Hi Pete,
Sorry, I think I didn't understand your original suggestion (Lukasz's link helped too!).
Re-reading your link, do you think this might go away with:
MALLOC_ARENA_MAX=1
? My understanding is that this should limit malloc to a single arena, which might be the source of our problem. Vincenzo, could you try this?
Thanks,
Brian
On Dec 22, 2011, at 3:40 PM, Peter Elmer wrote:
> Hi Brian, Vincenzo,
>
> On Thu, Dec 22, 2011 at 03:32:54PM -0600, Brian Bockelman wrote:
>> In CMS, Vincenzo has reported that using xrootd causes interesting memory spikes in our application. He wrote up a page here:
>>
>> https://twiki.cern.ch/twiki/bin/view/LCG/VIJemalloc
>>
>> Basically, the conclusion is that by switching malloc implementations, the memory spikes (which are large - 500MB of RAM!) go away. This likely indicates not memory leaks, but "abusive" memory allocation patterns or strange threading affects.
>>
>> Any idea where these might be coming from? Anything we can do to adjust this?
>
> The thought was that this was due to some bad interaction between memory
> arenas and the extra thread from the xrootd client.
>
> As I suggested in the CMS HN thread about this, did someone follow up
> on the variables that can be used to change that with the existing memory
> allocator? (The relevant bit from that CMS HN thread is included below.)
>
> Pete
>
> On Wed, Dec 14, 2011 at 04:12:30PM +0100, Peter Elmer wrote:
>> Hi Vincenzo,
>>
>> On Wed, Dec 14, 2011 at 04:08:17PM +0100, Vincenzo Innocente wrote:
>>> On 14 Dec, 2011, at 4:05 PM, Vincenzo Innocente wrote:
>>>> On 14 Dec, 2011, at 4:04 PM, Peter Elmer wrote:
>>>>> I ran across this posting:
>>>>>
>>>>> http://www.quora.com/Why-do-some-applications-use-significantly-more-virtual-memory-on-RHEL-6-compared-to-RHEL-5
>>>> on SLC5. I can also run on SLC6 if needed.
>>> btw per thread arena is what jemalloc uses ans it seems to help to keep xrootd memory under control.
>>
>> Hmm, I wonder if we can achieve the same effect on SLC5 via the
>> MALLOC_ARENA_* variables described here:
>>
>> http://www.centos.org/docs/5/html/5.4/technical-notes/glibc.html
>>
>> (see very last point in the 1.71.1 list, here for CentOS 5.4, but
>> presumably it applies also RHEL5.4/SL5.4).
>>
>> Pete
>
>
>
> -------------------------------------------------------------------------
> Peter Elmer E-mail: [log in to unmask] Phone: +41 (22) 767-4644
> Address: CERN Division PPE, Bat. 32 2C-14, CH-1211 Geneva 23, Switzerland
> -------------------------------------------------------------------------
>
> ########################################################################
> Use REPLY-ALL to reply to list
>
> To unsubscribe from the XROOTD-DEV list, click the following link:
> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1
########################################################################
Use REPLY-ALL to reply to list
To unsubscribe from the XROOTD-DEV list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1
|