At 07:10 AM 22/07/2004 +0000, you wrote:

>Hi folks,
>
>today I noticed some strange "performance behaviour".
>
>I ran a SELECT command like
>
> SELECT ID, ... FROM SomeTable WHERE ID = SomeID
>
>on a database running in CS mode (FB 1.0.3) on a Linux box. The
>client was W2K.
>It was rather slow. 500ms execution time.
>
>Running the same command from a different client on the same server
>and database returned after 20ms.
>
>This is reproducable.
>
>Running the same command on the same database (backup/restore) on a
>different Linux box returned after 20ms.
>
>Now it starts to get funny:
>
>Running "almost" the same command on the original server and database
>gave me 20ms!?
>
>What did I change?
>The original command contained 43 columns. I just added another ID
>column at the end of the column list.
>That's all.
>
>The same thing happens when I reduce the number columns to
>42 (the answer to everything ;-)
>or less.
>
>BTW, the overall number of columns of the table is 43!
>
>And so it's probably no surprise that "SELECT * FROM SomeTable" is
>also slow.
>
>Very strange.
>Any idea?

First thought is that, without the extra column, the record size fits into
one database page, while the extra 8 bytes tips it over into a second
page. Perhaps the different versions of the database have different page
sizes?

Or perhaps the restore cleaned up some accumulated back versions and/or
indexes...

/heLen