Subject | Re: Strange performance with FB 1.5 - Help |
---|---|
Author | bazarin |
Post date | 2003-12-30T17:12:16Z |
--- In firebird-support@yahoogroups.com, Daniel Rail <daniel@a...>
wrote:
I did the change in the CPU bios now and run the test again. This are
the results:
1) The processing time for one client decreased to 1:30 minutes (25%
less).
2) Running the system with 4 clients the time continued almost 4
times more (5:20 minutes).
3) Looking at the log for each client we could noticed:
3.1) Although each client started at the same time (we put 4
engineers starting the application at once in each microcomputer) the
second client started 17 seconds after the first, the third started
57 seconds after the second and the fourth 40 seconds after the third.
3.2) It seems that the FB could not start the four clients at once.
Is there something in the FB configuration that we could change to
solve this?
Regards,
Wagner
wrote:
> Hi,2:22
>
> At December 30, 2003, 08:07, bazarin wrote:
>
> > I did the changes in the FB configuration file as below:
>
> > CpuAffinityMask=1
> > DefaultDbCachePages=10240
> > EventMemSize=655360
> > LockRequireSpins=0
> > DeadThreadsCollection=50
>
> > These are the changes done from the default configuration.
>
> > We runned the application again, first with one client (it last
> > minutes) and with the all 4 clients that last 9:10 minutes. Itseems
> > that changing the CpuAffinityMask from 3 to 1 the total timeDaniel, thank you!
> > increased instead.
>
> Did you turn off the CPU's HT feature from the bios, as Aage asked?
> There are issues using Superserver on a processor with HT enabled.
> These issues are going to be addressed in FB 2.0.
>
> --
> Best regards,
> Daniel Rail
> Senior System Engineer
> ACCRA Group Inc. (www.accra.ca)
> ACCRA Med Software Inc. (www.filopto.com)
I did the change in the CPU bios now and run the test again. This are
the results:
1) The processing time for one client decreased to 1:30 minutes (25%
less).
2) Running the system with 4 clients the time continued almost 4
times more (5:20 minutes).
3) Looking at the log for each client we could noticed:
3.1) Although each client started at the same time (we put 4
engineers starting the application at once in each microcomputer) the
second client started 17 seconds after the first, the third started
57 seconds after the second and the fourth 40 seconds after the third.
3.2) It seems that the FB could not start the four clients at once.
Is there something in the FB configuration that we could change to
solve this?
Regards,
Wagner