PERC 3/QC and memory usage ... Michael_E_Brown at
Tue Jun 24 13:46:00 CDT 2003

You should be running the enterprise kernel and you will get your RAM back. 

It is likely that the PCI bus remaps 512MB of address space to use for your
PERC cards. This PCI address space must be below 4G. This creates a 'hole'
in the address space. This is fine. The only problem is that the SMP kernel
can only address up to 4G of mem. The 512MB that was remapped above 4G is
inaccessible to the UP and SMP kernels. Solution: use the enterprise kernel,
which is almost entirely identical to the SMP kernel with the exception that
it can address memory above 4G.

-----Original Message-----
From: Mark Watts [mailto:m.watts at]
Sent: Tuesday, June 24, 2003 2:56 AM
To: linux-poweredge at
Subject: Re: PERC 3/QC and memory usage ...

Hash: SHA1

> I'm going off what I know so here goes..
> #1 RHAS 2.1 says I have 3.5 g of ram
> The OS should be able to access all of it. This is part of a "highmem"
> in the kernel. You can try passing it as a kernel option but you shouldn't
> need to. It should automatically do this, especially for a new OS such as
> RH AS. 32bit Kernels and programs only have access to 3.5GB of memory as
> the other 500mb of used for Kernel allocation and other 32bit functions.
> Any program (32bit only) including the kernel is only able to access upto
> 3.5GB. This is a limitation of the 32bit architecture. 64bit Architecture
> has higher adderesssing limts, upto 64GB (not 100% sure??) Again I'm not
> 100% sure, this is what I have learnt. Any single process (aka kernel) can
> only use 3.5GB at any one time. All programs started after the kenel (aka
> init) fall under init itself so they too can only get to 3.5GB of ram.
> #2 Memory diags says I have 3.5 g of ram.
> This would be because the Kernel suffers the same limitation. If the
> can only access 3.5GB then any program under will be the same. This would
> also apply for any Dell diags as they are 32bit programs, run via a DOS
> shell or shelldos command boot disk or "utiliity" mode partition. These
> also have the same memory addressing fallback as it's a 32bit program.
> #3 Dell support says that each channel on the QC card take 128 mb of ram
> explaining the .5 g shortfall in memory.
> The Dell PERC cards have additional memory chips on the motherboard. These
> are SDRAM chips @ 133mhz ECC.
> Your normal memory would be DDR @ 300+MHZ ECC (at a guess). So what Dell
> are telling does not sound correct.
> All PERC's I know of have their memory modules on the motherboard seperate
> from the system memory.
> Well the QC is a quad channel card? Correct? So if you were to boot the
> server with 4 x 128mb chips in your main DIMM slots you would essentially
> have no ram for the OS?
> What happens if you boot the server with 4 x 128mb ECC chips? Does this
> mean you boot with 0mb of ram? If what Dell is saying is true that means
> you would boot OS (RH AS) with 0mb of ram as Each of those 4 x 128mb chips
> would be used by the PERC QC. I'd be suprised if the PERC QC could use DDR
> instead of the traditional SDRAM @ 133mhz
> Interesting conversation...
> Anyone have any excact info?
> Hope this helps (dont take all my info 100% correct)

I'm prety sure that you _only_ loose 0.5G to the PCI bus(ses) if you have
ram, regardless of what devices you have. The ram on the PERC cards has 
nothing to do with system ram - its purely a bigger cache than you get on
drives themselves, and does exactly the same thing.


- -- 
Mark Watts
Senior Systems Engineer
QinetiQ TIM
St Andrews Road, Malvern
GPG Public Key ID: 455420ED

Version: GnuPG v1.2.1 (GNU/Linux)


Linux-PowerEdge mailing list
Linux-PowerEdge at
Please read the FAQ at or search the list
archives at

More information about the Linux-PowerEdge mailing list