Hi

It is a threading issue that shows on multi CPU / dual core and even
single core hyperthreaded CPUs. Setting the affinity fixes the problem
in so much as the problem is significantly reduced (clients don't mind
55 instead of 45 minutes).

The variability seemed (we moved to FB 2) to be around the record size /
table size being processed. Sometimes the effect on small tables was
worse than big tables.

There has been discussion on the FB list about this issue. It was
something to do with the OS moving a thread from one 'core' to another
to improve performance and then moving it back again. Creating a
bouncing effect where the thread itself got little CPU time and the OS
used a lot. Supposedly effects other areas of FB but in our experience
it was worst with IB_Pump.

Rob


Geoff Worboys wrote:

>> Geoff is correct. This is actually a FB 1.5 (<=) issue that
>> is mostly resolved in FB 2. The issue appears on Multi CPU
>> / Core cpus with SuperServer. We had a DB that took 45mins
>> to pump, putting it on a dual core machine pushed this out
>> to 1.5 days. I think setting the CPU Affinity fixed this
>> issue. I would guess that the Vista machine is also
>> a new Dual / Quad core cpu.
>>
>
> Can confirm that it is a multi-cpu (even just hyperthread)
> issue. However cpu affinity does not fix the problem - it can
> alter the impact, but does not fix it.
>
> The variability of the problem is quite strange. Some days a
> transfer of a database will take place quickly and without
> apparent delays, on other days the same database can take
> hours. (Dont think I have had one take 1.5 days... thankfully.)
>
>