DNS reliability (Re: [oclug] RedHat Enterprise and HP servers)

[Andrew J. Hutton]

On Tuesday 22 June 2004 13:38, Rod Giffin wrote:
> Matthew Wilcox said:
> > On Tue, Jun 22, 2004 at 01:21:53PM -0400, Charlie Brady wrote:
> >> Not even that if it's a software failure which occurs. Most DNS
> >> outages  are software problems, aren't they?
> >
> > I don't think I've ever seen a DNS outage caused by problems with BIND.
> > Caused by badly configured software, yes.  Caused by router problems,
> > more frequently.  Caused by politics, most frequently.
>
> Certainly can't argue with that.
>
> One of the best ways to ensure DNS reliability is to put your DNS's on
> completely separate high availability networks, ideally a couple of
> thousand miles apart.  It makes no sense to operate highly available
> servers with single points of failure like a single network hub, or a
> single router.  You wouldn't achieve much reliability that way at all.

The 1000's of miles apart is actually not very useful in practise.  You're 
better of bringing in an extra DSL line from another carrier who doesn't use 
the same backbone service.  Considering most inter-Canada traffic already 
goes through New York you can have two servers '1000's of miles apart' even 
within the same office.  Yes, having them in different areas of the city is a 
good option due to power stability.

What I'd probably do is pay for some employees DSL and drop a server there for 
the 2ndary, and then negotiate 3rd and 4th nameservers with other 
organizations.