On 18-okt-2007, at 3:46, BELLEVILLE Ray wrote:
What ever happened to pushing on the traditional class A owners to free up their address space?
The ARIN lawyers say it can't be done. I don't find that a compelling argument, but unless something happens very soon in this area, it will be too late anyway.
I can't help but think that the issue has always been mis management of the early assigned address blocks. Look at Nortel's block for instance... How many addresses are actually reachable directly from the internet? /22 subnets as a standard block with 100 addresses assigned.... They MAY have had an argument 8 years ago when they had 120K employees, but at 25K now, its a bit ridiculous. Hundreds of addresses per employee? How many other blocks are unallocated?
Haha, that's a good one, posting from an Alcatel-Lucent email address! I'm not sure what Nortel address space you're talking about, though. Their name is not in the list of class A holders. But replace "Nortel" with "HP" and your argument becomes twice as strong, they hold nets 15 and 16, for a total of more than 33 million addresses or almost a percent of the usable IPv4 address space. (The US government holds about 5%, though, and they don't seem to be willing to give any of it back.) However, people who think that better managing the existing IPv4 address space is a solution should acquaint themselves with the toothpaste doctrine. A tube of toothpaste is never really empty: if you squeeze really hard, something will come out. But at some point, all the squeezing becomes tiresome and it's easier to buy a new tube and throw away the old one. RFC 3194 observes that in the past, networks generally expanded their address space when around 87% of the address bits was used up. Upto a HD ratio ( = log(addresses used) / log(possible addresses)) of 80% there are no problems. But past that, the cost of managing the address space quickly increases. Not sure what the latest domain survey figures for the number of IP hosts are ( http://www.isc.org/ds/ but I'm working offline right now), but we should be well above a HD ratio of 90% for IPv4 right now. And that's with more than a billion IPv4 addresses unused, so the HD ratio for the allocated RIR space is a lot higher than that. As John Klensin says: for all intents and purposes we're already out of IPv4 addresses. They're already too hard to get for many purposes.
V6 is a nice idea, but it only deals with the symptoms, not the cause.
IPv6 is not perfect but it has a lot of nice features. With IPv4 I need to make sure that my OSPF routers have IPv4 addresses in the same subnet prefix or they won't talk to each other. OSPFv3 and other IPv6 routing protocols simply use link local addresses and this limitation is gone. No need to think about subnet sizes: one size really does fit all. Automatic VRRP-like operation when multiple IPv6 routers are present. But none of these features is worth modifying everything that touches an IPv4 address, from code to ASICs to configs to human brains. The ability to connect new users for years to come is, however, so hopefully we'll find a way to get from where we are today (IPv4) to where we need to be in the future (IPv6) although so far we haven't. What is the cause, by the way?