Iwan Aucamp wrote:
Hi
The memory leak - or atleast what appeared to be a memory leak
(growing heap size) was only observed on libmalloc.
After I sent the initial message I ran with libumem for 7 hours - and
the heap size increase was sensible (7 MB for this time ) - as soon as
i switched back to libmalloc the heap size started growing at a rate
of more than 20 MB per minue - so this does seem to be related to way
in which application uses heap and way in which libmalloc manages the
heap.
I will continue to use libumem and see how things work out.
Sounds good. I'd recommend libumem anyway... the technology behind it
is well proven in the kernel. :-)
- Garrett
Thanks
-----Original Message-----
*From*: Garrett D'Amore <garr...@damore.org
<mailto:garrett%20d%27amore%20%3cgarr...@damore.org%3e>>
*To*: Iwan Aucamp <aucam...@gmail.com
<mailto:iwan%20aucamp%20%3caucam...@gmail.com%3e>>
*Cc*: opensolaris-code@opensolaris.org
<mailto:opensolaris-code@opensolaris.org>
*Subject*: Re: [osol-code] Can Heap Fragmentation Cause Heap Leak
*Date*: Fri, 07 Aug 2009 19:10:48 -0700
Iwan Aucamp wrote:
> Hi
> I have observed something which looks like a memory leak in an
> application - pmap shows heap size growing at a considerable rate.
> I have tried to track down the leak using several tools which
> identifies memory leaks - and even wrote one myself - none of these
> found anything:
> - Libumem
> - Valgrind
> (on linux since the application works on linux also - but runs on
> solaris in production)
> - Malloc interposer which keeps account of allocated memory and
> prints out summary file based on stack traces which lead to malloc.
> The application in question is a multithreaded (50 worker threads)
> application which quite heavily relies on malloc and does not really
> try to prevent heap fragmentation.
> My question is: could severe heap fragmentation leave heap space
> unusable and thus create something which appears to be a memory leak ?
I don't think so -- at least not with libumem. With other memory
management libraries this could happen.
- Garrett
> Regards
> ------------------------------------------------------------------------
>
> _______________________________________________
> opensolaris-code mailing list
> opensolaris-code@opensolaris.org <mailto:opensolaris-code@opensolaris.org>
> http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
>
_______________________________________________
opensolaris-code mailing list
opensolaris-code@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
