On Aug 18, 2010, at 4:57 AM, Simon Lyall wrote:
Sorry I'm a bit late with this question
That's OK, I'm a little late unsubscribing. (On a plane to China as I type.)
but I was wondering if you've had problems with
unusual DNS setups of any NZ providers?
In NZ specifically, I haven't noticed much other than what was discussed (and
hopefully fixed) as a result of my original post.
Things like forwarding DNS or http queries via
overseas servers so that it is difficult to use DNS to optimise delivery of http or (in
the case of http forwarding) geo-limiting break. Or people using the google DNS servers
Is there some "better practices" that ISPs can follow for the placement and
routing of DNS and http that will add content delivery?
Remember that Akamai - and most other content delivery networks, as well as lots of other
global load balancing companies / products - assume that the end user is "near"
the name server. We say we decide which node serves which end user ("mapping"),
but we really mean which node should serve which _name_server_. Akamai does this,
Limelight does this, L3 does this, etc. (Google's GCC is about the only one which
doesn't, but they do not serve everything out of the GCCs.)
If you use DNS forwarding to your upstream's NS, we will assume the user is in the
upstream. If you have a national network and all your NSes are in Auckland, we will
assume all the users are in Auckland. (Is this where I make a joke about "all the
users _are_ in Auckland"? :)
If you use something like Google (220.127.116.11), L3 (18.104.22.168), OpenDNS, etc., you are going to
be mapped .. uh, wherever that is. I am pretty certain none of those networks have an
anycast node in .nz, and some probably do not even have anything in .au either. How's
Tokyo or Los Angeles?
[BTW: I want to be clear I am talking about recursive, or caching, name service.
Authoritative NS has nothing to do with this, please do not get the two mixed up.]
There are easy ways to fix this that will actually improve performance & stability for
your end users, and (IMHO) make your life easier. Taking for instance a typical DSL
network, most people do not have IP packets until they get to the BRAS. For larger
networks with multiple BRASes in multiple cities (because if you have only one, the
problem kinda solves itself), just put a couple tiny boxes in each POP and anycast your NS
They make servers small enough that two actually fit in 1RU. They can be configured with
SSDs & laptop CPUs (so they are ridiculously reliable and ridiculously low power/heat)
for a couple hundred bux. Hell, for most networks, you could use a laptop to serve
recursive NS! This allows you to give _every_ user the same IP addresses for their NS,
yet they will all have ultra fast access to the NS since it is a few feet from the first
IP hop. Plus it will allow companies like Akamai to target individual POPs with our
mapping system. And you make the system ultra reliable because each POP backs all the
others up. Can't get much better than that.
Anyway, hope this gave you some insight into how content is being served on the Internet
these days. And perhaps some ideas on how to make your life easier.