keeping the routing table in check: step 1

Hopefully this thread will be quick and less convoluted. Rather than simply alluding to "one prefix per ASN", I'd like to detail an allocation scheme that works toward that.

Find the largest contiguous block. Split in half. Round to appropriate boundary. Assign. Space at the end of the block is reserved for expansion.

Ignoring special subnets for simplicity:

0/x, 128/x,
64/x, 192/x,
32/x, 96/x, 160/x, 224/x,
16/x, 48/x, 80/x, 112/x, 144/x, 176/x, 208/x

assuming all grow at equal rates. 96/x ends up growing quickly? No problem. Skip 112/x for the time being.

In short, allocate IP space logarithmically. Start with /1 alignment, proceed to /2, then /3, and so on. Keep the array as sparse as possible so an assignment can be extended without hitting, say, a stride 4 boundary.

Perhaps RIRs should look at filesystems for some hints. Imagine a filesystem that's 30% full yet has as much fragmentation as IPv4 space. Something is wrong.

Eddy

So while this may look nice and sound good and all that, I hate to ask the
obvious question... Who is going to obtain the authority and/or balls to
take everyone's currently allocated IP addresses away and start over?

Perhaps I missed something in an earlier discussion, but this to me sounds
like a very nice, very academic "Hmmmmm" thought process.

Unfortunately reformatting the Internet is a little more painful that
reformatting your hard drive when it gets out of whack.

I guess my question is, what's the point of asking this question now?

Scott

Date: Wed, 15 Feb 2006 23:05:20 -0500
From: Scott Morris

I guess my question is, what's the point of asking this question now?

IPv6 is still fairly green-field. Future IPv4 allocations will be made,
too.

Eddy

Find the largest contiguous block. Split in half. Round to appropriate

--Michael Dillon

This presentation was made at the ARIN meeting in Orlando earlier in the
year. It is also available at

http://www.arin.net/meetings/minutes/ARIN_XV/PDF/mon/mei-wang.pdf

Ray