OSEC

Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com
 
Re: DNS Load Balancing

From: Dean Strik (deanstack.nl)
Date: Mon Feb 02 2004 - 03:01:15 CST


Tobias Reckhard wrote:
> Dean Strik wrote:
>
> >Tobias Reckhard wrote:
> >
> >>Neither scheme provides deterministic load distribution. Both depend
> >>entirely on how clients handle multiple answers to DNS queries. You have
> >>no influence on that at all.
> >
> >Clients randomize.
>
> You hope. You don't know and you can't be sure.

Oh yes I do. And who cares if <10% of the MTA clients wouldn't. If 90% of
the mailers do, you won't see much of a difference.

> >RFC2821 even *requires* randomization by the client.
> >It just works.
>
> I know it works rather well in practise. But still I stand to my point:
> "neither scheme provides deterministic load distribution."

Load distribution doesn't have to be deterministic of course. Just 'near
50/50' is a nice goal, and esp. if the number of boxes increase (e.g.
for four boxes, it's getting near 25% per box continuously).

And I doubt the OP was even after strict 50/50. He just wants two boxes
that take care of the mail instead of one, distributing the load.

> Do you believe all SMTP clients conform to RFC 2821?

Of course not. But this is not about conformance (and strictly speaking,
they only have to conform to the current standard, RFC 821) - it's about
randomization. And I believe the vast majority of the mailers on the
internet do randomize.

> >Of course, distribution could be anything between 40%-60% one day and
> >60%-40% the other day, but *that* hardly makes a difference.
>
> That depends entirely on one's expectations.

If 40%-60% is unacceptable (I find that hard to believe), then yes,
buying expensive load-balancing equipment to make it nearer 50-50 is an
option. So is building some simple load balancing box itself. But
personally, if I had to spend money on new hardware for load balancing,
I'd just buy a third machine - distributing load over 3 boxes..

--
Dean C. Strik Eindhoven University of Technology
deanstack.nl | deanipnet6.org | http://www.ipnet6.org/
"This isn't right. This isn't even wrong." -- Wolfgang Pauli