On 10/26/06, Gavin Maltby <[EMAIL PROTECTED]> wrote:
for the t2000 to be shown the best light you need to run 24-32 copies of the single threaded app to shine.
James Dickens
uadmin.blogspot.com
Hi
On 10/25/06 10:24, Gino Ruopolo wrote:
> Hi All,
>
> ----- THIS IS NOT A FLAME!! ------
I'll take your word for it :-)
>
> I'm doing some serious comparison to decide what to do with some old USIII 900 servers.
What do you mean by the 900? 900MHz US-III - you quite 750MHz below.
> My application uses about 3MB of memory and is singlethreaded (for that test). All integer operations on a 400MB memory mapped file.
Sounds like memory config would play a part, too.
> Here is what I'm having (All systems wiht S10u1 or S10u2):
>
> gino$ time ./doit
ptime(1m) is supposed to be more reproducable, but does not time child processes (are there
any?) of the command; you could also look at timex(1m)
> Opteron 848 (2200Mhz) 5m 21s
> USIII 750Mhz 18m 01s
> Xeon 2.4 Ghz 35m 12s
> P4 2.0 Ghz more than on hour
> T2000 slow, really slow for that app :(
for the t2000 to be shown the best light you need to run 24-32 copies of the single threaded app to shine.
James Dickens
uadmin.blogspot.com
>
> Is that possible?? a USIII 750 faster than a Xeon 2.4Ghz??
> Could that be related to cache size?
Yes, or any number of other things. Even the old sparc processors have bigger l2 caches
than all but the very latest x86 cpus. The caches are different not just in size
but in architecture - where misses are filled to, indexing, cacheline size, associativity etc
and it looks like the combination on the USIII 750MHz *may* have suited this app very nicely.
But you also need to consider other factors such as system memory size. Was it the
same OS version used in each case?
> Also I've seen that Opteron 4CPU boxes are going slow when there are a lot of process runnning ...
That's never been my experience.
Gavin
_______________________________________________
opensolaris-discuss mailing list
[email protected]
_______________________________________________ opensolaris-discuss mailing list [email protected]
