Hi Geoff

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?

Rob Martin
Software Engineer

phone +64 03 377 0495
fax +64 03 377 0496
web www.chreos.com

Wild Software Ltd



Geoff Worboys wrote:

>>It was my understanding that the hyperthreading issue was
>>resolved in FB 1.5, at least that was the reason we moved
>>to FB 1.5 in the first place and I seem to remember our
>>testing showed that it was fixed.
>>
>>
>
>I am not not familiar with all issues that may have existed,
>but I can confirm that at least one HT issue remains with
>FB v1.5.2 (superserver).
>
>If HT is enabled and you do transfers between two databases
>using non-local connections (eg: TCPIP either remote or
>localhost) then the transfer seems to pause at intervals.
>(Everything just sits there, no CPU activity, nothing). After
>a short time (a few seconds to a minute or so) it wakes up and
>does some more records and then pauses again.
>
>The problem is unpredictable. It does not effect all tables,
>(seeming to effect mostly tables with a large number of rows),
>it does not pause in the same places each time nor for the
>same amount of time. In some situations it can be very severe
>making the transfer so slow as to be impractical, in other
>situations the pauses may be hardly noticed.
>
>Activity on other server connections (to the same or different
>databases to those involved in the transfer) seems to wake-up
>the transfer. So the problem is mostly seen during data-pump
>situations where no other users are operating - like my DBak
>appliciation :-(
>
>On single CPU systems with HT the solution is to disable HT.
>
>On systems with dual physical CPUs plus HT (so you see four
>CPUs in the task manager) it is possible that disabling HT
>will actually make the situation worse! Note that this is NOT
>the old SMP problem, it is specific to newer CPUs with HT
>capability. I have run FB on older SMP systems with no
>problem (with CPU affinity of course).
>
>The only way to minimise the problem on dual physical CPU
>systems seems to be to leave HT enabled and set the CPU
>affinity to at least the two logical CPUs on the same physical
>CPU (0 + 2 or 1 + 3). Even then the pauses seem to occur.
>Setting the CPU affinity to all four CPUs seems to stop the
>pauses but I have not yet worked out whether this then causes
>the old SMP CPU-thrashing problem to occur.
>
>I have seen the problem myself on Win2000 Server, WinXP Pro.
>I have heard other reports of the problem on WinXP and also
>Win2003 server.
>
>I have not seen the problem occur in other situations, such as
>normal user activity against the server, I have only ever seen
>it during data-pump. That is not to say it does not occur,
>only that I cannot seem to produce it in other circumstances.
>
>
>