> PBS?
>
> The list of hosts to run on is set by PBS. You must use this list and
> not the hostfile that you created yourself. The file with a list of
> hosts for your job is pointed to by the variable PBS_NODEFILE, so your
> mpirun command should look something like this:
>
> mpirun -machinefile $PBS_NODEFILE sander -np 8
>
> Of course, you must also indicate to PBS that you want to use 8
> processors. In the most common case of a cluster of 2-way nodes, this is
> done with:
>
> #PBS -l nodes=4:ppn=2
>
>
> --
> Karol Miaskiewicz [Contractor]
> Advanced Biomedical Computing Center
> NCI-Frederick, PO Box B, Frederick, MD 21702
> miaskiew_at_ncifcrf.gov, phone 301-8465664, fax 301-8465762
>
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu