Dave Cridland writes: > I do understand your argument, and you're correct in all its > assertions, but not the conclusion. I suspect that's the case for > everyone at this point. Not as long as I still see people claiming that 128 bits will provided 2^128 addresses _and_ that it can still be divided into multiple bit fields. > You state, loosely, that 128 bits will not realistically yield > 2**128 addresses, which is entirely true. Yes. > It's been pointed out that IPv6 wasn't designed for that, instead, > it was designed to yield 2**64 subnets, and even so, it's > acknowledged that a considerable amount of that space will be > wasted. People have agreed with this, but pointed out that the > "subnet" level can be moved down, since we're only using an eighth > of the available address space. I don't think many people appreciate just how quickly such policies can exhaust an address space--mainly because they keep emphasizing that 2^n addresses are available in n bits, apparently oblivious to the multiple factors that will waste most of the addresses. > Your conclusion, however, is that we should be switching to a > zero-wastage allocation mechanism preferably based on variable > bitlength addresses. That is one option. Another is to stop trying to plan the entire future of IP addressing today. As I've said, just adding one more bit to 32-bit addresses would hold the Internet together for years to come. Immediately blowing 2^125 addresses is absurd. > In response to this, several people have commented that this > is unworkable using both current hardware and any hardware > predicted to be available within the next few years. I don't > know about that, but I'm prepared to accept that opinion. I'll accept the opinion, but as long as it remains opinion, I can continue to assert the contrary. I don't see any insurmountable obstacle that would prevent this type of implementation. Indeed, I should think it would greatly simplify routing. > There's an additional unanswered question your argument has, which is > whether the - very real - issues you're pointing out with prefix > based allocations will cause actual operational problems within a > timeframe short enough for anyone to worry over for a few decades, > and - a related issue - would these problems hit sufficiently quickly > that a replacement for IPv6 couldn't be developed in time? In this respect I'm going by past history. As I've said, engineers routinely underestimate capacity and overestimate their own ability to foresee the future, often with expensive and defect-ridden results. The Internet gets bigger all the time, and the cost of these mistakes will be astronomically high in the future--more than high enough to justify changing this mindset. I'm just trying to limit the damage by suggesting changes as early as possible. Has anyone else noticed that the simplest standards tend to last the longest, and that complex, committee-designed standards are often obsolete even before the 6000-page specifications are printed and bound? I see that SMTP is still around, but I don't see too many people using X.400. I see people writing code in C, but not in Ada. _______________________________________________ Ietf@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ietf