Subject RE: [IB-Architect] Fw: Slow deletion on large tables
Author Ann Harrison
At 01:31 AM 5/29/00 -0400, Claudio Valderrama C. wrote:
> Thanks. Now it's clear blobs in this case aren't a great performance hit.
> How about updating blobs, please? It's a subject that has me "on hold".
> Also, when you say "If the blob is on page, release the
>line index indicator.", do you mean in the same page than the rest of the
>record? What's the size limit for that, the available space on the page, the
>segment size, some formula?

Yes, "on page" means on the same page with the primary record version.
A blob will be stored on page if there is room on the page for the
whole blob.

> > The index duplicate problem is architectural, but solutions exist.
> Then we need to collect a list of recommendations, probably.

Without seeing the code and thoroughly understanding the problem, finding
a solution will be very difficult. Essentially the problem is that
when removing an entry from an index, the database must determine which
of a list of duplicate key value entries represents the row being removed.
Since duplicates are stored LIFO, deleting an old row often requires
reading most of the chain.