[CentOS] [OT] Memory Models and Multi/Virtual-Cores -- WAS: 4.0 -> 4.1 update failing

[Peter Arremann]

On Saturday 25 June 2005 13:02, Les Mikesell wrote:
> On Sat, 2005-06-25 at 01:27, Bryan J. Smith wrote:
> > But at this point, let's just forget this whole thread because it's
> > obvious that you're not interested in hearing me out, just hearing
> > yourself out.
>
> Some real-world benchmark numbers would make the case more
> convincing.  Does anyone have some?  I'm particularly interested
> in anything with AMD vs. IBM's 64-bit xeon boxes.
The talk we had wasn't really about 64bit... 

There are a few benchmarks out there comparing the two - but none I've seen 
actually test 64bit linux tuned for Opteron/Xeon (instead of the stock RHEL 
kernels, see below) with large memory (>4GB) running heavy IO. That's where 
you'd see the biggest issue with intels implementation because of the lack of 
IOMMU. 

Here is a list of benchmarks and why they are not idea to demonstrate the 
differences in how the 64bit extensions work:
http://www.anandtech.com/IT/showdoc.aspx?i=2447 - No heavy IO, no opteron 
tuned kernel but EM64T kernel for both. 
 http://www.gamepc.com/labs/view_content.asp?id=pciews&page=1&cookie%5Ftest=1 
- Windows, 32bit. 
http://www.anandtech.com/IT/showdoc.aspx?i=2347 - windows, 32bit
http://www17.tomshardware.com/cpu/20040927/index.html - windows, 32bit.


More interesting my opinion is the tuning though... There was a couple of 
threads about mtune differences between k8 and nocona (i.e.  
https://www.redhat.com/archives/fedora-devel-list/2005-June/thread.html#00719)
and the RHEL3U2 release notes where EM64T was first supported 
(http://www.redhat.com/docs/manuals/enterprise/RHEL-3-Manual/release-notes/as-amd64/RELEASE-NOTES-U2-x86_64-en.html 
- look in the kernel info section).

Peter.