On Wed, 2005-08-10 at 16:31 +0000, Eduardo A. Salgado wrote:

> --- In firebird-support@yahoogroups.com, David Johnson <johnson_d@c...>
> wrote:
> > If centralization is required, you will want to use classic on as
> large
> > a *nix box as you can get.
>
> Here the customer has been providing the hardware according to their
> internal standards. Not sure about their using *nix.
>



> > Common application design principles are simple.
>
> We are following these OK, using Delphi. We have some 160+
> installations already running OK, with lots of happy users, but none
> approaching any limits. This one is a biggy and we have some other big
> ones in the horizon.
>
> Our customers use our programs to do proposals based on light number
> crunching. So the load on transactions is not that heavy. But the
> main database has grown to some 6 Gig. We have two other databases
> involved in the program. One is getting big with historical number
> crunching data (about 2 Gig) and another smaller one.
>
> I am trying to see what are the practical limits for concurrent users
> for SS and Classic if we have 8 processors and 3.5 Gig of RAM. Then
> based on that, we can see if we need to do something else to support
> this client.
>
> -Eduardo

2 to 3 gigs is practically empty, as long as your data model is good and
your regular maintenance (sweeps, transaction counts, etc) are up to
date. Even 6 gigs is not too bad.

Superserver does not work well on SMP systems (yet). Vulcan is supposed
to take care of that eventually.

As Helen said, if ($processorCount > 1) $platform = classic

There were some good articles on scaling firebird a couple of months
back. Check out messages in the archives about scalability - one
gentleman described for me the guidelines he uses for large scale
installations. Also check out Helen's Firebird book.

To improve scalability and performance, you can pool connections and
prepared statements behind Francois Piette's midware suite. see
www.overbyte.be