Hi Dimitry,

> As you described below, you actually need not replication, but

high

> availability of database.

Yes, and it's most current data, it's a realtime auctioning system ,
so every auction transaction (=Buyer purchase) is important, as well
as available supply. The main thing is that the auction has to be
able to continue, only a small hickup may be the result of a server
crash !

>
> Theoretically - yes. In practice nobody use it. They just set up
> replication every N seconds where N is reasonably small.
>

Ok, in my cas N would be 1 or 2 seconds ;-)

> >In my case the target database is a sleeping database, up until the
> >master server dies, then the target becomes the master database.
>
> From here a bit more details, please! How? Your application can
> recognize server failure and switch to another one? Or something

else

> like DNS round-robin?

Well both servers are dual boot (master and slave) so when the master
crashes, they restart the slave as master, which gives it the
master's ip address. I did suggest for the software to switch
automatically, but they (our customer) prefered it to be a manual
switch.

>
> >I built the sync-server app with Kylix 3 using UIB components.
> >Now that it's finished I wonder wether I should throw it away and

go

> >for the IBReplicator tool anyway... Especially the IBevent driven
> >feature makes me doubt (so no polling needed !)
>
> If I were you, I'd think about three-tier architecture...

That would make the transition master/slave lighter, true, but I'd
have to build a trigger in it so client apps would be aware of the
switch, and customer wanted the switch to be manual, not
automatically, with full restart.


Dirk Naudts.