Out of Memory in Event Generation

[cweiss]

The memory usage of Whizard increases continuously increases until the process is killed. This was first observed in the context of NLO event generation, but also occurs e.g. for the process ee -> Zbb for n_events > 4M.

The process will be executed with the nagfor memory tracing option. Maybe this gives some information.

[Juergen Reuter]

Could this be even related to the pointer problem in #772?

[kilian]

I also suspect that this is caused by the broken polarization assignment, unless it turns out to be an old problem. If true, this will go away with the reimplementation as promised in #772.

You may check if the problem is already present in version, say, before r7461.

[cweiss]

The memtrace shows that a lot of objects of the state matrices are not finalized. This probably yields the memory leak. Therefore, I do not think that this problem is related to #772. However, we should wait until WK's refactoring is finished before we address this issue.

[kilian]

Replying to cweiss:

The memtrace shows that a lot of objects of the state matrices are not finalized. This probably yields the memory leak. Therefore, I do not think that this problem is related to #772. However, we should wait until WK's refactoring is finished before we address this issue.

Yes, those state matrices are our Sorgenkind ...

Refactoring would be necessary - but is out of question currently, there are more urgent tasks on the agenda, given limited manpower. Just notice that #395 would benefit from a generalization of the interactions data structure, so there is at least some more concrete motivation.

[Juergen Reuter]

Milestone renamed