How MRBench works?

[Gaurav Dasgupta <gd...@gmail.com>]

Hi All,

I want to run "MRBench" benchmark to test the performance of my Hadoop
cluster. The command with parameters which I am submitting is:
*hadoop jar $HADOOP_HOME/hadoop-test.jar mrbench -numRuns <value> -maps
<value> -reduces <value> -inputLines <value> -inputType *random

For different number of "maps value" I am submitting, I can see that the
actual number of maps running are different.
According to my assumption, MRBench should use the parameters exactly which
I submit. But there's a difference in number of Maps.

Is my assumption correct? Am I doing some mistake?
Can someone describe/suggest how MRBench works and how I can evaluate
MRBench results to test the performance of my cluster?

Thanks,
Gaurav Dasgupta