Hi Ferda!

Does this slowdown appear after a massive delete or update (regardless
of which transaction deleted/updated the records), maybe on a table
with low selectivity indexes?

I just seem to remember reading something about it being a good idea
to run a select count(*) on tables after doing such deletes, since
garbage collection otherwise could be left to a more or less random
transaction.

If this is your problem, then others on this list will know lots about
it (it should be a pretty common "problem"), it just happens to be one
of the things that very rarely affects me since I hardly ever do
massive updates/deletes.

Since no-one have answered you, I guess it is more likely that this
only was a problem some versions ago and that it is completely
irrelevant for your problem. But it's better to receive possibly wrong
advice than no advice, so I'll send it to the list just in case,

HTH,
Set

--- In firebird-support@yahoogroups.com, Ferda Brabenec wrote:
> Hi,
> we are running Firebird Superserver 1.5[1-2] on a Linux server. It
> serves data to a web application. We are accessing FB from a
> multithreaded CORBA backend server through ODBC (UnixODBC driver
> manager + Easysoft ODBC FB/IB driver).
>
> Everything works fine until moderate load appears. Then randomly
> some of SQL* calls are processed 100-1000x slower then usually. We
> started to log timestamps of some execution points when the
> processing gets too slow. It seems that the slowdown may happen
> with any arbitrary ODBC call such as SQLNumResulCols,
> SQLDescribeCol, SQLBindCol, SQLExecute. Sorrounding ODBC calls from
> a given thread are regularly fast.
>
> This slowdown appears even if a low number of db connections (such
> as 5) is being used. Our proprietary connection pool allocates
> additional connections when needed but this is not the case.
>
> We have tried 2 Xeon-based servers with various setup combinations
> of HW : single/double cpu, , HT on/off
> kernels : 2.4/2.6, smp/non-smp, pthread/nptl
>
> and also we tried to collocate/dislocate application modules on 1-3
> hosts and nothing seems to affect these slowdowns.
>
> Classic server perfomance was noticably weaker so we did not
> experiment with that build at all.
>
> So I want to ask if anybody has/had similar problem or if anybody
> has a clue how to cope with this.
>
> TIA, Ferda