[Bf-committers] Seemingly hugepage-related performance issues

Jonas Wielicki j.wielicki at sotecware.net
Fri Nov 23 09:15:46 CET 2012


Dahlia,

On 22.11.2012 22:11, Dahlia Trimble wrote:
> This sounds like something tcmalloc might help with. I believe you can link
> it in at run time with a command line switch. It's at
> http://code.google.com/p/gperftools/
Wow, that works like a charm. Blender memory use may have increased a
bit though, but I'm not sure. Gotta check that later, will finish my
bake now first. No more stalls between the simulation frames.

> Oh and a request to blender devs: if it works and it's something you
> want, please make it optional because it doesn't like to release
> memory back to the OS.
>From reading into tcmalloc, it has options to release the unused memory
to the operating system, so that's something which could be called after
a full bake or alike. Otherwise, tcmalloc seems to be releasing memory
gradually[1] (similar to what the memory manager of sun java does).

Thanks for pointing that out,
Jonas

   [1]:
https://gperftools.googlecode.com/svn/trunk/doc/tcmalloc.html#compiletime


More information about the Bf-committers mailing list