[ic] Cluster and/or load balancing question

[Dan Browning]

At 11:41 AM 7/13/2002 -0600, you wrote:
>Quoting Dan Browning <dbml@kavod.com>:
>
> > At 08:46 AM 7/11/2002 -0400, you wrote:
> > >On Thu, Jul 11, 2002 at 07:20:46AM -0400, Bill Carr wrote:
> > > > On Thu, 2002-07-11 at 04:19, Joachim Leidinger wrote:
> > > > > Hi List,
> > > > >
> > > > >
> > > > > I've searched for any answer in the mail archivs to get a right
> > > > > using/installation/configuration of IC with 2 servers as a cluster
> > and
> > > > > later with a load balancing solution. But I'm unsure about the right
> > way
> > > > > to the solution.
> > > > >
> > >
> > >...
> > > >
> > > > I use a LVS-DR setup like this:
> > > >
> > > >           Internet
> > > >              |
> > > >         ----------
> > > >         | LVS-DR |
> > > >         ----------
> > > >              |
> > > >       ----------------
> > > >       |              |
> > > > ------------    ------------
> > > > | www/ic 1 |    | www/ic 2 |
> > > > ------------    ------------
> > > >       |              |
> > > >       ----------------
> > > >              |
> > > >         ---------------------
> > > >         | MySQL/NFS Server  |
> > > >         ---------------------
> > > >
> > > > I would start with this and once you get something like this working
> > > > consider load balancing or replicating your database. You will have a
> > > > single point of failure in your database server so through your best
> > and
> > > > most sturdy hardware at it.
> > > >
> > > > Please go for it and become an expert. I think that there is only
> > myself
> > > > and Dan Browning that run any kind of rig like this now.
> > > >
> > >
> > >I'm curious as to what level of activity justifies load balancing?
> > >Or are you doing it just for redundancy?
> >
> > For us, the nature of the application required a lot of processing power
> > per user, but it would have been done anyway for the redundancy (or reduced
> >
> > failure percentages anyway).
> >
> > >I'd think NFS would cost you more performance in the above setup than it
> > >would be worth vs more beef for handling session files.
> >
> > We use MySQL for sessions, and it seems really fast.
>
>
>Dan, have you thought of posting how to use MySQL for sessions instead of the
>traditional method? That alone would help many people with clustering issues.
>Getting the sessions off the hard drive of one machine and into an SQL 
>server is
>most peoples breaking point.
>
>I had spoken with Mike about this a long time ago...
>
>He mentioned his method was to have an Apache server up front. It would 
>take the
>http requests and hand off new requests to a session server. That server would
>issue the session number. Then the client was handed back and forth 
>between all
>the Interchange servers, since they all shared a common session database. You
>just needed to go to the main session server first. Then, if I recall 
>correctly,
>he had a MySQL server as well. Depending on the needs.
>
>So to sum it up. One Apache server, one Interchange initial session server,
>several Interchange servers, and one MySQL server.
>
>With that model, if you stepped it up to say 2 MySQL servers. One for sessions
>and one for the Interchange tables, you could have a nice little boost as 
>well.
>Again the only thing would be to publish how to use MySQL for the sessions as
>opposed to the default methods.

Consider it done (once I get back from Kentucky, that is).
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| Dan Browning, Kavod Technologies <db@kavod.com>
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~