Processor Synchronisation

Hi.

I have set up a very small cluster (2 nodes, 4 processors on each node). I am trying to monitor CPU, memory and network resource usage for it. I have installed Ganglia monitoring tool on it which is quite cool. I now need to deploy a few benchmark HPC(High Performance Computing) C++/MPI applications to benchmark them. The cluster has Ubuntu desktop server installed on it at the moment.

Can anyone please advise how I can get the processors in the nodes to communication with each other and deploy the application so all 8 processors work together to compute the problem?

I have been advised I may need to write a synchronization code which I have no idea how to do (total newbie programmer!) or I should look for a job scheduler?

I came across the following list of free software but I really don't know which is the most appropriate to what I need to do.

Download Free Windows Clustering Open Source Software - SourceForge

Can someone please advise?

Writing your own synchronization code will be very difficult specially for a newbie programmer. Since you will have to use a lot of semaphores to avoid the race condition and divide the jobs among the two schedulers. You will also need to study the benchmarking application's code(given that it is open source).
A job scheduler is what you need. I am not too experienced in the practical application of these kind of systems though I know the theory. Sorry I wasn't much help.

Don't know anything about the subject but you might want to check this part of your topic: The cluster has Ubuntu desktop server installed on it at the moment.

It's either desktop or server, pick one

Correction.

You can tell I'm a total newbie can't you!? Thank you. Its Ubuntu desktop.

In addition I am in the process of installing SLURM (job scheduler) to tackle the issue.

Are you just trying to judge the performance from your cluster? If so, it's probably easier to find some kind of distributed processing software than to write your own.

The short of it mihir has, you need a scheduler and it has to know how to manage the remote jobs (or remote manage the local schedulers).