Eric Barber wrote:
> I need the above to know how to size the database for an application.
> I know the size of the primary key, the row size and the number of
> rows. I've used -
> PCTFILL, primary key (pkey), urowid 10 + pkey, index pointer size,
> block overhead size for cache and transaction layers and for data
> block.
>
> given the number of rows I calculate the number of blocks needed at
> the leaf level using the block overhead and adding the urowid to the
> row size -
> rowperblk = int((blocksize - blockoverhead)*PCTFILL/(row size +
> urowid))
> this gives me the number of pointers needed at the higher levels of
> the tree -
> pointers = int(numberrows/rowperblock + 1)
> and this gives me the number of layers above the leaf nodes -
> layers = int(log(pointers)/log(pkeyperblock) + 1)
> where
> pkeyperblock = int((blocksize - blockoverhead)*PCTFILL/(pkey +
> urowid))
>
> I can now work out a size using various assumptions about the degree
> of fill of the index and leaf blocks as rows are added and deleted.
>
> I've not seen anything on working out the size of index-organised
> tables, and wonder if I'm on the right track here.
>
> thanks, Eric
Why bother? You will always be wrong. Even if you had it
right at the beginning, you will be off next year.
Estimate total length, plus 25%, plus PCTUSED overhead
Add salt.
--
Regards, Frank van Bortel
Received on Tue Mar 18 2003 - 12:00:48 CST
