John Aldrich wrote:
>
> On Fri, 14 Jul 2000, Kevin Wood wrote:
> > Hey guys, got a question. I have a customer who is requesting a quad
> > zeon system running RH 6.2 and 3GB of Ram. My last impression was that
> > the current kernels will only support 2GB of ram and the Alphas will
> > support more only with a patch. Is this true? The reason I ask is that
> > Dell is quoting the same system and said that they gurantee that the 3GB
> > of ram will work under RH6.2. Is this possible without going to an
> > experimental kernel? Any information will be greatly appreciated.
> >
> I think you have to have a custom-compiled kernel to get more than
> 2GB of ram.
> John
>
RH is using a back port of bigmem. (VA uses a different version of the
Bigmem patch than RH, and my remarks are based on that patch.) There
are 2 ways to boost the amount of RAM beyound 1 Gig in 2.2.x. The first
is to adjust the amount of memory directly address by the kernel. This
is availble in 2.2.12 or greater. I know that this option is
incompatible /w some versions of Oracle. (Something involving adding
shared memory segments.)
The Bigmem option is availble only if the kernels been patched for
bigmem. It some be noted that only 1 gig is directly addressable by the
kernel. This mean that isn't not really useful on a file server. (You
can't buffer more than 1 G.) You can enable both, however which allows
user space programs to use up to 4 gigs, the kernel up to 2 Gigs, and
may break Oracle. (I reported the issue to Oracle, but never heard back
for them. They may have fixed it in later releases.)
A quick glance at RH's kernel rpm reveals that CONFIG_2GB is not set,
and CONFIG_BIGMEM=y in kernel-2.2.14-i686-smp.config. So a default 6.2
would appear to be able to make use of up to 4 Gigs for userspace
programs, and up to 1 Gig for the kernel.
>From make xconfig:
ftp://ftp.valinux.com/pub/support/flory/2gig.jpeg
ftp://ftp.valinux.com/pub/support/flory/bigmem.jpeg
--
99 little bugs in the code, 99 bugs in the code,
fix one bug, compile it again...
101 little bugs in the code....
Samuel Flory [EMAIL PROTECTED]
--
To unsubscribe: mail [EMAIL PROTECTED] with "unsubscribe"
as the Subject.
