next up previous contents
Next: sub clustering using PBS Up: Clusters of clusters Previous: allocating groups of nodes   Contents

sub clustering using multi-user mode

Additionally you may subcluster the multi-user nodes by editing the files in the /opt/score/etc/ndconf/ directory corresponding to the node numbers. The contents of these files look like this (e.g for node 3):
#define HOST_3 cpugen=i686 speed=501
#define ETH_3 00:50:56:E6:5C:32
The first line gives a cpu name and a nominal speed. You can use an arbitrary name for the cpu and an arbitrary number for the speed. For example you can give a group of cpus the name ``bigmem'' or ``smallmem'' and then you can specify them at runtime using:

[nrcb@vmserve mpi]$ scrun -nodes=2..bigmem ./mpitest
will run mpitest on the bigmem nodes,
[nrcb@vmserve mpi]$ scrun -nodes=2..smallmem ./mpitest
will run mpitest on the smallmem nodes,
[nrcb@vmserve mpi]$ scrun -nodes=2 ./mpitest
on any nodes. See the online manpage for scrun for more details.