Re: PIO Lifetimes (was: Re: draft-ietf-6man-slaac-renum: Heuristics to deprecate stale information)

[Fernando Gont <[e-mail]>]

Hi, Loba,

I had changed the Subject line because the discussion had deviated from 
the original thread: reviewing the algorithm that I posted.

Comments in-line....

On 18/8/22 16:47, Loba Olopade wrote:
> I think the algorithm as suggested works, under the current ND 
> specifications.

Are you referring to the algorithm that I posted? (the one in the 
message that started this thread).

If so, could you comment/describe specifically why you think "the 
algorithm doesn't work".?



> That said, I believe the current specifications should be expanded to 
> enhance both the host and router side heuristics. For instance, a new 
> option in RA indicating that the complete information has been sent by 
> the router would simplify the algorithm.

But in order for that to work, all routers would need to be updated. -- 
and that's a non-starter.

Aside: Why, specifically, do you think the algorithm is complex?
It is seems super simple: "Keep track of the info you have received from 
a router. If you receive partial information, refresh/validate the 
information with a unicasted RS".



> The host could be definite when 
> it has a complete set of information, rather than waiting for a random 
> period. 

The only "random period" we wait is simply to have the possible RSes 
spread over time -- i.e., if you have a network of 100s/1000s of nodes, 
you'd probably prefer that hosts spread their queries over a few 
seconds, rather than have all of then query the local router at the same 
time.



> On the router side (aware that Fernando mentioned that the draft doesn’t 
> address this side, but wanted to put my thoughts down), the question for 
> me is how they detect that hosts are using stale information and then 
> initiate some clean ups. The possibility of hosts including previously 
> received information in their RS could serve as a feedback mechanism, 
> allowing for this router side heuristics.

The idea is simple: we need to address the problem on the basis that the 
hosts should do the magic. Routers, and specifically CPEs, will probably 
take ages to be updated.

And the worst of all possible options are those that need implementation 
in both the host and the router -- since they require changes on both 
sides to be able to see any improvements.

Thanks,
-- 
Fernando Gont
SI6 Networks
e-mail: [e-mail]
PGP Fingerprint: F242 FF0E A804 AF81 EB10 2F07 7CA1 321D 663B B494