NEW cluster tarkil with NEW scheduling system PBS Professional available

NEW cluster tarkil with NEW scheduling system PBS Professional available

Dear users,

we would like to introduce new scheduling system PBS Professional (PBS Pro), which is available for testing in a new experimental environment accessible from its own dedicated frontend tarkil.grid.cesnet.cz.

In future, we plan to replace current old Torque scheduling system with new PBS Professional, so we highly recommend you to try this new testing version.

Reasons for changing Torque to PBS Pro:

• current Torque system run into principal problems with decreased throughput and scalability during significant increase in number of jobs (this caused decrease in speed)
• source code of commercial PBS Pro, that we left due to financial reasons, became available for free
• PBS Pro scheduling system fulfills almost every of our demands - supports scalability and offers compatibility with others PBS Pro systems as well as other so far unsupported functionalities

Differences of PBS Pro compared to Torque:

• PBS Pro comes with new qsub syntax "select" which differs in major aspects from old Torque syntax (details)
• offers highly advanced opportunities for specification of required resources
• better support for planning of parallel jobs and also supports Docker containers
• Please note, PBS Pro environment is not supported at that moment in PBSmon web application, we work on integration. Please use qstat on frontend in the meantime.

Using the PBS Pro in MetaCentrum experimental environment:

• PBS Pro server: arien-pro.ics.muni.cz (server is not directly accessible)
  
• Dedicated frontend: tarkil.grid.cesnet.cz
  
• Available resources: NEW cluster tarkil[1-16].grid.cesnet.cz, 16 nodes (384 CPUs), each:
  • CPU: 2x 12-core Intel Xeon E5-2650v4 (2.20GHz)
  • RAM: 128GB
  • disk: 2x 4TB
  • property: CESNET
  • SPECfp2006 performance of each node: 790 (32.9 per core)
• Experimental environment supports temporarily only short jobs with max 2 days walltime
• qsub syntax (Request for job with two processors on each of 3 chunks (nodes), 1 GB of RAM, 1 GB of local scratch, 1 hour walltime, 1 licence for Matlab):

  • qsub -l select=3:ncpus=2:mem=1gb:scratch_local=1gb -l walltime=1:00:00 -l matlab=1 skript.sh

Documentation for new scheduling system PBS Professional can be found at wiki https://wiki.metacentrum.cz/wiki/PBS_Professional

Comments and questions please addrress to RT: meta@cesnet.cz

However, we believe that new possibilities introduced with PBS Pro will help users to better specify their jobs within MetaCentrum and therefore gain significant results in their research more easily.