Re: [RRG] End user network size [ [Q] draft-farinacci-lisp: IPv4 address depletion]

[Iljitsch van Beijnum <iljitsch@muada.com>]

On 27-sep-2007, at 14:05, Robin Whittle wrote:

> In the absence of a new architecture (LISP/eFIT-APT/Ivip/TRRP), year
> by year, X number (X will grow with each year) of end-user networks
> will want and/or need stable IPv4 addresses, for the purposes of
> multihoming and/or so they can choose ISPs without renumbering.

(which can be described as "serial multihoming")

> The question is:

>   What proportion Z of these end-user networks could be run (in a
>   way which reasonably meets the end-user's needs) with
>   significantly less than 256 IP addresses?  (Say 128, 64, 32, 16,
>   8, 4, 2 or even 1.)

The only thing that absolutely, positively eats up addresses is  
servers that are reachable over SSL. For pretty much everything else  
you can have multiple logical servers share an IP address. With load  
balancers you can have multiple physical servers share an IP address.  
In my experience, there are significant numbers of organizations that  
have to be creative to get that coveted /24 block from RIPE (other  
RIRs have slightly different rules). That's today. Bandwidth is  
getting cheaper, address conservation is getting better, dependence  
on the internet keeps increasing.

> I think the IPv4 address crunch will just lead to more and more
> inventive ways of using address space efficiently.  Without a new
> architecture, this will surely bloat the IPv4 global routing table
> to double and beyond its current size in a few years.

> No-one will be happy with an IPv6-only address until 99% or more of
> other Internet users have an IPv6 address too.  There seems to be
> little reason to get an IPv6 address at present, for most people,
> and I can't see why this would change.

What IPv6 brings to the table is a simpler network. We tend to forget  
this because we still assume that IPv4 will always be present, but if  
you compare an IPv6-only network to an IPv4 network, the IPv6 network  
is much easier to build and maintain. As IPv4 addresses become harder  
to get and IPv6 implementations become more mature, this difference  
will only become more pronounced.

Fortunately, the 99% issue is only applicable to a small subset of  
all applications. IPv6 deployment issues depend on the type of  
application: client/server or peer-to-peer, and one-to-few or one-to- 
many.

Client/server one-to-few would be email: you only talk to a very  
small number of mail servers. If those servers that you use are dual  
stack, you can run IPv6-only on your workstation and still read and  
write mail without trouble.

Client/server one-to-many is web: you talk to many different web  
servers. You can only be IPv6-only once ALL these servers are dual  
stack.

Peer-to-peer one-to-few is BitTorrent: you download pieces of files  
from different peers, but you don't really care from which peer. So  
as long as the tracker (coordinating server) and a reasonable subset  
of all peers is dual stack, you're in business.

Peer-to-peer one-to-many is VoIP: in theory, you can dial any number,  
so you can only run IPv6-only once your SIP server and all possible  
peers are dual stack.

Turns out you can easily move an application from the one-to-many to  
the one-to-few category by implementing proxies. So once your ISP (or  
IT department) has set up a dual stack HTTP/HTTPS proxy you can reach  
the web (in theory everything that uses TCP with the HTTPS proxy) and  
if your VoIP provider has an IPv6-capable gateway towards other  
networks, you can make calls over IPv6, too.

So in an environment where only a subset of all possible applications  
must be supported (i.e., an enterprise network) it's entirely  
possible to ditch IPv4 and proxy to the IPv4 world at the edge of the  
network.

> I think we are still going to be clamouring for IPv4 addresses in
> ten or fifteen years time.  Maybe we will be stuck with it forever.
>  I hope a new routing and addressing architecture can be implemented
> around 2012.

I find it strange how people can make predictions that far in the  
future. Less than 1% of all people that are now running IPv4 were  
doing so 15 years ago. If you can go from 0 to 4 in 15 years, why not  
from 4 to 6 in as many or fewer years?

--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg