HP XC System Software Administration Guide Version 2.1

n
m
l
inner-iteration-value
outer-iteration -value
cpus-per-node
cut-off
detailed-output-flag
The following table describes these parameters and indicates their default values.
Parameter Default Value Description
n
1000
Thesizeofthematrix
m
1000
Thesizeofthematrix
l
1000
Thesizeofthematrix
inner-iteration-
value
6
The number of inner iterations (timed)
outer-iteration
-value
10
The number of outer iterations (instance)
cpus-per-node
1
The number of CPUs per node; this parameter is used to calculate
performance.
cut-off
90
The point at which nodes should be flagged.
detailed-output-
flag
.false.
A flag indicating whether or not the additional log files, one for
each CPU is generated (.true)) or not (.false.)
Use the following procedure to confirm the efficiency of the nodes in the allocation:
1. Change to a system-wide directory:
# cd system-wide-directory
2. Load the mpi/hp/default m odu le file:
# module load mpi/hp/default
3. For XC3000 and XC4000 systems only, also load the mlib/pgi/5.1 module file:
# module load mlib/pgi/5.1
4. Invoke the following command if you are superuser:
# mpirun -srun -v -p lsf -n max /opt/hptc/contrib/bin/dgemm.x
Invoke the following command if you are not superuser:
$ bsub -nmax -o ./ mpirun -srun -v -n max /opt/hptc/contrib/bin/dgemm.x
The max parameter i s t he maximum number of p rocessors av ailable to you in the lsf
partition.
No warning messages appear w hen all the specified nodes are performing at their peak
efficiency.
15.4 Using the System Interconnect Diagnostic Tools
There are various tools that allow you to diagno se the system interconnect.
Some tools are provided by the system interconnect m anufactu rer and are discussed in the
Installation and Operation Guide (the hardw are documentation) for your system. Be sure to
consult the appropriate Web page fo r these system interconnect tools:
Myrinet
http://www.myrinet.com
Quadrics
http://www.quadrics.com
15-4 Using Diagnostic Tools