Hi Michael,

Thanks again for the information and the links, especially the one about
Firebird locking which is really useful. Now I understand a bit more about how
the semaphores are used.

Tom

On Thu, 30 Jul 2009 20:22:19 Michael Weissenbacher wrote:
> Hi Tom!
> Well, i've used the following information sources:
> http://ibdeveloper.com/issues/issue-2-oct-17-2005/locking-firebird-and-the-
>lock-table/
> http://www.idevelopment.info/data/Oracle/DBA_tips/Linux/LINUX_8.shtml
> http://forums.devshed.com/firebird-sql-development-61/tuning-firebird-datab
>ase-for-large-db-on-dualxeon-with-2gbt-202214.html
>
> Frankly, i cannot remember how exatly i arrived at those values. I
> conducted some load tests and settled for them after i saw my locking
> problems go away. I think the default values for the lock manager are
> pretty low for today's powerful hardware.
>
> hth,
> Michael
>
>
> -------- Original-Nachricht --------
> Betreff: Re: [firebird-support] LockSemCount
> Von: Tom Munro Glass <tom@...>
> An: firebird-support@yahoogroups.com
> Datum: 29.07.2009 23:39
>
> > Hi Michael,
> >
> > Thanks for the information. Clearly LockSemCount = 500 is significantly
> > higher than the default value of 32 - how did you arrive at this value?
> > I'd really like to know what factors should be considered when changing
> > this value - can any of the Firebird developers explain the significance
> > of this?
> >
> > Tom
> >
> > On Wed, 29 Jul 2009 20:59:16 Michael Weissenbacher wrote:
> >> Hi Tom!
> >> I have been running a heavily loaded Firebird Classic 2.0 on a (Gentoo)
> >> Linux Server for about 3 years without a hitch with the following
> >> settings: firebird.conf:
> >> LockSemCount = 500
> >>
> >> /etc/sysctl.conf:
> >> kernel.shmmax = 1073741824
> >> kernel.sem = 1024 32000 256 1024
> >>
> >> /etc/xinetd.d/firebird:
> >> service gds_db
> >> {
> >> disable = no
> >> flags = NODELAY KEEPALIVE IPV4
> >> socket_type = stream
> >> wait = no
> >> user = firebird
> >> server = /opt/firebird/bin/fb_inet_server
> >> cps = 0 0
> >> instances = 1000
> >> per_source = 300
> >> log_type = FILE /var/log/firebird/xinetd.firebird.log
> >> }
> >>
> >>
> >> I had some problems in the start with the Lock Manager, before
> >> increasing the default values. Another important thing are the
> >> "instances" and "cps" settings of xinetd, which may prevent new
> >> connections under certain circumstances.
> >>
> >> hth,
> >> Michael
> >>
> >> -------- Original-Nachricht --------
> >> Betreff: [firebird-support] LockSemCount
> >> Von: Tom Munro Glass <tom@...>
> >> An: firebird-support@yahoogroups.com
> >> Datum: 28.07.2009 06:35
> >>
> >>> Running FB classic 2.1.2.18118 on Mandriva 2009 with an unmodified
> >>> firebird.conf.
> >>>
> >>> I have had a corrupted database, and prior to this firebird.log
> >>> contains the "Fatal lock manager error: semaphores are exhausted"
> >>> message. The system would only have had one or two users connected at
> >>> the time, so can someone please tell me what operations cause a
> >>> semaphore to be used? For example is the number of semaphores used
> >>> affected by the number of active transactions, and does running nbackup
> >>> use any semaphores?
> >>>
> >>> Also, does this message mean that FB has used all of semaphores as
> >>> determined by LockSemCount (default 32 I assume) or does it mean that
> >>> Linux has run out of semaphores?
> >>>
> >>> I'm trying to determine if I need to increase the value of
> >>> LockSemCount, and if so, what should I increase it to.
> >>>
> >>> Tom Munro Glass
> >