Re: A little baffled by "chained rows"....

> On a table-by-table basis, it's possible to do something as you suggest,
> but that doesn't truly fix the problem. CLOBS merely formalise the process
> of chaining: they still end up being stored 'out of line', with a link
from
> the text/numbers bit of the row to the CLOBby bits. Sometimes, especially
> when CLOB/BLOB data is involved, chaining is just an inevitable fact of
> life; the nature of the beast. And you just live with it.

Hi!

I think that CLOBs can make performance even worse in this situation, beacuse with "normal" storage using index scan for example, you traverse index blocks + do a sequential read of datablock, then find out about chaining, then do another sequential read, etc..

But with CLOB, you traverse index, do a sequential read of datablock to find lob locator, traverse lobindex blocks for finding lob item's inode which contains lobitems physical location(s), then read these locations. (and performance may vary depending on chunk size and whether LOB item has grown after inserting e.g. can the whole LOB be read in with single multiblock IO or several ones).

And you possibly have to recode some parts of your app.

So as Howardman ;) suggested, go with bigger block size and if this doesn't help, then decrease your rowsize problem by redesigning.

Tanel. Received on Tue Aug 12 2003 - 16:30:56 CDT