[arin-ppml] CIDR v2.0

[bmanning at vacation.karoshi.com]

On Tue, Sep 16, 2008 at 12:37:14PM +0200, Iljitsch van Beijnum wrote:
> In the early 1990s IPv4 address space was running out. They fixed this  
> by changing routing protocols. Maybe we should try to do that again.
> 
> <history>
> 
> In the early days of the internet you could only use IPv4 addresses as  
> blocks of 16777216, 65536 or 256 addresses. 256 was too small for most  
> people so 65536 was a popular choice, but there are only some 16000 of  
> these class B blocks and it was looking like those would be exhausted  
> by the mid-1990s. So they started giving people a bunch of class C  
> blocks (256 addresses each) but now the routing tables started to  
> explode because a university that needed a single class B block before  
> now used something like 16 class C blocks, which had to appear in  
> routing tables individually. To fix this, routing protocols,  
> especially BGP, were changed to be able to work with address blocks of  
> arbitrary power of two sizes so address space and routing table slots  
> could be managed much more efficiently. (This is "classless  
> interdomain routing".)
> 
> </history>
> 
> Now that IPv4 address space is becoming scarce again, why not do the  
> same thing again? But now rather than arbitrary powers of 2, we modify  
> the protocols to work with arbitrary address ranges. I.e., 192.0.2.27  
> - 192.0.2.43.
> 
> To accommodate legacy routers that can't do lookups based on ranges we  
> can put in backward compatibility mechanisms from BGP5 to BGP4 similar  
> to the ones from BGP4 (with CIDR) to BGP3 (no CIDR).


	Thats a whole lot of work to support bitstrings in routing protocols.
	If its just to support IPv4 then I am not sure I see the long-term value
	in making such a change.  Now if we were going to start supporting 
	variable length addresses then I could see a compelling case for making
	such a change.  

	Mind you, I am a fan of variable length addressing.

--bill