César Ávila
Tue, 25 Oct 2011 09:28:30 -0700
2011/10/25 Michael Johnston <michael.ap.johns...@gmail.com> > > > On Tue, Oct 25, 2011 at 4:44 PM, César Ávila <clav...@gmail.com> wrote: > >> There are some specific cases where I have observed memory leakage in >> Adun. One of them is when you run a single simulation using the same output >> path as a previous simulation. In this case the increase in memory demand is >> very sharp and AdunCore will quickly freeze your computer as it runs out of >> memory. >> > > Strange - maybe because it reads in the existing data? Some questions: What > size simulation exists at the path?; how much memory gets used?; does it > continually increase as the simulation progresses or is it one sharp jump? > > It looks like instead of dumping the generated data to a file, it keeps storing it in memory as the demand continually increases during the simulation. You can try it yourself with any simulation, if you give it enough time (depends on the size of the information that needs to be stored).
> Now I am seeing a small memory leakage while running RepEx simulations. I >> would like to debug if this is related to the plugin or just to the >> forcefield (some memory allocation that is not released upon switching the >> systems). Is there a quick way to debug this kind of problems? >> > > I'm trying to think of one and can't so that probably means theres not ... > > Some possibilities: > > Valgrind > > Pass OutputMemoryStatistics YES as an argument this will print out memory > allocation info at every allocation.In combination with debug output > (SimulationLoop etc) it may help you spot the routine where the footprint > increases > > Xcode has a good memory debugger for object-allocations but this won't work > with parallel RepEx - however I think there is some mechanism in gnustep for > tracking object allocations. Although this will only help if its an Objc > related problem not a C related one. > I'll see if I can remember what it is. > > To test if its your force-field - create a script where you set up > system+force-field. Then create a loop where you call 'reloadData' on the > force-field a couple of hundred/thousand times and see if the memory goes > berserk - if its stable then its not the configuration exchange part. > > > I will give them a try. > > > >> >> _______________________________________________ >> Adun-dev mailing list >> Adun-dev@gna.org >> https://mail.gna.org/listinfo/adun-dev >> >> >
_______________________________________________ Adun-dev mailing list Adun-dev@gna.org https://mail.gna.org/listinfo/adun-dev