In message <1140706118.355705.30350_at_t39g2000cwt.googlegroups.com>, Mark
D Powell <Mark.Powell_at_eds.com> writes
>To expand on Ken's response. If you need to make the estimate before
>the tables are build and test data or better yet some real sample data
>can be loaded then
>
>For single byte character sets
>
>Each date is 7 bytes for the date plus a null indicator so 8 bytes
>
>Each varchar2 column less than 128 bytes is should be defined as the
>average length of its contents plus 1 byte for the length. For lengths
>of 128 bytes and up I believe that you need 3 bytes for the lengh/null
>indicator.
>
>Number are roughly the average number of digits / 2 plus 1, but
>negative numbers need an extra byte.
>
>Every row has a 3 byte header and every data block has both fixed and
>variable overhead. Using 104 would not be a bad guess.
>
>If you take the average row size times the expected number you can then
>divide by the Oracle block size minus the overhead to get an idea of
>how many blocks the table will require. You can then adjust that for
>your storage allocation method: uniform extents, autoallocate, or
>dictionary.
>
>HTH -- Mark D Powell --
>
And don't forget to allow for indexes. A single index may take as much
space as a table. The database at my current client uses three times as
much for indexes as for tables.
--
Jim Smith
I'm afraid you've mistaken me for someone who gives a damn.
Received on Thu Feb 23 2006 - 13:51:31 CST
