Oracle FAQ Your Portal to the Oracle Knowledge Grid

Home -> Community -> Mailing Lists -> Oracle-L -> LIOs from 10g when using CPU_TEST.SQL

LIOs from 10g when using CPU_TEST.SQL

From: Tim Onions <>
Date: 2006-01-10 15:42:02
Message-id: BAY105-F26FA822560CC6E3F5214FA99250@phx.gbl


I am using the CPU_TEST.SQL script discussed before Xmas (found on to see what effect moving to 10g will have on our current Win2003 8 CPU systems. Taking into account all the good words said about this script I'm only using it as an indicator. However, the results I get puzzle me. I have 2 DBs on the same server (only one up at any point in time) - a and a Both are 8k block size and everything else as near as the same as I can get them.

I run the script for it takes 3 seconds, does ~30000 LIOs/sec/CPU via 40300 LIOs.
I run the script for it takes .25 seconds, does only 25000 LIOs/sec/CPU but (and here is the confusion) only uses 1500 LIOs!. So the LIO/sec is slower but the overall result is much quicker due to the vastly reduced number of LIOs needed.

I run and re-run this and also checked the data on other Dbs, the same number come up for a DB.

So what am I missing, why does 10g require less LIO? It clearly is much quicker whatever it is doing. Is there something missing from the v$sessstat the script uses (and yes it gets then via stat name not statistic#). It is almost as if 10g is getting mutlple blocks per LIO (in a similar way to how db_file_multiblock_read_count does for physical IO).

Whilst happy with the vast performance improvement I am seeing I need to understand why and in so understanding see how this might help/hinder a real appliation.

Many thanks

Tim Onions

PS I made some very slight adpatations to the original script to display the numbe rof LIOs, total time etc. If you want to see exactly what was run then it is available (for now) on

Be the first to hear what's new at MSN - sign up to our free newsletters!
Received on Tue Jan 10 2006 - 15:42:02 CST

Original text of this message