I am planning to do the Parallel Computing (Grid Computing) using the Globus and Condor.
I just want to find out is it possible to do the following.
1) If I submit a job either to Globus or directly to condor. The job has to be divided (either by globus or by condor) and sent across different available resources (nodes) and collects the results, combine and send the combined result back to the client. That is for example if a job takes ordinarily 10 mins, if the same job submitted in the above scenario I like to see definitely less than 10 mins say may be 5 mins suppose if the job is divided across 5 resources.
2) Does the globus or condor will automatically do this for us, or do we need to write a separate program to do this or the job it self should be written in special way to accomplish this.
3) Does the cluster will divide and send the job across its nodes and collect the result and send the combined result back.
4) What is MPI, MPICH and what for it has been used.