understanding applications and PAE

[Chris Kloiber]

On Tue, 2003-12-09 at 08:16, James Olin Oden wrote:
> On Mon, 8 Dec 2003, Vanco, Don wrote:
> 
> > While I hate to use this as a support line I've gotten no help from the 2
> > local lugs I've queried, and know that there's some heavy hitters lurking
> > here.
> > 
> > I understand that the Linux kernel can (through PAE) provide memory paging
> > beyond 4GB of physical memory on 32-bit architecture.  What I need to
> > understand is: what are the requirements on the application side to utilize
> > this memory?  I'm looking for documentation / links that talk about
> > applications and PAE.
> > 
> > Case in point: Lotus Notes on OS400 can address huge amount of RAM, but when
> > it's running on Linux does it need to be compiled in a specific manner, or
> > with some specific bit of code / API to address more than the architecture
> > physically supports?
> >
> Hi Don,
> 
> We use the PAE feature where I work.  The deal is that this functionality 
> only makes it possible for the kernel to access and use all this memory,
> but each process still can only address 4Gig of memory less the amount of
> memory under 4Gig the kernel is using.  If you wanted a single app
> to use more than 4Gig, at this point you would need to create several
> processes that share memory.  Even this this is very specific to the 
> application as to whether such a scheme could be used effectively (you
> would in effect be shuffling data around processes as needed).
> 
> Does anyone know of any other solutions than PAE?

Can you do some tricks like Oracle does? They can use very large shared
memory segments using /dev/shm I've seen customers use /dev/shm up to 6
gigs on a 8 Gig box, and 14 Gigs on a 16 Gig box. This is probably
slower than normal ram access, but faster than swapping to disk.
Remember I'm in Support, I can't help you program this.

-- 
Chris Kloiber
Red Hat, Inc.