Re: Fragmented vs. Contiguous vs. Single-extent segments

Bruno-

No, I don't think you are really missing anything. The original post referred to research performed simply to ascertain the performance impact of fragmentation on full table scans. I think it is fair to surmise that random IO is unrelated to this issue.

Random IO means that a head positioning operation is going to take place for each fetch and performance will average out over time. Although Nick suggests that full table scans don't normally happen, for decision support systems they happen all the time.

Check the related never-ending thread on defragmentation and find the recent post that references an excellent paper by Juan Loiza. He talks about optimal extent sizes and performance. Don't discount this paper. Juan is a serious heavyweight inside Oracle and is no intellectual slouch. Good reading, too.

• ricky

Bruno Jargot wrote:

> On Sat, 14 Apr 2001 17:14:50 +0100, Nick P wrote:
>
> >Interesting, but not a 'real life' experiment. Very few systems will do full
> >table scans. I would more interested to see how access speeds using indexes
> >would be affected.
>
> The fragmentation should not have any impact on speed when the access
> is done by rowid or I miss something.
Received on Sun Apr 15 2001 - 18:04:04 CDT