Niall Mullen's Weblog

specweb 2005 configuration: Building the fileset

This blog entry forms one part of a larger guide to specweb2005 which
is indexed here.

Building the Fileset

Building the fileset is quite straight forward and is well documented here .

The only interesting caveat I found is the level to which one should parallellise
the building of the fileset. The building can be parallellised by generating a
series of

In the example below the script will build a fileset of 24000 for the SPECweb_Banking
workload, the only variables which need to change are dirscaling which takes the values
1000 for banking, 50 for ecommerce, and .25 for support workloads.

#!/bin/sh

i=0
dirscaling=1000

while [ $i -lt 24 ]; do

i=`expr $i + 1`

sessions=`expr $i \* 1000`

dir=`expr $sessions \* 50`
dir=`expr $dir + 1`

cat template |sed -e "s/XXX/$dir/" -e "s/YYY/$sessions/" >bank_usercheck_props.$sessions

done

for file in bank_*
do
java -server -Xms384m -Xmx384m -jar /102/wafgen/Wafgen.jar $file &
done

rm bank*

This launches 24 processes in this case to build the fileset in parallell.
I have found that one process per cpu on lower end machines and a little less at
the high end generates the best fileset build time.

Posted at 03:29PM May 25, 2006 by nmullen in General  |