Re: PROs/CONs of SIP/PIP/TUBA

[Ross Callon <[e-mail]>]

	> >1)  Since hierarchical addresses tend to be deployed in a "sparse" manner, 
	> >it unclear whether a 64 bit address space will adequately scale to meet 
	> >every conceivable future addressing scenario. 
	> 
	> It is pretty clear that any fixed length will never be enough.

	Yes, in the sense that it is possible to conceive of an addressing scenario
	that uses an unbounded number of bits.  Once we have infinitely fast routers
	with infinite memory, joined by links of infinite bandwidth, we could consider
	implementing an internet protocol with unbounded address length.  Until then,
	we'll have to make an engineering choice of maximum address length.  SIP's
	64-bit address space is 10^7 times larger than the international telephone
	numbering space, and if managed sensibly (i.e., not just allowing everyone
	to cobble up their own favorite octet string and expect the world to be able
	to route on it) it will scale to handle many thousands of computers in every
	room and vehicle in this solar system.

I think that this is such an important point that I will break my own rule
(against public "my protocol is better" discussions) and reply (with apology 
to folks on all three lists): 

It seems from recent discussions that when folks talk about "scaling to a
huge network" there are two different things that people mean by this:

 - Some folks seem to mean being able to find some way to set up addresses 
   and some way to set up the topology so that it is possible to run routing 
   protocols and deliver packets to everyone.

 - Some people mean being able to adminster, configure, and run the network
   of this size, including routing, but also including host configuration,
   address administration, subnet address determination (mapping IP-level
   addresses to subnet addresses) and whatever other address-related problems
   that you will face in whatever type of network that you might happen to 
   find in a huge Internet (for whatever topology the world actually ends up
   setting up).

The first problem therefore is a subset of the latter.

I sort of think that the first problem can probably be solved with 64 bit 
addresses. My concern is the latter problem.

For example:

 - How do you autoconfigure host addresses?

 - If this requires active address assignment at the time that the host is
   hooked up (rather than a more passive automatic address discovery based
   on already assigned parts of the address), then how do you assign these 
   addresses, and how do you coordinate redundant address servers (assuming
   that you will need redundant servers for reliability)?

 - How do you assure that when a system is re-attached to the same network
   after some time away (for example, when I take my laptop with me for a
   two week vacation and two more weeks of business trips, and then return
   to the office), that when I return I will get the same address assigned
   to me again? (or if I don't, then how do you deal with filters?) 

 - If you have many public service providers, each with 10,000,000 
   customers, where 9,999,900 of the customers are small stub networks 
   (homes and/or small businesses) attached in only one location, then 
   how do you map from the IP-level address to the addresses used in the 
   service provider network? Will this work with the networks that people 
   are already talking about building (and over the phone network, given
   that the cost of networking homes may mean that POTS is the networking 
   infrastructure for much of the world for many years)?

I can think of very simple solutions to these sorts of problems if I have
large addresses (and can also think of straightforward ways to compress
headers and/or forward large headers at high speed). I would really like 
to see complete proposals for each of these problems with small addresses
(and I would expect that SIP advocates would want to see solutions for
header compression and high speed forwarding for the proposals with larger
headers and addresses). I understand that the SIP folks have started working 
on the first of these items (host autoconfiguration), but I think that it is 
worth waiting for solutions to all of these problems before we try to decide 
which solution is the best one for the next 20 years.  

Ross