Subject | Re: scalability how to? |
---|---|
Author | Richard Salt |
Post date | 2004-09-14T17:04:49Z |
--- In firebird-support@yahoogroups.com, "Photios Loupis (4T)"
<loupisp@a...> wrote:
The default LockMemSize under Linux (96K) is lower than on Windows
(256K). I raised ours to 256K but still got out "Lock manager out of
room" errors. So I went to the extreme(!) and raised it to 1024K.
Using the fb_lock_print util, I can now see that the actual usage is
approx.350K, so 1024K is overkill, but still peanuts on current
hardware.
We also got an EventMemSize error that caused the server to crash.
The default size is 64K. I set ours to 1024K and this seems to have
solved it. I have no idea of finding out the actual usage.
HTH
Richard Salt
UK
<loupisp@a...> wrote:
> --- In firebird-support@yahoogroups.com, "Richard Salt"out
> <Richard.Salt@q...> wrote:
> > Does anyone have any suggestions on how to take the guesswork
> of scaling/load balancing the number of databases/connections perLinux
> Firebird server?
> >
> > We currently have 60 fdb's with a potential of 150 connections
> being serviced by one Firebird 1.5.1 (nptl) server running on a
> (Mandrake 10.0) box with 2GB ram. Our delphi apps use ibobjects:the
> transaction objects are all set to tiCommited and we avoid longI
> transactions. It has been necessary to increase LockMemSize &
> EventMemSize. It typically works well, but we do occasionally have
> unexplained crashes.
> >
> > The number of databases & connections is growing each month and
> am mindful of if/when to balance the load onto another physicalmemory
> server.
> >
> > I have written scripts that log the fbserver's processor and
> useage, but this seems fairly meaningless without knowing theuseage
> patterns of the users. Is there any Firebird monitoring softwarethat
> might help?the
> >
> > Best Regards
> > Richard Salt
> > UK
>
> Hi Richard
>
> We too are experiencing a similar problem (see earlier posting in
> group). We have a large number of concurrent connections andthese
> periodically 70% of the connections get EXTREMELY active. This
> causes our processor utilisation to go through the roof.
>
> I am interested in the settings you have used for LockMemSize &
> EventMemSize, please could you elaborate on your findings with
> settings?Hi
>
> Thanx in advance
The default LockMemSize under Linux (96K) is lower than on Windows
(256K). I raised ours to 256K but still got out "Lock manager out of
room" errors. So I went to the extreme(!) and raised it to 1024K.
Using the fb_lock_print util, I can now see that the actual usage is
approx.350K, so 1024K is overkill, but still peanuts on current
hardware.
We also got an EventMemSize error that caused the server to crash.
The default size is 64K. I set ours to 1024K and this seems to have
solved it. I have no idea of finding out the actual usage.
HTH
Richard Salt
UK