Memory intensive Jobs

Jobs requiring more than 2 GB per process have several alternatives. They might use

1. a node with more memory,
2. a parallel environment with fewer processes per node,
3. increased virtual memory, or
4. swap space (still experimental).

The free space on the nodes of a job can be listed, e.g. by

qstat -F mem_free|grep -B 2 <job-id>

there are 2 nodes with 64 cores and 256 GB of memory each which are accessible via the highmem.q. Specify in your job script:

#$ -q highmem.q

several nodes on the VSC-2 have 64 GB or more instead of the 32 GB of the standard nodes. To use one of these nodes, just add '-l mem_free=50G' or '-l mem_free=100G' to qsub.

See also memory usage of running jobs!

Jobs using more than 2 GB per process can be executed in one of the parallel environments

• mpich : 2 GB per process
• mpich8: 4 GB per process
• mpich4: 8 GB per process
• mpich2: 16 GB per process
• mpich1: 32 GB per process

Please keep in mind that on VSC-2 jobs are node-exclusive and therefore your contingent of CPU-hours will be computed by full nodes and are therefore significantly more expensive. For example,

qsub -pe mpich2 64 my_job.sh

requests 64 cores, each with 16 GB: this is fulfilled by 32 nodes: each hour of this job costs 32*16=512 CPU-hours.

#$ -pe mpich2 4
mpirun -machinefile $TMPDIR/machines -np $NSLOTS_REDUCED -ppn 2 ./executable

The parallel environment option (pe) includes the type of the requested environment (mpich2) and the number of processes (4). The number of processes per node (ppn) is 2. The machine file and the variable NSLOTS_REDUCED are automatically modified such that they reflect the number of slots requested (4), whereas the variable NSLOTS is set to the number of physical cores allocated in the queueing system. This latter number NSLOTS corresponds to the cost calculation in the previous paragraph: NSLOTS = 4 processes * 8 cores = 32 cores = 2 nodes.

#$ -pe mpich2 4
export OMP_NUM_THREADS=8
mpirun -machinefile $TMPDIR/machines -np $NSLOTS_REDUCED -ppn 2 ./executable

In this example, each process allocates 8 cores and starts 8 threads.

Some programs allocate more memory than they use. This was especially true in old FORTRAN 77 programs, which had to decide at compile time how much memory will be used. These programs are allowed to allocate 50% more memory than available by '#$ -l overcommit_mem=true' in the job script. Unfortunately it might happen that the whole node crashes, reboots and leaves the queuing system, so please use this option wisely!

A novel feature of the VSC-2 is remote swap space (implemented using 'SCSI RDMA Protocol', SRP), which is used by specifying '#$ -l swapsize_GB=32' or a multiple of 32. Each node of a job gets the same amount of swap space.