[DNA] Re: Flash renumbering

[Erik Nordmark <erik.nordmark@sun.com>]

Sathya Narayanan wrote:

>>Based on the comment above, we would also have to forbid the routers
>>from using learned prefixes as part of the DNA solution.
> 
> 
> ... why? The problems becomes worse with wrong state for upto 7 days
> only with reassignment. The above comment is only about flash
> renumbering which will be corrected within 90 minutes and is not that
> bad (A)

I guess we need to first agree what goals we are trying to satisfy. I 
see three possible levels:
1. do nothing special for flash renumbering and immediate reassignment 
(other than telling network admin to not immediately reassign prefixes)
2. do something so that a host can recover, but it might take a while 
(e.g., 90 minutes)
3. handle it without any delay

Which goal are you attempting for? (And we should ask the same of the 
whole WG.)

I was assuming you were trying for #3, since in some cases you'd recover 
in zero time. But in order to accomplish #3  we'd also have to remove 
any notion of using learned prefixes in the routers.

And if we want to reach #2 then I don't think we need to affect the 
efficiency of the normal DNA operations, but instead have some slower 
sanity check which can recover.

> By suggesting that this is suspicious, we are suggesting that the
> prefix-lifetime being greater than 90 minutes is not necessarily right -
> in effect putting an upper limit on the prefix-lifetime and overruling
> 2461. Right? Is it OK to do that?

Good question. I wish I had a firm answer. We have an inherent conflict 
between RFC 2461, which was designed we long default lifetimes just so 
that a host wouldn't throw away its addresses and prefixes just because 
a router died for a short time.

But our need to quickly and reliably detect movement, if we combine it 
with the renumbering and reassignment issue, is in conflict. We can't 
solve both.

Hence the best I think we can do is to use the absence of a 'link up' as 
a way to tell the host to stay in the RFC 2461/62 "stability is 
important" mode.

Note that your suggestion to use a RS to reverify each prefix has the 
same issue as triggering a sanity check after 90 minutes, since it would 
also remove a prefix from aggressively than 2461 if the advertising 
router is temporarily down.

> In support of my argument above marked (A), this solution doesn't
> address the flash renumbering problem - only flash renumbering with
> early reassignment. But, I do agree that this simpler defensive scheme 
> can handle the low-probability occurance of flash-renumbering with early
> reassignment.

Flash renumbering without any reassignment wouldn't cause any additional 
problems for DNA, would it? I think it is only an issue/feature of RFC 
2461/62.

   Erik