Memory Limits on Linux on x86

Jeff Layton laytonjb at
Fri Jul 19 10:42:16 PDT 2002

Mark Hahn wrote:

> > problem we have. I know that 32-bit x86 Linux has
> > a memory limit of 3 Gigs (3.5 with a patch) that it
> 3.75 is pretty trivial.  don't let the "patch" scare you off -
> there's nothing exotic about it, it's been around for years,
> consists of just a couple lines of code change, and some dist/kernels
> come with it as a CONFIG_ option.

Yes, I've seen the patch (very simple).

> > can address. Is that 3 Gigs per user process or 3 Gigs
> > total?
> ia32 hardware is inherently limited to 4G to a process, always, ever, period.
> the 3/3.5/3.75 comes from the fact that the kernel likes to have an
> aperture within that space so that it can do IO.  (consequently,
> doing 3.75 with lots of fast IO would be a mistake.)

I was thinking of 3.5 Gigs with some pretty hefty IO it a Ramdisk
from the extra memory. Do you think this would work OK?

> the kernel can use up to ~64G physical ram.
> even if you have just one 3G proc running, this ram may still
> be worth using, since you can do IO and have the 60G of pagecache.
> you can also use that ram for Shm, though of course no one proc
> can have more than a few gigs mapped at once.
> note that doing lots of IO through highmem can be a performance problem
> as well, since not all IO hardware can deal with 64b addresses.
> if you have 21 3G procs running, using something like MPI,
> there will be no problem using all 64G ;)  if you manage,
> please do brag to the list...

I wish! I'm hoping to do two 3.5 Gigs processes and a 4 Gig
Ramdisk for better IO throughput (instead of a disk).



> > With the new E7500 and GC chipsets with large
> > amounts of memory, I'm hoping it's 3 Gigs per user
> > process.
> it is.

More information about the Beowulf mailing list