Subject | RE: [ib-support] Child Tables Problem Revisited |
---|---|
Author | Doug Chamberlin |
Post date | 2001-11-01T23:18:32Z |
At 11/1/2001 04:25 PM (Thursday), Bill Morrison wrote:
Glad to hear my suggestion of committing every write has eliminated the
visibility problem.
The ever growing database problem is a puzzle to me. I thought that there
was an algorithm in place to pack records into available pages which also
re-uses deleted space. Perhaps you have hit on a usage pattern which
defeats this. Perhaps it is an outright bug.
I know there is a configurable option for setting how filled each database
page is when the database is initially built. If pages are not completely
filled then the records which are located in each page have some room to
grow without causing them to need to spill over to a new page. Perhaps
fooling with this "loading" parameter might help. People with more in-depth
knowledge of the engine would have to advise you about that.
> I've modified it so it is doing the transactions the way you described,I think everyone would agree that the two problems are entirely separate.
>and I still have the same problem with the database size creeping up. Note
>that it does cause a small performance issue, but once I resolve this other
>problem I can restore it and see if it's a culprit.
>
> If I watch the allocated pages for the database, it'll increase by 1 or so
>every 500 records. Only the deletions and additions are occuring.
Glad to hear my suggestion of committing every write has eliminated the
visibility problem.
The ever growing database problem is a puzzle to me. I thought that there
was an algorithm in place to pack records into available pages which also
re-uses deleted space. Perhaps you have hit on a usage pattern which
defeats this. Perhaps it is an outright bug.
I know there is a configurable option for setting how filled each database
page is when the database is initially built. If pages are not completely
filled then the records which are located in each page have some room to
grow without causing them to need to spill over to a new page. Perhaps
fooling with this "loading" parameter might help. People with more in-depth
knowledge of the engine would have to advise you about that.