Re: SLAAC renum -- revised algorithm (Fwd: New Version Notification for draft-gont-6man-slaac-renum-04.txt)

[Fernando Gont <[e-mail]>]

On 12/3/20 10:12, Philip Homburg wrote:
> In your letter dated Wed, 11 Mar 2020 10:38:43 -0300 you wrote:
>> The original algorithm (Algorithm #1) is now Section 4.5.1, while the
>> new algorithm (Algorithm #2) is in Section 4.5.2.
> 
> I think your new algorithm fails if a router sends two prefixes in two
> RAs and one RA gets lost.
> 
> Note RFC 4861 says (Section 6.2.1):
> MaxRtrAdvInterval
>                       The maximum time allowed between sending
>                       unsolicited multicast Router Advertisements from
>                       the interface, in seconds.  MUST be no less than 4
>                       seconds and no greater than 1800 seconds.
> 
> So if one RA gets lost, the next one may come only after 1800 seconds.

Yes, this is noted in the I-D where we say LTA_IVALID_DEFAULT could be 
1800 seconds if we wanted to be more conservative.

I see a number of possible ways forward:

1) Keep "as is", since in order for this to happen, this woudl require:
   + that implementations use a non-defualt value for MaxRtrAdvInterval, 
*and*

   + that they spread PIOs among multiple RA messages, *and*

   + one of such messages with PIOs get lost


2) Increase LTA_INVALID_DEFAULT to 1800, as noted in the I-D. I wouldn't 
mind, although this comes at the expense of reduced responsiveness

3) Keep the current default timer, but not that an implementation may 
want to send a unicast RA to the router when expiration is imminent 
(this is probably in RFC4861, already)

Thoughts?



> I forgot one issue, a router may send two RAs, one with a ULA, and one
> with a routable prefix. If the routable prefix accidentally gets deprecated,
> then the host loses connectivity.

This is a good point, and we will fix it (thanks!). Essentially, this 
behavior should be triggered by "RAs that contain one PIO that 
advertises a non-ULA prefix".

Or, even better, ULA addresses should be deprecated by PIOs that carry 
other ULA prefixes (but not the currently used one), where global 
addresses should be deprecated by "RAs that carry PIOs that advertise 
global prefixes, but not the currently used ones".

Thoughts?

Thanks!

Cheers,
-- 
Fernando Gont
SI6 Networks
e-mail: [e-mail]
PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492