There are some problems in doing this:
- v$sql gets reset when SQL is reloaded
- rows_processed does not reflect work being done, but
logical reads do. I am not sure how you can corelate the
two. For example, if you do a select count(*) from emp,
rows_processed will be 1, but logical reads would be as
many blocks as the full table scan needs to cover.
So this ratio would be meaningless in this case and it
skews your average.
Overall, I think # of LIO per second is a very useful measurement
which I also collect it myself to understand the workload profile
and trending. But I am not sure averaging that into # of rows
processed (rows returned is a more acurate term) tells you much.
Received on Fri Jun 03 2005 - 13:09:17 CDT
