[OSM-dev] mmap() (was: Re: Some statistics for file-format -development)

[Marcus Wolschon]

2008/11/25 Sascha Silbe <sascha-ml-gis-osm-dev at silbe.org>:
> On Tue, Nov 25, 2008 at 08:47:16AM +0100, Marcus Wolschon wrote:
>
>> Okay, on a big AMD-multicore-laptop the limit in Java seems to be
>> 1GB of memory-mapped files.
>
> That's pretty small given that kernel/user split is usually 2G/2G or even
> 3G/1G...
>
>> Ulimit was "unlimited" and --XX:MaxDirectMemorySize was set to 32GB
>> -Xmx was a few GB. I am using the current Java6 -patchlevel 10 from
>> Sun.
>
> Have you tried using a smaller -Xmx size? I have the impression that Java
> allocates the maximum allowed size directly on startup, relying on the OS to
> do lazy allocation of pages. If that's the case, reducing -Xmx should
> increase free virtual address space, allowing bigger memory mappings.

It does not. Xmx is the maximum size, Xms would be the initial and
it used no more then a few dozen megabyte or heap-space acording to
jconsole.


>> I'd like to try this on a 64bit-Linux on that box but don't know how to
>> do that yet. Any help on turning a debian 32bit into a 64-bit or running
>> a 64bit-VM in a 32bit-Linux?
>
> I don't think that's possible. Both 32bit and 64bit userspace and emulation
> work with a 64bit kernel, but at least 64bit userspace on a 32bit kernel
> isn't possible.

I'm not talking about native execution but a virtual machine with
qemu or Vmware that emulates all of the hardware.
>From what I could find at least qEmu should be able to emulate all
the guest-architectures (including x86-64, PPC and Sparc32/64) on an
x86-32 host.

Marcus