For those interested I've had a reply off Dell and Redhat for this
issue:

 

We have done some research  & this is what our software tech & RH have
stated --> i.e.: you REALLY needs to use a 64bit OS. 

This is a typical behavior of 32 bit systems with a lot of memory. In
Linux, the system memory is divided into 3 zones, namely: zone DMA
(first 16 MB), zone Normal (from 16 MB - ~1 GB) and zoen HighMem
(remainder 1 GB - 64 GB here). Unfortunately, in the 32 bits intel
platforms, the kernel is restricted to do all its memory allocaltions in
the zone Normal. 

With 64 GB of memory, more than half of the zone Normal is consumed by
data structures used to describe and manage the rest of the memory.
Since those are kernel data structures, they cannot be moved to Highmem.
This leaves very little room for the kernel to run its operations. 

This is why the OOM killer is triggering despite the fact that the
system still has 48 GB of memory free (MemFree: 48190756 kB). However,
the free memory is all in the Highmem zone which cannot be used by
kernel alloctions. Kernel allocations are restricted to the Normal zone
(low memory) where you have only 50 MB free (LowFree: 51084) and I guess
there was even less when the OOM killer triggered. 

The RHEL 5 PAE kernel is support with up to 16 GB of RAM, and should
work correctly with up to 32 GB of RAM, but beyond that, it is strongly
recommended that you switch to a 64 bit kernel. 

We do not ship a kernel-hugemem in RHEL 5 because of several reasons: 

1. the RHEL 5 PAE kernel handles memory in a better way and we do not
require the 4GB/4GB address space split used by -hugemem 

2. it became extremely difficult to maintain a reliable -hugemem kernel
in addition to the Xen patches in RHEL 5 

3. almost all new servers are 64 bit anyway and we'd like to discourage
running large memory systems in 32 bit mode because it causes too many
problems 

4. now, it is very safe to run 32 bit applications on 64 bits kernels

 

Kind Regards

Nick Lunt

Managed Services and O/S Analyst

Patech Solutions Limited

Tel: 01543 444 707

Fax: 01543 444 709

Tame House, Fradley Park, Lichfield, Staffordshire, WS13 8RZ

www.patech-solutions.com <http://www.patech-solutions.com/home.htm> 

 

 

From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of Nick Lunt
Sent: 24 September 2008 11:46
To: Red Hat Enterprise Linux 5 (Tikanga) discussion mailing-list
Subject: [rhelv5-list] PAE Kernel 32 bit RHEL 5.2 64GB RAM

 

Hi folks,

 

We've got a RHEL 5.2 32 bit server with 64GB of RAM. We had to use 32
bit Redhat due to the application.

Were using 2.6.18-92.1.10.el5PAE kernel and the OS can see all 64GB, we
can also stress the server and it will use all 64GB.

 

The problem is oom-killer is going ballistic, killing processes all over
the place.

 

I've logged a call with Dell about this and they say we cannot use >
16GB of RAM on a 32 bit kernel. RHEL 5 doesn't have a Hugemem kernel
hence the PAE kernel.

 

I got the following info from here
http://en.wikipedia.org/wiki/Physical_Address_Extension

"The x86 processor hardware is augmented with additional address lines
used to select the additional memory, so physical address size is
increased from 32 bits to 36 bits. This increases maximum physical
memory size from 4 GB to 64 GB."

 

So can anyone confirm that we 'cannot' use 64GB RAM on a 32 bit system
with the PAE kernel, or offer any other advice please ?

 

Kind Regards

Nick Lunt

Managed Services and O/S Analyst

Patech Solutions Limited

Tel: 01543 444 707

Fax: 01543 444 709

Tame House, Fradley Park, Lichfield, Staffordshire, WS13 8RZ

www.patech-solutions.com <http://www.patech-solutions.com/home.htm> 

 

 

_______________________________________________
rhelv5-list mailing list
rhelv5-list@redhat.com
https://www.redhat.com/mailman/listinfo/rhelv5-list

Reply via email to