Mailing List Archives

Public Access

 		[Computer Systems Lab]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] java machines with > 2gb

This issue has everything to do with 32 vs. 64-bit operating systems. With 32-bit OS's no process can have more than a 3GB address space. If you need more than 2GB of Java heap space ensure you are submitting your job to a 64-bit machine with a 64-bit OS. If you don't need more than 2GB's, follow the workaround. 

_joe



Matthew Farrellee wrote:
A quick read of the work around[1], a few runs of java -Xmx... NOTHING, and it seems this isn't entirely Condor's fault. 

$ java -Xmx4g NOTHING
Invalid maximum heap size: -Xmx4g
The specified size exceeds the maximum representable size.
Error: Could not create the Java Virtual Machine.
Error: A fatal exception has occurred. Program will exit.

vs

$ java -Xmx1g NOTHING
Exception in thread "main" java.lang.NoClassDefFoundError: NOTHING
...
On my 32 bit system, the JVM is not able to handle the 4GB max heap size I've requested, but it can handle 1GB.
Condor is trying to be helpful and set the max heap size for you by default, specifically to sysapi_phys_memory()/ sysapi_ncpus(). The problem is sometimes that number is too big for the JVM. In those cases you need to stop Condor from trying to help you out. This is what setting JAVA_MAXHEAP_ARGUMENT to nothing is all about. 

 From src/condor_c++_util/java_config.C:

    tmp = param("JAVA_MAXHEAP_ARGUMENT");
    if(tmp) {
       arg_buf.sprintf("%s%dm",tmp,sysapi_phys_memory()/ sysapi_ncpus());
       args->AppendArg(arg_buf.Value());
       free(tmp);
    }
If it isn't set, i.e. "JAVA_MAXHEAP_ARGUMENT=" in your config file, then no max heap is passed to the JVM. 

The work around mentions setting your own max heap with:
JAVA_EXTRA_ARGUMENTS = -Xmx1906m
That is well and good, as long as you are setting the max heap to a size that the JVM can handle! If you were to JAVA_EXTRA_ARGUMENTS = -Xmx4g on my machine you'd still have a problem.
So, you can either not specify a max heap, and deal with whatever fall out there is there if any, or make sure you are setting it to something reasonable with the little java -Xmx... NOTHING test above. Be careful though, this test doesn't seem to be fool-proof. From my testing: 

$ java -Xmx2670m test
Error occurred during initialization of VM
...
$ java -Xmx2669m test
Exception in thread "main" java.lang.NoClassDefFoundError: test
...
$ java -Xmx2670m test
Exception in thread "main" java.lang.NoClassDefFoundError: test
...
$ java -Xmx2671m test
Exception in thread "main" java.lang.NoClassDefFoundError: test
...
$ java -Xmx2672m test
Error occurred during initialization of VM
...
$ java -Xmx2671m test
Error occurred during initialization of VM
...

Best,


matt

[1] http://www.cs.wisc.edu/condor/manual/v7.1/3_13Java_Support.html#29892

Michael Rusch wrote:

I share your surprise that it hasn't been fixed after 5+ years, especially
now that machines with 2+gb of RAM are not really extraordinary.  Condor
team: are there any plans to fix this?

I'll probably wind up following suit and reverting to the vanilla universe.
We have a mixed linux/windows pool, so I'm assuming I'll need to write two
bootstrap scripts named with the appropriate magic so that I can write the
submit file to use the correct one based on target machine's architecture.
Is there any way that I can get around having to make sure Java is installed
in the same place on all of the machines of the same architecture (and then
hardcoding that path into the bootstrapping scripts)? 
Michael


-----Original Message-----
From: condor-users-bounces@xxxxxxxxxxx [mailto:condor-users-
bounces@xxxxxxxxxxx] On Behalf Of Filip Defoort
Sent: Monday, September 08, 2008 1:34 PM
To: Condor-Users Mail List
Subject: Re: [Condor-users] java machines with > 2gb

The only solution I found was bypassing the java universe altogether
and starting your own JVM via a regular java invocation.
Works, but kinda defeats the purpose of the java universe.

Quite surprised this hasn't been fixed yet - this bug has been there for
years.
- Filip


On Mon, Sep 8, 2008 at 11:25 AM, Michael Rusch <mcrusch@xxxxxxxx> wrote:

Has anybody found a way with recent versions of condor to get Java nodes
with >2GB of RAM to show up?  The workaround from the manual is not

working

for me.



Michael

_______________________________________________
Condor-users mailing list
To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with

a

subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/condor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/condor-users/



_______________________________________________
Condor-users mailing list
To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/condor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/condor-users/

_______________________________________________
Condor-users mailing list
To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/condor-users
The archives can be found at: https://lists.cs.wisc.edu/archive/condor-users/ 
_______________________________________________
Condor-users mailing list
To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/condor-users
The archives can be found at: https://lists.cs.wisc.edu/archive/condor-users/