MQbench
Summary
0
MQbench is a graphical benchmark for Cluster with Message Passing Interface.
It contains a full implemented connection benchmark embedded in a GUI.
Man pages for each benchmark routine, along with command for runnig and
compiling, are provided in doc directory.
MQbench needs a Qt implementaion available from <Trolltech>,
an implementation of MPI such as <mpich>
and <Qwt Widget> for technical applications.
How it works
As MQbenchs opens the GUI from the first MPI rank, therefor this node has to provide a xdisplay.
You can choose between different MPI calls and between different groups
of node groups.
Due MQbenchs capability of creating MPI_Comms, which initalize a subset of the given processor network,
it should not be nessesary to create a machinefile for system selection. You will be able to join the group of benchmark systems in the GUI together.
After starting the benchmark, MQbench will create a new MPI_Comm containing the selected nodes.
Now MQbench will transfer all nessesary information to this nodes and completes the benchmark. During that, MQbech begins to fill the ResultBrowser
with time- and speed-information.
Thanks to
Qwt - Qt Widgets for Technical Applications, they done a great job.
Please visit them <Qwt Widget>.
|