On 18 Feb 2008, at 22:53, John Summerfield wrote:

> Sharpe, Sam J wrote:
>> On Tue, 2008-02-19 at 01:52 +0900, John Summerfield wrote:
>
>>> Seriously, how long to run eight of them at once?
>>>
>>> The number of CPUs and cores is irrelevant unless you're running
>>> multiple threads/processes.
>>>
>>> I would speculate that F8 and opensuse 10.2 have newer gcc and
>>> that that makes a significant difference.
>> I think the issue was:
>> Fast: ( big hardware + (opensuse 10.2 or Fedora 8)) OR (slow
>> hardware +
>> older RHEL4)
>> Slow: ( big hardware + (RHEL5 or latest RHEL4))
>> Note that (Slow Hardware + older RHEL4) beats (fast hardware + latest
>> RHEL4) - the other comparisons aren't interesting.
>
> See my previous note. The number of cores/CPUs is irrelevant, and we
> don't know the capabilities of the individual processors.
>
> Running eight together will show the benefit of the additional CPUs
> and cores.

see my next note...

>
>> I'm not into reading C, so I have no idea what the test case
>> attempts to
>> do, however:
>
> My quick impression is that it's a benchmark of a kind intended to
> exercise some aspects of the CPUs capabilities.

err, yes. Got that far, just wasn't checking /how/ it did that.

>
> Changing compiler _and_ CPU doesn't give a valid measurement of the
> CPUs, though it _is_ a valid measurement of the benefits of changing
> the hardware and software platform, both.

I should have been clearer. My test systems were an HP DL385 running
RHEL5: It has two Opteron 250 CPUs and 1GB of RAM and an HP xw9300
running Fedora 8: it has two Opteron 248 CPUs and 4GB of RAM. The
systems aren't totally comparable because one is a rackmounted server
while the other is a high-end workstation, but they do at least use
the same memory technology and have roughly equivalent CPUs. In this
case the notable difference is:

Less capable system: F8 32 bit + 4GB RAM = faster
More capable system: RHEL5.1 64 bit + 1GB RAM = slower

Naturally RAM was the one thing I looked at, because that is the place
the xw9300 scores highly. It also has better interleaving as it's 8 x
512MB, rather than 2 x 512MB - the next steps are looking at the glibc
and compiler versions and knowing (which I don't) if the fact my
"slower" system is 64 bit makes any difference to what that code is
attempting to do.

If I had exactly matched hardware running different OSes I'd give that
a look but this was as close as I could get ;o)

--
Sam

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