Java 64 bit consumes excessive native memory (c-heap) due to JIT Compilation

Serguei Mourachov smourachov at gmail.com
Wed Aug 29 09:05:33 PDT 2012


On 29-Aug-2012 08:30, Andrew Haley wrote:
> On 08/29/2012 04:49 AM, Serguei Mourachov wrote:
>> That could also be a result of glibc bug
>> http://sourceware.org/bugzilla/show_bug.cgi?id=11261
> That one is not a bug, as Ulrich explains.  It's a misunderstanding
> of what is supposed tp happen.
>
> Andrew.
>
Right, actually I meant the MALLOC_ARENA_MAX race condition issue.
We have observed behavior similar to what  Ashish has described in two 
occasions.
In first case it was jdk6 running on RHEL 5, at that time we could not 
figure out real cause of the problem and just switched to jdk7, that 
apparently is using somewhat different memory allocation strategy.
Second case was about our code running  on jdk7 and calling malloc/free 
using Unsafe  from multiple threads, in that case  we had to switch to 
RHEL6 and start using MALLOC_ARENA_MAX=1.
After we found actual source of the our problems , we switched the first 
application  to use MALLOC_ARENA_MAX=1 on jdk7/RHEL5 and we saw some 
substantial reduce in memory consumption as well, but not as significant 
as in our second case.

Serguei Mourachov


More information about the hotspot-compiler-dev mailing list