The MXQ Cluster Environment
Currently there are about 1700 CPU nodes available in our cluster.
Submitting Jobs
To submit a job you use the mxqsub command:
mxqsub --threads=<n> --memory=<mem size> --runtime=<time> --stdout=<filename> --stderr=<filename> <command + command line options>
Example:
mxqsub --threads=1 --memory=2G --runtime=60m --stdout=date.txt --stderr=/dev/null date +%H:%M
If you do not specify those options, very restrictive default values will be used. More options are available - run mxqsub -h
to see them.
BEWARE: your job will be automatically killed if you try to allocate more memory than specified or your job runs longer than specified.
BEWARE:
it is very well possible that hundreds of your jobs start up at the very same time. Loading the program or data over the network can slow down or even damage the network or hardware. Therefore do NOT access data in your home-dir or project-dir - instead use an appropriate dir in /scratch/cluster
.
Monitoring
You can get information about your queued/running/finished jobs using mxqdump
. See mxqdump -h
for advanced options.
A more readable page with all this information can be accessed here:
http://afk.molgen.mpg.de/mxq/mxq
and the load can be monitored here:
http://twiki.molgen.mpg.de/foswiki/Edv/LoadMXQ
Killing Jobs
If things go wrong and you have to kill jobs use the above web-site to find the corresponding group_id
of your jobs and kill them with:
mxqkill -g <group_id>
At this time it is not possible to kill single jobs individually.
Official Documentation
The official documentation of the queueing system can be found here:
http://twiki.molgen.mpg.de/foswiki/Edv/MarIuX64ClusterHowTo
Further Help
If you need further help please contact: mailto:helpdesk@molgen.mpg.de