IPv6 Pain Experiment

Fri Oct 4 04:38:18 UTC 2019

I'm going to reply in some detail to your points here because they are
very common arguments that have real answers. Those who have heard all 
this before are free to move on.  :)

You sound like someone who doesn't have experience with IPv6. I don't 
intend any criticism, I'm simply saying that once you start working with 
it, you learn it, and almost all of these concerns evaporate. Just like 
what happened when you learned IPv4.

On 10/3/19 8:20 AM, Naslund, Steve wrote:
> I don’t think the issue is the readability of the addresses (although
> hex does confuse some people), mainly it is the length and ability to
> deal with any string of numbers that long for a human,

Coming from IPv4 world, it's very common that when you're working with 
addresses directly most or even sometimes all, of the bits are 
different. By and large this isn't the case with IPv6. If you sized your 
RIR request properly, you'll have the same /32 (or shorter) prefix 
across your entire network. That covers the first two hextets of the 
network portion of the address (that is, half the network portion).

One of the great things about IPv6 is sparse allocation. The next hextet 
(third of four in the network section of the address) are the bits from 
/33 through /48. Since except for all but the largest sites you'll 
assign a /48 per site (65,536 /64s) that hextet will be the same across 
the entire site. For a really large office site, or a medium size data 
center, you might assign a /44 (1,048,576 /64s), so 3/4 of that hextet 
will be stable across that site. For a large data center you might 
assign a /40 (16,777,216 /64s) so half the hextet will be stable across 
that site.

So let's say a site is allocated 2001:0db8:1234::/48. A common practice 
is to use the top of the space for infrastructure, so 
2001:0db8:1234:0/49 (32,768 /64s) would be the same prefix used 
everywhere at that site, and every site would have the same admin 
prefix. Hopefully by now you can see the opportunities ...

> and I do
> realize that you can do static addressing in IPv6 (but I sure would
> not want to since the manual entry of the addresses is going to be
> error prone on both the host and into DNS). 

Copy and paste is your friend. Seriously. If you're typing things in 
you're doing it wrong. Obviously there are a very few situations where 
you don't have a choice, but seriously, copy and paste.  :)

> It is just way harder
> for a human to deal with hex v6 address than to easily memorize four
> decimal numbers in v4.  Most system admins and engineers can rattle
> off the IPv4 address of a lot of their systems like gateways, DNS
> servers, domain controllers, etc.  Can you imagine keeping those v6
> addresses in your head the same way? 

Why would you bother? That's what DNS is for. Also, see above, where you 
can establish patterns that make this easier for the very few situations 
where you actually have to use addresses, and also easier to spot check 
them when you do.

> Think about reading them over
> the phone, typing them into a support case, typing a configuration
> sheet to be entered by some remote hands etc.  I am not saying it is
> insurmountable, it is just something people need to get used to.  To
> me, that is the biggest reason not to do more manual assignments than
> we need to. I do understand why they need to be the way they are but
> I can't see anyone thinking IPv6 addresses are easier to read and
> handle.

No one said easier. Just different. And not hard to learn as you work 
with them over time.

> It is not a misconception that most server guys are used to static
> addressing and not auto-assignment. 

See the other messages where we talked about how static addressing for 
services is not a problem for IPv6.

> I also takes some time to get
> people to stop hardcoding static addressing into system
> configurations.  There are lots of applications that have dialog
> boxes asking for addresses instead of names.  That needs to stop in
> an auto-configured or DHCP environment

Yes, things are different ... one could even argue better, but yes, 
different.

> (yeah, I know all about DHCP reservations but I hate them).

They aren't that bad, but made much easier with a good IPAM. :)

> Your comment regarding small networks not needing IPv6 is exactly my
> point.  The original post was talking about MANDATING the use of IPv6
> to the exclusion of (or taxation of) IPv4.  My point is that there is
> not really a need to do so in a lot of use cases.
> 
> Is there a huge advantage to stop using RFC1918
> addressing within our network?  No, not really.

You've obviously never had to renumber two large internal networks after 
a merger.

> Would I build a
> completely new enterprise on IPv4...probably not.   Would I recommend
> that every global enterprise eradicate the use of IPv4 in the next
> couple of years....no.

No serious person is doing that.

hope this helps,

Doug