[LINK] Census Config [Was: Personal comments on the eCensus form]

[Saliya Wimalaratne]

On Wed, Aug 02, 2006 at 05:02:25PM +1000, Roger Clarke wrote:

> Currently, www.census.abs.gov.au resolves to 202.81.31.30, 62, 94 and 
> 126 (round-robin on one sub-net?!), but times out on pings.  The DNS 
> entry is below.
> 
> Whereas census.abs.gov.au says 'Bad name, address or endpoint 
> (OTBadBameErr)'.  (I'm using Interarchy 8.1.1 at 16:50-55 UT+10).

Hi Roger,

> Am I alone in being astonished that:
> (a)  there's no duplicate name-server (on a separate sub-net, of course)

Well, it's not the best-practice, but... if even a single nameserver is
working and reachable, then in practical terms you probably wouldn't 
notice it.

I'd say they've spent their redundancy budget where it's needed - 
on the actual www hosts themselves (there are currently 4). 

> (b)  there's no name-server entry for the domain-name without 'www.'

Yes. 

Blame MS for this one - if you put an A record in for the 'domain', 
some old Exchange servers will IGNORE your MX records and try connecting
directly to the SMTP port on your A host when sending mail to user at domain.

Meaning that if you _do_ put in an A record for a 'domain' (it's not
really that; but the term will do :) you are virtually guaranteed to
have random mail problems at some point down the track when mail to
user at domain passes through an old Exchange server (it happens!). 

> (c)  all of the servers appear to be on the same sub-net, and hence
>      the service is seriously fragile

They're all on the same /24; but that doesn't mean that the service
is fragile. The theory here is that they are all traffic to the
subnet is going to be delivered over the same physical infrastructure,
and hence will be 'only' as reliable as the least-reliable link. But:

* they may be on separate smaller nets routed via different links (most
upstreams are routing down to /29 these days :P)
* the link(s) they are on may be seriously over-engineered (maybe 
a gig-E cross-connect into Telstra's core with multiple redundancy?)

We don't know whether either of the above possibilities are true; 
but a single /24 != fragile, is all I'm trying to say.

Regards,

Saliya