MPI process

The MPI benchmarks are performed with:
- A high memory bandwith scientific code: 1 step corresponds to one iteration of the main program loop.

Memory usage summary over 25 steps:
heap total: 8843158354924, heap peak: 805454928, stack peak: 52080

            total calls     total memory     failed calls
malloc| 1261103     8843158354924       0
    free| 1261069     8843158267004       0

- The Message Passing Interface is essentially used as a convenient tool to start and stop N identical processus in parallel. As a result, it does not involve significant communications during the run and does not increase significantly the overall running time of each processus.
- For N varying from 1 to the total number of available cores, the figure shows the mean computing time T per processus over 25 steps as a function of N. For i=1,...,N, if t[i] is the physical time (in seconds) for the processus i to complete 25 steps, T is defined as
    T = (t[1] + t[2] + ... + t[N])/N
and should remain independent of N.

CMSSW process

Typical CMSSW process.
Benchmarks howto