[ILUG] Class C Addresses

[John P. Looney]

On Wed, Nov 07, 2001 at 06:06:59PM -0000, Treanor, Ciaran mentioned:
> >  I'm certainly not joking. It's how DNS works, one of the 
> > most reliable time critical systems around. It's not going to work for 
> > everything, but people should be aware of it...
> Can you name some other things that use the approach you mention?

 SMTP comes to mind. NNTP can also be configured to such news from
multiple locations.

 Ages back, I was working on a HA webmail system. We had three webservers,
running PHP behind a load balancer, all lovely.

 However IMP, the webmail client talked to a MySQL database for temporary
storage of session info. It talked to an IMAP server, and an SMTP server
for getting/sending mail. The IMAP server authenticated against an oracle
server.

 However, if the MySQL server went down, we were screwed. We had two IMAP
servers. So, I hacked IMP to do the round-robin thing (when behind a load
balancer, you can't access load balanced services, due to NAT
shortcomings) talking to IMAP (SMTP works like that, as mentioned
earlier). Same had to be done to the MySQL thing, once I setup a backup
server. 

 NAT just doesn't work with Fileservers or Database servers. Anything
where locking is a problem. Round Robin DNS isn't the answer to most
stuff. DNS propagation kills you, on the big bad internet. But pulling out
the big guns straight away is not a good idea.

> > It's just I've seen the pain caused trying to load balance stuff like
> > MySQL, because application programmers don't want to be bothered
> > considering a scenario where they can read from many servers, and only
> > write to one...
> Sorry mate, but if you having problems with MySQL then it's not the
> fault of the programmers that wrote MySQL - that's for sure.

 Problems like the hard disk dies in the MySQL box ? Because that's what
I'm talking about.

> You've got to ask yourself why you are using MySQL - legacy hangover?

 Now, you have me trying to work out are you a troll or an arsehole.

> You've got to choose the right tools for the job - maybe the tools
> that were originally chosen either:
> - Weren't the right ones to choose - ever.
> - Were the right ones at the time but the plans never accounted for
> expansion/

 I see plenty of that. I was talking to a customer a while back that
wanted to load balance his WinNT/sqlserver/COM application. It just wasn't
possible - he made no attempt to address the concept of a database being
down, or lock the database when in use by the application so what could we
do? Tell him to rewrite it, or drop it ? No layer 7 magic is going to save
you there.

> That said, I am one of those people who developed some CORBA-load
> balancing stuff using the CosTrading a couple of years ago. One of the
> reasons I chose that approach was that IIOP is one of those protocols
> that firewalls and NAT doesn't like. It was interesting for sure, but
> I'm glad to say that there are plenty of app servers on the market now
> that happily take care of load balancing (and probably make a better job
> of it than I did!)

 But it's so much more pain to develop a layer7 load balancer than to
develop a decent protocol in the first place.

Kate

-- 
_______________________________________
John Looney             Chief Scientist
a n t e f a c t o     t: +353 1 8586004
www.antefacto.com     f: +353 1 8586014