OpenJDK and boxes with lots of RAM

[Jan Just Keijser]

Florian Weimer wrote:
> * Jan Just Keijser:
>
>   
>> I would like to report an issue with OpenJDK 1.6 (and Oracle Java 6u29)
>> that we have run into on our grid computing farm:
>>     
>
> The same topic has recently been discussed on the hotstop-dev list.
>
>   
>> the latest&greatest worker nodes have 12 cores and 48 GB of RAM ; we
>> offer a maximum 12 jobs slots on these worker nodes, in order to
>> accomodate small and large jobs (users can requests 1 - 12 cores). To
>> ensure that the different jobs don't interfere with each other we also
>> set a VMEM limit for each job slot. The current VMEM limit is 48 / 12 =
>> 4 GB RAM per job slot. Each single core job that starts has a 'ulimit
>> -v' of 4,000,000 .
>>     
>
> Hmm.  Does this mean that Hotspot should call getrlimit(RLIMIT_AS) and
> limit itself to some fraction of that value if it is not unlimited?
>
>   
I like that idea, e.g. 60% of RLIMIT_AS would be good (instead of 25% of
PHYS_MEM) - it would be very nice if a user (or sysadmin) can make a
global setting to affect the percentage.

The environment variable _JAVA_OPTIONS (and/or _JAVA_TOOL_OPTIONS) would
be suited for this, as long as there's a way to get rid of the annoying
message :)

JJK / Jan Just Keijser



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/jdk6-dev/attachments/20111118/aa115d59/attachment.html