VM Size greater than allocated for a Java Heap

[Paddy_1]

Hi,

We have an rp class box with 2 gig of RAM. The machine is extremely slow to work with.

As usual we started running glance and we see that memory utlisation is almost always at 95,99 or 100%. The system runs approximately 10 Java Virtual Machines of varying sizes but none exceeding 256 M.

The interesting part is that inspite of limiting the maximum amount of memory that can be consumed by a Java process say 256, the VSS column reports a consisten use of 450 M. I might be missing something elementary out here in explaining this phenomenon.

Any stab at explaining the behavior is greatly appreciated.

Thanks
Paddy

[RAC_1]

How do you limit the memory that will be used by JVM?? (you control that with initial heap size, max heap size options on java command line.)
Are you doing that?? Check both sizes - initial heap size and max. heap size.

Also you should be up to date on java patches.

Anil

[Paddy_1]

Anil,

Yes. We use the java command line of -Xmn and -Xmx to set the initial and max heap size.The current values are 0 and 256.I experimented with increasing the minimum size which only reflected a change in the RSS but no effect on the VSS.

I am not sure of the lastest HotSpot Patch on HP.The following is my output of java -version
-----------------
java version "1.3.1.13"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.3.1.13-040210-15:41)
Java HotSpot(TM) Server VM (build 1.3.1 1.3.1.13-_10_feb_2004_17_15 PA2.0, mixed mode)
--------------

[RAC_1]

what is the output of the following.

ps -efx|grep [j]ava

This would show the values you used on command line.

Anil

[Paddy_1]

I am attaching the output as a file since there are too mahy and mostly the minimum setting is 50 M and max of 256M.

[Paul Torp]

hi.. i have recently been at a preformance course in websphere..

U should use PRM and WLM and have one java-prosess stuck at a single cpu.

Java should NEVER swap.

One could also look at the Garbage collectors in the j2ee environment..

Typical for a badly written java application is that it collects objects until one reaches the memomory limit, and then it uses ALL resouces to delete the objects it has collected..

typical indication is running top and look at the load-average compared to idle.

if load-average is high and idle is good u have a GC problem.

paul

[RAC_1 · Accepted solution]

The last process in your attachment, has option -xms50m, (50 mb) it certainly does not have option for max heap size. you need to start that processes with max. heap size option.

The other three processes are old.

Anil

[Paddy_1]

Anil,

For the last process After the URL Protocols jar file you'll find the -Xmx directive of setting to 256M (I admit that visually its kinda confusing.

Meanhwile I have asked my system folks to upgrade the system with the attached list of patches. I have got this list after running HPjconfig.Some of them are present already but some missing.

[Dietmar Konermann]

Paddy,

I think what you're seeing is quite usual for JVMs. First of all VSS is the virtual set size... if you see memory bottlenecks then you should have a look at the resident set size (RSS)!

However, if you look at the memory regions in Glance you should see that the largest (VSS) regions are of type "MEMMAP/Priv". These memory mapped files are used for JVM internal memory management and they are allocated using mmap()'s MAP_NORESERVE flag... which means even no waste of swap due to layz swap allocation.

Best regards...
Dietmar.

[Paddy_1]

Thanks.That might be interesting as the problem might be somewhere else then.If thats true I am looking at the wrong problem..

How come the same behavior isnt reflected in AIX ..(probably the jvm implementations, I guess)...

[Paul Torp]

points-assignment would be nice.. tnx in advance.

[Ng Jiunn Jye]

Hi Paddy,

I am experiencing the same problem now. Have you got the answer to this problem.

For my case, I have JVM running on both 1.3.1.13 and 1.4.2.03. The problem happened to both JVM on itanium machine. Testing on an other HP machine does not show this behaviour. The memory is limited to 100m (-Xmx100m) but TOP show java is taking up to 900m. This cause the system to run out of memory and the application need to be restated periodically.

Please advise.

Thank you

Rgds,
jay