Hi,

At May 22, 2003, 04:23, Fabrice Aeschbacher wrote:

> I ran the test again during last night, this time with smaller video
> slices: I insert a new video record for each 1MB read from stream.
> Once again, this completely filled up my 20GB disk during the night;
> the "real" total blob size is 1GB.

> Below the interesting part of gstat output (when the disk is full):

> Oldest transaction 40298
> Oldest active 40299
> Oldest snapshot 40247
> Next transaction 41004

Have you tried to perform a sweep? I know you have sweep set to 1000,
but that simply means that the sweep is performed after 1000
transactions. Also, is it possible that you have a transaction that
is still active, because your oldest active and next transaction is
700+ transactions apart.

> VIDEO (131)
> Primary pointer page: 144, Index root page: 145
> Average record length: 13.69, total records: 4029
> Average version length: 51.99, total versions: 3018, max
> versions: 1
> Data pages: 1330, data page slots: 1341, average fill: 24%
> Fill distribution:
> 0 - 19% = 1111
> 20 - 39% = 83
> 40 - 59% = 1
> 60 - 79% = 0
> 80 - 99% = 135

> By the way, SELECT COUNT(*) FROM VIDEO returns 1004, and not 4029.

Probably the total records includes the record versions.

> I really can not understand what I'm doing wrong. I cannot imagine
> that what I'm doing is not possible with FB. The problem is that I
> cannot spend a month with this

Reading your earlier message, you mention each process use one DB
connection. Do you mean one connection for DELETE, another connection
for updating the data and another connection for SELECT COUNT(*)? Or,
they all use the same connection? Also, what are you using for
connection in your application?

--
Best regards,
Daniel Rail
Senior System Engineer
ACCRA Group Inc. (www.accra.ca)
ACCRA Med Software Inc. (www.filopto.com)