[torqueusers] Java job getting killed in MOM as soon as it gets
velayups at email.uc.edu
Fri Mar 3 07:47:08 MST 2006
Ronny T. Lampert wrote:
> I have to take some guesses as I am not actively working with java or
> x86_64s myself.
>> 1) When the VM is started with no default -Xms and -Xmx values, jobs get
>> killed when they get to about 84 MB of RES memory limit (this is very
>> repeatable as I have checked this several times), but the manpage on
>> java says that the default value is 64 MB.
> You forgot about the mem the VM needs for itself.
> 20MB seems like the numbers I usually see for an "idle" java VM.
>> 3) When I set -Xms200m and -Xmx3500m or -Xmx4000m, job gets killed in
>> less than 10 seconds of getting started.
> It depends on which kind of mode you run your apps on the x86_64s.
> If in standard 32bit/compat mode then you wont be able to have more than
> ~2.5GB of memory to use PER programm image (32bit = 4GB total; top or lowest
> 1GB = direct kernel mapping; just above 2.5GB or so the shared library
> mapping begins).
> Also java's model per se is 32bit IIRC.
> So I suggest making sure you use the native x86_64 64bit java VM and also
> have a look at this java VM option:
> -d64 use a 64-bit data model if available
> Hope this was of some help.
You hit it on the head Ronny, Thanks. It was indeed 32-bit Java VM that
was being used. I did not give even a moment's thought to check that
first. I installed Java 1.5 for x86_64 and the jobs are getting more
than 4 GB allocated.
More information about the torqueusers