Subject Re: Fb/1.5.2 and Win2003 - first impressions
Author Steve Rice
--- In firebird-support@yahoogroups.com, Aage Johansen
<aagjohan@o...> wrote:
> (Not a support q., but anyhow...)
>
> I'm in the process of moving databases from an old (DELL) machine
to a new
> one (DELL 2850). The old one has 2 Xeons (approx. 800MHz) and
512MB RAM,
> the new one has 2 Xeons (3GHz, and larger cache than the old ones)
and 2GB
> RAM. Disk speeds are the same: 15000rpm (but different sizes).
Probably
> different specs for the RAID controllers as well - both are
DELL's. Disks
> are mirrored.
> The old box is running Windows 2000 and Fb/1.0.3, the new one has
Windows
> 2003 (SP2 I believe, it says version 5.2) and Fb/1.5.2 (SS).
> [Yes, I can see a huge gain in speed with a faster server and a
newer
> Firebird version]
>
> I haven't switched the HyperThreading off yet - just to see the
effect of HT.
>
> I observed cpu activity with the TaskManager while restoring a
database.
> Filesize of gbk: 1.4GB. Filesize of restored database: 2.5GB.
The restore
> lasted for about 45 minutes.
> Pagesize 8KB, cache pages=8K (if that's relevant for a restore).
> Default config file (except for the alias section) which means
then
> fbserver will use only one cpu.
>
> While records were loaded TaskManager reported (in the "Process"
window)
> 11-16% of cpu for the fbserver process, and 4-9% for gbak. The
rest was
> mostly the Server Idle process.
> The four processor kernels (is this the correct name?) [K0-K3]
showed in
> the "Performance" window (approx!):
> fbserver: K0=60% K1=20% CPU 0
> gbak: K2=25% K3=25% CPU 1
> The split between kernels is different for fbserver (60+20) and
gbak (25+25).
> When indexes were built, the picture was somewhat different:
> fbserver: K0=60-100 K1=0
> gbak: K2=0 K3=0 (naturally)
> Building an index started out at about 60%, and then quickly rose
to 100%
> (and stayed there), but only on one kernel.
> The indexes aren't declared 'unique', but in practice they are (or
nearly so).
> I didn't watch the whole process so I may have missed something.
>
> Is there a "ping-pong" effect only when loading records? Why?
> Do you think there will be any benefit in turning the HT off?
>
>
> In my limited test, I haven't observed any delays when connecting
from
> remote clients. So far everything seems great, and I'm planning
to move
> production to this server during the weekend.
>
>
> --
> Aage J.

Aage,

I'm not sure if this will help with the issue you are seeing, but
you may want to test with it to see what affect it does have. We
had experienced issues with Windows 2003 Servers and Firebird 1.5.x
versions. The Firebird process would spike to 99% and our
application update would take very long to complete. I've been
aware of the HT issue and that it can be positive to turn this off;
however, in our user environment we often don't have control of how
the users configure their server.

I'd read a post some time ago that mentioned a similar issue and
recommended that the CpuAffinityMask be set to 3. We tried this and
it has worked very well for us. It eliminated the problem we were
seeing and to this point (about 6 months) we've not heard of any
problems with users running with this setting or produced any issues
in our test environment. We have this setting in the Firebird.conf
file as:

CpuAffinityMask = 3

I would be interested what affect this does have for your issue. We
had not seen experienced this as an issue with other operating
system environments other than Windows Server 2003 and this setting
has resolved our issues. I do understand that this setting is
contrary to what I've read the normal setting should be (1), but
again it did resolve our issue and we've not found any adverse
affects from using it.

Regards,
Steve