We are running firebird 1.5.3 under Linux Fedora Core 5.

The machine:
Xeon 3.2 GHZ (HT disabled)
2 GB RAM
SCSI disk

We turned off selinux.
We are using the default configuration file.

We have up to 300 concurrent connections all coming from
a windows web server running Apache2, PHP and GDS32.dll from
the windows version of firebird 1.5.3.

Our database starts out empty, but every connection uploads
data with a lot of inserts. Forced Writes are OFF.

Firebird works great for 2-10 hours and then stops accepting
connections. At this point CPU usage goes to almost 100%
and stays there.

Sweeps are set to the default 20000, and they happen regularly before
Firebird dies, I checked.

We left the server up to see if it would recover, it did not recover
after about 10 hours of just sitting there with 100% CPU.

There is plenty of harddrive space.

There are no unusual messages in firebird.log or var/log/messages
firebird.log just has a lot of errno 9 socket errors, but those
happen all the time with Firebird is working properly. When Firebird
stops accepting connections, it does not write anything to firebird.log


Any help would be greatly appreciated.


PS, running our app on the windows version of firebird, we see
worse performance (Forced Writes are ON) and eventually the database
throws BUGCHECK (290) deadlock - error during savepoint backout, could
this be the same problem with different result on linux?

Thank You,