| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
 |
 | |||
| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
|
| |||
Home -> Community -> Mailing Lists -> Oracle-L -> Re: ASSM in 10g RAC doesnt seem work that well
Christo,
To be 100% sure all timings and statistics I am collecting are correct I did my tests with and without 10046.
> In my testing if you run with 10046, the waits are >
lost in the CPU cost of maintaining the trace.
Here, as I said I did not concentrate on waits at all.
Waits are something that are not reproducable across
different systems, even on the same HW/SW you are not
100% sure everything is the same.
But again on my system ASSM outperform non-ASSM anyway
with or without waits.
CPU usage was lower and waits were lower.
I saw that Tom Kyte tests on Linux also gave him the
same CPU usage.
Maybe something related to the way how bitmap
management is done.
Because of it I concentrate on the pure CPU usage that
I found different.
In your tests on Linux your CPU usage was the same for
both tests.
For me it was different.
I heard that ASSM is going to help in high-DML concurrent environments bu have never seen one article telling that ASSM is making CPU usage improvements.
I have got a big difference in pure CPU usage when using ASSM over non-ASSM in LMT environemnts.
> Did you repeat the exact same test ? With 2 inserts
> on different nodes
YES. I also did the tests on the single instance with
9i and 10g just to check that the CPU time is
different.
> ? non-ASSM was still a LMT tablespace, right ?
YES, because I could not find one my database (have
around 20 in my env) with dictionary based SYSTEM
tablespace, so could not have anything except LMT.
What is left to me, to check carefully one more environment I have, HP-UX and also to make the test with the bulk DML.
going to check this thing because it is going to overcalculate/underestimate all my future capacity sizing figures, that are going to be affected if this ASSM/non-ASSM is so different in CPU timing.
Regards,
Zoran
> Martic,
>
> In your test, did you run with 10046 ? In my testing
> if you run with
> 10046, the waits are lost in the CPU cost of
> maintaining the trace.
> Thus my sampling approach .
>
> Did you repeat the exact same test ? With 2 inserts
> on different nodes
> ? non-ASSM was still a LMT tablespace, right ?
>
>
> Christo
-- http://www.freelists.org/webpage/oracle-lReceived on Tue Mar 29 2005 - 03:24:44 CST
 |
 |