SUMMARY: Strange Performance results E450 vs E6500 (with SRM)

From: Alex.De-La-Salle@nmp.nokia.com
Date: Mon Apr 26 1999 - 11:34:15 CDT


Original Question:

E450/E4500 (4x 400Mhz CPU/4GB) on average performs a compilation in half
the time of a E6500 (10x 336Mhz(6offline)/10Gb). Veritas VolMgr and VxFs
in use on both systems, disks were RAID 0+1, and swap was 2Gb on each
system. Solaris Resource Manager was used to control CPUs and VM to try
and replicate the E450/E4500 configuration, however this only meant a
further 10% added to the compilation time. With SRM removed from the
E6500 and the CPU's offlined so there were only 4 online (to avoid
performance loss due to the compilation running on different CPUs
throughout the compile time (lots of switching)) the gain in performance
was 10% as expected.

 Machine Time for Compile

 E450/E4500 28 mins
 E6500 56 mins

SUMMARY:

Well, thanks to all that responded, unfortunately we have been unable to
investigate it further and are unlikely to at this stage.

Mostly it was mentioned that the bus & cpu speed differential would show
a substancial difference in performance, approx 20%, but you should not
see a 50% differential. We were aware of that the E6500 has a 82.7Mhz
bus where as the E450/E4500 has a 100Mhz bus, but did not expect to see
such a difference.

Another possibility mentioned was that the I/O could be the bottle neck,
but we had used BMC's BEST/1 and iostat/vmstat/sar to check for
performance bottlenecks and the I/O didn't seem to be one, see as the
E6500 was using A5000 with 10Krpm disks, I wouldn't expect to see too
much problem there.

There was obviously something in the way the machine had been setup that
caused the problem, but unfortunately as I said, time ran out for the
performance testing so we couldn't persue the E6500 tests any further.

So alas, there is no conclusion to this question, the only thing to note
is that another application was tried on the E6500 and the E450/E4500 and
the performance difference was around 12% in favour of the E450/E4500...
so this looks to be application dependant!

Thanks again for the input.

Rgds, Alex

-----------------------------------------------------------------------

Alex de la Salle
Nokia Mobile Phones
Tel: 01252 866521
Mobile: 07970 117647
Work Email: Alex.de-la-Salle@nmp.nokia.com
Home Email: alex@openit.clara.net

-----------------------------------------------------------------------



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:13:18 CDT