Hello Diego,
I asked our Oracle specialst from IBM Expert Labs and this is his answer:
The first starting point is always to set up the LPAR and DB according to "best practices". Attached are 2 documents for this from my US colleagues.
Otherwise, you can of course always get the result you want from such tests:
If you look at "With which settings (especially parallelism) do I achieve the best result on x86 (until x86 can't go any further)", and then apply these settings to the Power, x86 may look better, but is 50% idle.
This means that even when testing on Power, you have to play around with the settings to get the best results.
Of course, also the same memory equipment and settings on the DBs, and from the cores (from an Oracle license point of view) with e.g. 8 cores x86 -> 4 cores AIX, SMT-8, and as written, screw around with the settings until the maximum is achieved on AIX.
Simply optimizing on x86 and then measuring 1:1 on AIX will certainly not produce such a good result on Power.
My suggestion:
Depending on how much the customer would like to use Power or you as a BP would like to sell Power/keep the customer on Power, you can also hire someone from Expert Labs to tune the Oracle on Power environment.
------------------------------
Birgit Röhm
------------------------------
Original Message:
Sent: Wed December 13, 2023 02:56 PM
From: DIEGO AGUILAR
Subject: Oracle on Power PoC with Swingbench
Hi Community:
I have a customer looking to run a PoC on Power10, AIX and Oracle 19c database.
As a requirement Swingbench must be used for the load test, in order to compare versus results on x86 platform.
Is there any recommendation for the test on Power10?
Thanks for any comments.
------------------------------
DIEGO AGUILAR
------------------------------