Hi Ivaylo,

>>At a specific time in the month some of the clients perform big and
>>ugly queries, which take more than 30 minutes to finish and return

results.

>>These SQLs are blocking the server.

>>At the same time the client applications are using a large number of
>>small queries (I mean small, fast SQLs), which are essential for the
>>UI manipulation.

>>So the blocked server is practically stopping the work of all the
>>clients. That's what we are trying to avoid.

We had the same issue, and almost changed over to classic server, but a
couple of small changes suggested by Helen resolved our issues to the
point that our client s stayed on SuperServer.

Firstly, we upped the sweep interval, so that there was less chance of a
sweep occurring in the middle of a query. We had it set to something
stupid. Changed it back to the default of 20000.

Next we had a look at our transaction handling. Changed over to
read-only transactions where the queries were not modifying any data.
That improved our "oldest active transaction" and "oldest interesting
transaction" statistics enormously.

All our queries were indexed properly, but going back and have a 100th
look at them now and again eeks out a bit more performance enhancements.

Now the server's CPU still runs at almost 100% capacity, but the other
users do not notice much speed degradation while the big disgusting
queries are running.

Cheers
Maya