On Tue, 2008-02-19 at 00:25 +0000, Sharpe, Sam J wrote:
> I believe the OP said that fully updated RHEL5/4 were consistent, but
> not as "fast" as Fedora 8 or non-updated RHEL4 systems.

My interpretation was that ran his test application on the Intel
hardware using the most recent versions of RHEL4, RHEL5, and CentOS, as
well as other OS's, and seen a significant difference in performance.
He named a specific time (9 minutes) that this system ran when using the
"latest" versions of RHEL4/5/5.1/CentOS5.  I tried to compare my updated
systems which had very similar hardware (dual-core Intel) which should700
give somewhat similar results, however, all of my results were under 3
minutes, no where close to his 9 minute result and more in line with the
results he saw from the other distros.  This implies to me a strange
interaction with the RHEL kernels and his hardware.

If he had stated that an older version of RHEL4 or 5 on the same Intel
hardware ran faster than the "latest" version then that would have led
me down a different path, however, he didn't say that.  He compared his
"slow" results on the Intel hardware to an older RHEL4 kernel on
completely different Opteron based hardware, which isn't all that
useful.

Of note, the Dell 6850 result which I posted earlier was actually Redhat
4.4 with kernel 2.6.9-42.0.10smp.  I though it had been upgraded but I
was incorrect.  Sorry for the mistake.  However, that makes it pretty
close to his Opteron system with 2.6.9-42.0.2.ELsmp which he stated ran
in 2m 45sec.  The Intel based 6850 results with a very similar kernel:

$ time ./sorttest 20
real    2m30.838s
user    2m29.109s
sys     0m1.687s

Based on the consistency of my results I can't imagine why he is seeing
such a disparity in performance between distros unless there is
something unique about his hardware that is interacting poorly with
RHEL4/5 kernels.  Perhaps he has a NUMA setup this is doing strange
things with his cache/memory with the RHEL kernels.  More info on the
hardware would help and perhaps a post of the dmesg output.

Later,
Tom


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