RE: unaccounted for time in a tkprof output
Date: Mon, 10 Aug 2009 16:22:37 -0500
Generally the unaccounted for time breaks out into these five buckets:
1 Measurement intrusion effect (> 0, is negligibly small)
2 Quantization Error (¡ú 0, is negligibly small)
3 Time spent Not executing (unconstrained)
4 Un-instrumented time (¡ú 0, if your kernel is patched)
5 CPU double-counting during OS calls (¡Ü 0, is negligibly small)
Most likely it¡¯s #3 and it would make some since that you are on the CPU but not executing as much on Production as you do in Test since I will make the grand assumption that there are a lot more folks doing things on production then there are on test. Oracle doesn¡¯t know how long you spend on the CPU but in the ready to run queue and not actually on the CPU doing work. The more folks on the system, the more time you will likely spend in the queue while on the CPU.
If you want to know more about this, get the book ¡°Optimizing Oracle Performance¡± and/or come to a Hotsos training class (Oracle Performance Management
Using Response Time Profiling) where we cover this in great detail.
Ric Van Dyke
March 7 ¨C 11, 2010
From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Dba DBA
Sent: Monday, August 10, 2009 4:23 PM
Subject: unaccounted for time in a tkprof output
I have this on otn. I didn't want to copy it to email since I can't properly format the tkprof outputs.
How do I have extra elapsed time with out wait events to account for it? I am trying to figure out why inserts take longer in production and I don't have a wait to go on.Received on Mon Aug 10 2009 - 16:22:37 CDT