Oracle performance on HP-UX 11iv3 and HT enabled

[Brem Belguebli]

Hi all,

Has anyone already run comparative benches on Oracle between 11iv2 IA64 and 11iv3 IA64 (hyperthreading enabled).

HT not being available on 11iv2, I was wondering what could be the gain or the loss to upgrade to 11iv3 with HT enabled on a pure CPU point of view.

I know the 11iv3 IO stack has completely changed, and the gain is visible on this point of view.

My question concerns pure CPU aspects.

Thanks

[Brem Belguebli]

No answers...

At least, are there any Oracle DB running on 11iv3 with HT enabled ?

[Steven E. Protter]

Shalom,

You must deal with general Oracle performance tuning. HT enabled is a standard feature on Itanium Class and other Intel based CPU's.

Oracle tunes itself reasonably well and takes advantage of this configuration.

http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1259497

http://www.scribd.com/doc/7089/HPUX-Kernel-Tuning-Guide

http://wordpress.com/tag/hp-ux/

SEP

[Brem Belguebli]

Salam Steven,

HT doesn't work (no plan to backport from HP) on 11iv2 IA64 but only on 11iv3 IA64.
On other intel CPU's like last Xeon, HT is not active.

My question was about a potential pure CPU gain I could get by upgrading to 11iv3 with HT enabled.

Brem

[VK2COT]

Hello,

Lab tests show that enabling HT on HP-UX 11.31
by default is not a great idea.

If the application is not HT-aware, it actually
can affect the performance in negative way.

Cheers,

VK2COT

[Hein van den Heuvel]

Enabling HT on Itanium has the potential to improve THROUGHPUT for CPU bound Oracle applications under HPUX.

The proof is in the public HPUX / ORACLE benchmark subnmission. For example:

4M tpc/s TPC #107022701 :

http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=107022701

30K users SAP=SD #2006089 :

http://download.sap.com/download.epd?context=40E2D9D5E00EEF7CD8B6D3F654B770E65ECB8C760DF622B83267AA820A5C171B

If HT did not help, it would not have been used for those results. Trust me :-).

Mind you, _heroic_ tuning efforts may or might not have been used for those benchmarks. Notably process-to-cpu bindings to facilitate rad-ness and or thread-co-locating.

hth,
Hein van den Heuvel
HvdH Performance Consulting.