> I have a customer who uses my accounting system, who has a server (FB 2.5.1) and 4 clients, distibuted and connected via Internet to this server.

The Firebird remote protocol isn't really optimized for WAN usage.

> People there complains about poor performance. I use Firebird Transaction Statistic Logger to see the problem. One and only one of the clients (figured out which) somehow foods my server. While this client is using my software, somehow the gap between OIT and NT increases 200 transactions each 5 seconds. I backup their data each night, and after this I start a gfix.exe -sweep that makes things ok.

Is the OAT stuck as well? If yes, then you have a long-running active
transaction. Use the monitoring tables to identify the offending
transaction.

> the server log (Linux) says a few " INET/inet_error: read errno = 104" problems, but just a few each day.
>
>
> Have anybody encountered something like this ? Is this a hardware or a software problem ?

Usually bad transaction management in the client application. If OAT
moves forward and it's only OIT which got stuck, this can happen with a
write-able transaction which rolled back a huge data manipulation
operation. What's your Sweep Interval setting?

If you are using distributed transactions across two or more databases,
then a failed 2PC transaction will stuck the transaction counters (at
least OIT) as well. But as the sweep cures the problem, I think this
isn't the problem in your case.


--
With regards,
Thomas Steinmaurer (^TS^)
Firebird Technology Evangelist

http://www.upscene.com/

Do you care about the future of Firebird? Join the Firebird Foundation:
http://www.firebirdsql.org/en/firebird-foundation/