I am trying to run a compiled program that is supposed to be running on multiple processors. But with the same data, sometimes this program runs in parallel and sometimes it won't (with the identical PBS script file!). I am suspecting that something is wrong with some of the compute nodes that won't let it run on parallel (I don't get to choose the compute node I want). How can I troubleshoot if this is a bug in the program or it is problem with the compute node?
As per the sys admin's adivce, I am using ulimit -s 100000, but this don't change anything. Also, this program is not an mpi program (runs only on a single node, with multiple processors).
The code that I run is as follows:
quorum_error_correct_reads -q 68 \
--contaminant=/data004/software/GIF/packages/masurca/2.3.0rc1/bin/../share/adapter.jf \
-m 1 -s 1 -g 1 -a 3 --thread=32 -w 10 -e 3 \
quorum_mer_db.jf aa.renamed.fastq ab.renamed.fastq ac.renamed.fastq ad.renamed.fastq ae.renamed.fastq af.renamed.fastq ag.renamed.fastq \
--no-discard -o pe.cor --verbose
Thanks for any advice you can offer. I will greatly appreciate your help!
PS: I don't have sudo access.
EDIT: I know it is supposed to be using multiple processors because, when I SSH into the node and do top -c I can see (above command) sometimes running like 3200 % CPU (all the time) and sometimes only 100 % CPU all the time. This is the only step involved and there are no other sub-process within this program. Also, I am using HPC, where I submit the job to a compute node, each with 32 procs, 512GB RAM.
Related
I am trying to test a script I have developed locally on an interactive HPC node, and I keep running in this strange issue that mclapply works only on a single core. I see several R processes spawned in htop (as many as the number of the cores), but they all occupy only one core.
Here is how I obtain the interactive node:
srun -n 16 -N 1 -t 5 --pty bash -il
Is there a setting I am missing? How can I make this work? What can I check?
P.S. I just tested and the other programs that rely on forking to do parallel processing (say pigz) are afflicted by the same issue as well. Those that rely on MPI and messaging work properly, it seems.
Yes, you are missing a setting. Try:
srun -N 1 -n 1 -c 16 -t 5 --pty bash -il
The problem is that you are running the parallel commands within a bash shell that is allocated on a single core, so the bash process is spawned on only one of the cores requested by srun.
Otherwise, you can first allocate your resources using salloc and once you obtain them run your actual command. For instance:
salloc -N 1 -n 1 -c 16 -t 5
srun pigz file.ext
I am using Intel MPI and have encountered some confusing behavior when using mpirun in conjunction with slurm.
If I run (in a login node)
mpirun -n 2 python -c "from mpi4py import MPI; print(MPI.COMM_WORLD.Get_rank())"
then I get as output the expected 0 and 1 printed out.
If however I salloc --time=30 --nodes=1 and run the same mpirun from the interactive compute node, I get two 0s printed out instead of the expected 0 and 1.
Then, if I change -n 2 to -n 3 (still in compute node), I get a large error from slurm saying srun: error: PMK_KVS_Barrier task count inconsistent (2 != 1) (plus a load of other stuff), but I am not sure how to explain this either...
Now, based on this OpenMPI page, it seems these kind of operations should be supported at least for OpenMPI:
Specifically, you can launch Open MPI's mpirun in an interactive SLURM allocation (via the salloc command) or you can submit a script to SLURM (via the sbatch command), or you can "directly" launch MPI executables via srun.
Maybe the Intel MPI implementation I was using just doesn't have the same support and is not designed to be used directly in a slurm environment (?), but I am still wondering: what is the underlying nature of mpirun and slurm (salloc) that this is the behavior produced? Why would it print two 0s in the first "case," and what are the inconsistent task counts it talks about in the second "case"?
I have PBS command something like this
#PBS -N marcell_single_cell
#PBS -l nodes=1:ppn=1
#PBS -l walltime=20000:00:00
#PBS -e stderr.log
#PBS -o stdout.log
# Specific the shell types
#PBS -S /bin/bash
# Specific the queue type
#PBS -q dque
#uncomment this if you want to debug the process
#set -vx
cd $PBS_O_WORKDIR
ulimit -s unlimited
NPROCS=`wc -l < $PBS_NODEFILE`
#export PATH=$PBS_O_PATH
echo This job has allocated $NPROCS nodes
echo Cleaning old files...
rm -rf *.png *.plt *.log
echo Cleaning success
/opt/Lib/openmpi-2.1.3/bin/mpirun -np $NPROCS /scratch4/marcell/CellMLSimulator/bin/CellMLSimulator -ionmodel grandi2010 -solverType CVode -irepeat 4 -dt 0.01
gnuplot -p plotting.gnu
It got error something like this, thrown by the PBS error log.
/var/spool/torque/mom_priv/jobs/6265.node01.SC: line 28: gnuplot: command not found
I've already make sure that the path of GNUPlot is already been added to the PATH environment variable.
However, the strange part is, if I interchange the sequence of command, like gnuplot first and then mpirun, there isn't any error. I suspect that some commands after mpirun need some special configs, but I dunno how to do that
Already following this solution, but no avail.
sleep command not found in torque pbs but works in shell
EDITED:
it seems that the before and after mpirun still got error. and this is the which result:
which: no gnuplot in (/opt/intel/composer_xe_2011_sp1.9.293/bin/intel64:/opt/intel/composer_xe_2011_sp1.9.293/bin/intel64:/opt/pgi/linux86-64/9.0-4/bin:/opt/openmpi/bin:/usr/kerberos/bin:/prog/tools/grace/grace/bin:/home/prog/ansys_inc/v121/fluent/bin:/bin:/usr/bin:/opt/intel/composer_xe_2011_sp1.9.293/mpirt/bin/intel64:/opt/intel/composer_xe_2011_sp1.9.293/mpirt/bin/intel64:/scratch7/feber/jdk1.8.0_101:/scratch7/feber/code/apache-maven/bin:/usr/local/bin:/scratch7/cml/bin)
It's strange, since when I try to find the gnuplot, there is one in the /usr/local/bin
ls -l /usr/local/bin/gnuplot
-rwxr-xr-x 1 root root 3262113 Sep 18 2017 /usr/local/bin/gnuplot
moreover, if I run those commands without PBS, it seems executed as I expected:
/scratch4/marcell/CellMLSimulator/bin/CellMLSimulator -ionmodel grandi2010 -solverType CVode -irepeat 4 -dt 0.01
gnuplot -p plotting.gnu
It's very likely that your system has different "login/head nodes" and "compute nodes". This is a commonly used practice in many supercomputing clusters. While you build and launch your application from the head node, it gets executed on one or more compute nodes.
The compute nodes can have different hardware and software compared to the head nodes. In your case, gnuplot is installed only on the head node, as you can see from the different outputs of which gnuplot. To solve this, you have three approaches:
Request the system administrators to install gnuplot on the compute nodes.
Build and install your own version of gnuplot in a file-system accessible from the compute nodes. It could be your home directory or somewhere else depending on your cluster. In general, the filesystem where your application is will be available. In your case, anywhere under /scratch4/marcell/ would probably work.
Run gnuplot on the head node after the MPI jobs finish as a post-processing step. PBS/Torque does not provide a direct way to do this. You'll need to write a separate bash (not PBS) script to do this.
Since the beginning of November, I'm stuck in to run a parallel job in a Linux cluster. I already search A LOT on the internet searching for information but I simply can't progress. When I start to search for parallelism in R using cluster I discovered the Rmpi. It looked quite simple, but now I don't now more what to do. I have a script to send my job:
#PBS -S /bin/bash
#PBS -N ANN_residencial
#PBS -q linux.q
#PBS -l nodes=8:ppn=8
cd $PBS_O_WORKDIR
source /hpc/modulos/bash/R-3.3.0.sh
export LD_LIBRARY_PATH=/hpc/nlopt-2.4.2/lib:$LD_LIBRARY_PATH
export CPPFLAGS='-I/hpc/nlopt-2.4.2/include '$CPPFLAGS
export PKG_CONFIG_PATH=/hpc/nlopt-2.4.2/lib/pkgconfig:$PKG_CONFIG_PATH
# OPENMPI 1.10 + GCC 5.3
source /hpc/modulos/bash/openmpi-1.10-gcc53.sh
mpiexec --mca orte_base_help_aggregate 0 -np 1 -hostfile ${PBS_NODEFILE} /hpc/R-3.3.0/bin/R --slave -f sunhpc_mpi.r
And this is the beginning of my R program:
library(caret)
library(Rmpi)
library(doMPI)
cl <- startMPIcluster()
registerDoMPI(cl)
So here is my questions:
1- Is this way I should initialize the processes (i.e. using starMPIcluster whitout a parameter and using at the command line -np 1)?
2- Why when I use this commands the MPI complains with it's frase?
An MPI process has executed an operation involving a call to the
"fork()" system call to create a child process....
OBS: He said that for all the 64 processes (because there are 8 nodes with 8 cpus and I'm creating 63 processes)
3- Why when I use this commands on a machine of 60 CPU's he just spawn two workers?
Finally, I got it!
To run a parallel program in R using the Rmpi in a cluster you need to configure the job script according to the system. Next on the command line:
mpiexec --mca orte_base_help_aggregate 0 -np 1 -hostfile ${PBS_NODEFILE} /hpc/R-3.3.0/bin/R --slave -f sunhpc_mpi.r
You have to modify to:
mpiexec -np NUM_PROC -hostfile ${PBS_NODEFILE} /hpc/R-3.3.0/bin/R --slave -f sunhpc_mpi.r
On the R code, you must not detail anything 'startMPIcluster()' So, the code will exactly as I wrote above.
I am doing MPI programming on a cluster with 8 nodes and each having a Intel Xeon hexcore processor. I need three processors for my mpi code.
I submit the job using qsub. When I check on which processors the job is running using "qstat -n" it says something like cn004/0*3 .
So does this mean it is running it on only one processor ??
Because it is not speeding up than when I use a single processor(This is when the domain size is the same for both cases)
The script i use for submitting is as follows
#! /bin/bash
#PBS -o logfile.log
#PBS -e errorfile.err
#PBS -l cput=40:00:00
#PBS -lselect=1:ncpus=3:ngpus=3
#PBS -lplace=excl
cat $PBS_NODEFILE
cd $PBS_O_WORKDIR
mpicc -g -W -c -I /usr/local/cuda/include mpi1.c
mpicc -g -W mpi1.o -L /usr/local/cuda/lib64 -lOpenCL
mpirun -np 3 ./a.out
"qstat -n" it says something like cn004/0*3.
Q: So does this mean it is running it on only one processor ??
The short answer is "no". This does not mean that it runs on one processor.
"cn004/0*3" should be interpreted as "The job is allocated three cpu cores. And if we were to number the cores from 0 to 5 then the cores allocated would have numbers 0,1,and 2".
If another job were to run on the node it would receive the next three consecutive numbers "3,4, and 5". In the qstat -n output this would look like "cn004/3*3".
You use the directive place=excl to ensure that other jobs would not get the node, so essentially all the six cores are available.
Now for your second question:
Q: it is not speeding up than when I use a single processor
In order to answer this question we need to know if the algorithm is parallelized correctly.