Using gstat to check, the sweep interval is already set to 0 on these
database files. I do a nightly gbak to make a backup. I'll check into
the nightly sweep idea.

Alexandre Benson Smith wrote:

> tmc@... wrote:
>
>>Am running Firebird 1.5.3 Classic on Redhat 8.0 with a Dell 2650 (dual 2.4GHz,
>>4MG RAM). Have about 15 million records in various tables. One in a while
>>(like once a week), the database becomes very busy and slows to where MS-Access
>>programs can't connect properly (using open source odbc V.2.128). After about
>>20 minutes, the load average (which gets upwards of 28) calms down to around 2
>>and the system is not as busy. Requests start to be processed at some point
>>during this, so it's not quite an 'outage', but becomes very slow.
>>
>>What is this doing? We were on Interbase before migrating to FB1.5 with a
>>backup and restore. It had the habit of becoming slow over time. Then we
>>would do a backup and restore in interbase and it would get faster again.
>>
>>What are the mechanisms at work here and how can I tune them to not cause an
>>interruption of service? Even knowing what parameters to monitor would help
>>predict when the interruption would likely occur.
>>
>>Thanks!
>> Tom
>>
>
> Tom,
>
> I think it's due to the sweep process that get started because you reach
> the sweep interval (defined per database).
>
> How to check it:
> Look for your database statistics (use gstat) when the server becomes
> slow and watch the gap between your transactions.
>
> How to avoid it:
> 1.) Keep your transactions as short as possible
> 2.) Turn off the automatic sweep setting the sweep interval to 0 (use
> gfix for this)
> 3.) Run every day (night) a manual sweep to remove the garbage left on
> the database, schedule it to a low usage period.
>
>
> see you !
>