On Thu, May 30, 2013 at 9:57 AM, Jed Brown <[email protected]> wrote:

> Cody Permann <[email protected]> writes:
>
> > Yeah, I guess I need to spend more time with Massif.  I tried it last
> week
> > for the first time as well and was overwhelmed by the amount of output.
> >  From what I've seen just scanning over the output is a detailed log of
> > when and how much memory is used.  Is there a way to flip that tree
> > upside-down to see what structures hold onto memory?  That would be
> > useful...
>
> How would you identify those structures?  PETSc can capture the stack
> (in debug mode) or the function (debug or optimized) where things are
> allocated, as well as the Stage (PETSc profiling).  Massif is more
> complete, of course.
>

I have no idea, and once you allocate memory you can always transfer
ownership so you couldn't rely on it anyway.  Just wondering!

In one of my secondary tasks, we are tracking down a memory issue with TBB
where our usage appears to go up and up even though we don't see any leaks.
 The memory usage is completely flat without threads.  However, with
threads, the memory usage curve appears to be flattening out, but I haven't
run out enough timesteps on these big problems to see if it reaches
some asymptotic limit.  Without an obvious leak, we are trying to track
down which thread local structures are growing over time which is tricky...
 Trying to sort through Massif output is an option, but not very easy.  I'm
thinking about turning of threading in various structures in MOOSE to see
if we are caching data in some odd way since TBB ranges can vary from call
to call.

Cody
------------------------------------------------------------------------------
Introducing AppDynamics Lite, a free troubleshooting tool for Java/.NET
Get 100% visibility into your production application - at no cost.
Code-level diagnostics for performance bottlenecks with <2% overhead
Download for free and get started troubleshooting in minutes.
http://p.sf.net/sfu/appdyn_d2d_ap1
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to