IPv6 vs IPv4 (Re: Sprint NOC? Are you awake now?)

Iljitsch van Beijnum iljitsch at muada.com
Wed Sep 3 08:43:02 UTC 2003


On dinsdag, sep 2, 2003, at 23:18 Europe/Amsterdam, Nenad Pudar wrote:

> Again my point is that your site (or any other that use the same dns 
> for ipv4 and 6) may be "blackholed" by ipv6 (it is not the question 
> primary about the quality ipv6 connction it is the fact that your ipv4 
> connection which may be excelant is blackholed with your ipv6 
> connection which may not be good and to me the most obvious solution 
> is not to use the same dns name for both)

First of all, why are you repeating everything the previous posters 
said? This is a waste of bandwidth. Not only on the network, but also 
where it really matters: in the synapses.

The real problem is that your software assumes that if there are 
several addresses in the DNS, it can just pick one and assume that 
address works. That has never been a good idea, but in IPv4 you can get 
away with it. In IPv6, you can't. IPv6 hosts are required to support 
more than a single address per interface, and when people actually use 
this then it's only a matter of time before address #1 becomes 
unreachable while address #2 is still reachable. So this means you have 
to try them all.

The new name to address mechanisms for IPv6 are such that you can ask 
for IPv4 addresses, IPv6 addresses or both for a certain FQDN. If you 
choose both, you'll usually get an IPv6 address first.

I don't see how it would be reasonable to have separate FQDNs for all 
these addresses and have the user try them all rather than simply have 
the application walk through the list of addresses and try them all 
until it gets a live one.

(And yes, I've suffered from decreased performance because of 
non-optimal or even nonexisting IPv6 connectivity, but that's the price 
of being an early adapter.)

Now if your argument is that it's not a good idea to depend on 
applications handling this they way they should _today_ that is 
something I'm willing to discuss, although I don't necessarily agree.

BTW, my IPv6 connectivity for www.bgpexpert.com is in some ways better 
than IPv4 as there is an extra path available over IPv6 that isn't 
available over IPv4.




More information about the NANOG mailing list