Track CPU time for a given run

[ykempf]

The CPU time used by a run should be computed and saved, e.g. to the logfile. This would give us at last more solid values for our large runs as we haven't tracked that very accurately so far.

Can this be done by phiprof? Or shall we make that an independent vlasiator feature?

[galfthan]

I had not thought about adding it to phiprof, but that would be an option.

I originally thought we would write the CPU hour stuff (accumulated over all segments, for this runsegment) into the restart files. We just have to be careful not to discard too much of the simulation time. If we measure and write it as a last parameter into the restart then there should not be too much that is unaccounted for at the end.

[galfthan]

Let's add better tracking of CPUh before the next big campaign. Then we could see how many CPUh each part takes, how much in total, and we would be able to more easily make estimates. This would include:

• Possiblity of naming run in the vlsv file
• Store in restarts array with info on each simulation segment (restart to restart)., e.g.: date, num-of-procs, tot-blocks, machinename(?), simulated t, walltime t, and CPUh This could be stored in restart vlsv files,
• additionally also be written in easy-to-read format in logfile when writing restart/ending vlasiator as a report.