MRPerf
======
Guanying Wang, Ali R. Butt, Prashant Pandey, Karan Gupta
{wanggy,butta}@cs.vt.edu, {ppandey,guptaka}@us.ibm.com
2009.6
What's MRPerf
-------------
MRPerf is a simulation tool to model Map/Reduce application performance on
large clusters. The current simulation mimics Hadoop, an open-source Map/Reduce
implementation.
Get Ready to Run MRPerf
-----------------------
To run simulation, ns-2 must be installed and patched. A patch is available in
ns.patch/, and it's against ns-2.33. The patch is not tested against other
versions of ns-2. Since xml configuration files are used in MRPerf, libxml2
must be available to compile patched ns-2. A copy of libxml headers is included
in the patch. In the future, it is desirable to specify dependency through
automake tools. Once ns can be found and run in command line, MRPerf is ready
to run.
How to Run MRPerf
-----------------
MRPerf source files are in src/. It's highly recommended to copy the source
files into a new dir and run simulation there. Run "make topo", "make gen", and
"make" in turn to run the simulation. The first two commands convert some
configuration files from .py or .xml to .tcl for ns-2 to read, and "make"
launches ns-2 on .tcl simulation files.
How configuration files are created, and their relationship is specified in
conf.files.relationship. Manually modified configuration files are
metadata_gen1.xml, topology1.xml, hadoop_conf.py, parameters.tcl, and
hadoop.job.tcl. Some sample configuration files are in exp/.
Disclaimer
----------
This is a pre-release/preview version of MRPerf. Some part of the code may not
work as expected. The authors will not provide support or bugfix of the code.