How is the native memory limit (malloc) computed?

Question

On Windows (32 bits), with the SUN JVM, the memory space that can be allocated in native code (e.g. with malloc in C) is limited and depends on the maximum heap size, specified with the -mx JVM argument. The larger the heap size, the smaller the native code memory limit.
With JRockit, this limit is higher: it is possible to allocate more memory in native code than with the Sun JVM.
However, on some machines the memory limit is not higher. The problem is that I see no difference in the configuration of these machines. They also have 2GB of RAM, with the same amount of free memory (virtual and physical), the same OS and the same OS version. On both machines, the same application is used, with the same JVM arguments.
On some machines, we can use a -mx600m and allocate a given amount of memory in native code, and on other machines, for the same amount of native memory, we can only use a -x300m.
Do you have an idea, what the difference can be? Is it possible on these other machines to use a higher -mx value?
Thanks,
Laurent
We are using jrockit-R27.2.0-jre1.5.0_10 on windows XP.

Answer 1

It's possible that some DLL is eating up process address space. Start JRockit with -Xverbose:memdbg to get an ascii graphic over the address space. Or use something like Process Explorer to see what is loaded into memory and where.
-- Henrik