Re: Lazy Write-Back??

In article <James.Lawrence.26.001332F1_at_epamail.epa.gov>, James.Lawrence_at_epamail.epa.gov (Lawrence James) wrote:

>
> What else was going on? Is it possible that it didn't have to do physical
> reads for the first part of your second query because the DB block buffer
> already had them from the first query. Then on the third query it couldn't
> find them in the buffer so had to read them again. This would happen even
> without any other activity if you buffer is smaller that the total blocks
> since the least recently used blocks would have been the first part of the
> table and would have been reused for the end of the table in your second query.
>
> If you want a repeatable measure of physical reads you need to flush any
> possible existing blocks. I do it by selecting from a really large table
> unrelated to my test. That fills the buffer up with that table and I know
> that none of my test tables are in it.
>

It is my understanding of Oracle, that full table scans put datablocks at the tail end
of the LRU chain. So it shouldn't matter. All 3 queries were full table scans, and since
there was no index on the table (and it was larger than the small_table_threshhold)
no activity on the table should persist in the cache.

Any other ideas?

Chip Dawes Received on Tue Nov 01 1994 - 19:23:29 CET