On Thu, 2005-04-07 at 15:47 -0700, Nishanth Aravamudan wrote:
> ppc64: 8-way pSeries (1.5 GHz Power4) with 64 GB RAM
> x86: 4-way NUMA-Q (360 MHz PII) with 3GB RAM
>
> Please ask questions/request more details, if you need them.

It would be interesting to see these on slightly more modern hardware.
That ppc64 machine has *HUGE* L2/L3 caches, and the "NUMA-Q" really
behaves like a regular-old 4-way PIII Xeon.  Plus, its 2MB L2 caches
will hide lots of cacheline bouncing problems, and some of that is
certainly going to be an issue when you bloat a data structure like
'struct page'.

I'd suggest trying these on a couple of different pieces of hardware.
First, a real NUMA NUMA-Q.  A 16-processor 4-node system should do.
Secondly, a non-Xeon x86 system.  These have smaller caches, and will
have the cache problems show more effectively.  Finally, as large and
NUMA-ish of a ppc64 system as you can get.

Also, all of your results appear to have relatively bouncy results.
Perhaps you can run more iterations and post the averages.

Lastly, I wouldn't consider these results to be too valid at all until I
see the system loaded up with a bunch (say 100 or 1000) CKRM classes.
There are a ton of linear searches, and I look forward to seeing them
choke.

-- Dave



-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click
_______________________________________________
ckrm-tech mailing list
https://lists.sourceforge.net/lists/listinfo/ckrm-tech

Reply via email to