> Awesome fault report! This exactly the problem we experience.
> We set up a new empty database and then pump data from another
> database into it (TCP/IP connection), we get the pause problem.
> With hyperhtreading this database pump takes 20hrs+ on one
> clients machine. Without HT it takes about 10minutes. Our
> fixes are similar to Geoffs.

> I really hope this is fixed for FB 2. Do you know if it is
> scheduled to be Geoff?

Hi Rob,

I brought this up on the fb devel list recently. The core
developers are aware of the problem but...

Since disabling HT on dual CPU systems makes the problem worse
it would seem to indicate that the problem is not really HT.
The problem does not occur on older (PIII) dual CPU systems so
it must be some other feature on these new HT capable CPUs.
My guess is that it may be Intel specific - perhaps instruction
reordering or some such.

None of the core developers have access to a suitable machine
to try and reproduce the problem and test solutions. Jim
offered his dual CPU system, but its not Intel.

It is possible that the solution may simply be a change in
the compiler settings (I think the new compilers have stuff
relating to HT and instruction reordering etc). Or it may be
that there is a real timing problem involved here. Until one
of the core developers that really understand this stuff is
able to investigate the problem in detail I am guessing that
the problem may hang around (or it may mysteriously disappear
as a result of other changes - you can always hope :-).

--
Geoff Worboys
Telesis Computing