Re: [v6ops] new draft: draft-yc-v6ops-solicited-ra-unicast

[Alexandru Petrescu <[e-mail]>]

Le 22/07/2015 09:20, Andrew 👽  Yourtchenko a écrit :
>> On 21 Jul 2015, at 16:08, Alexandru Petrescu <[e-mail]> wrote:
>>
>>
>>
>> Le 17/07/2015 18:57, Andrew Yourtchenko a écrit :
>>>
>>> On 17 Jul 2015, at 09:34, Fred Baker (fred) <[e-mail]> wrote:
>>>
>>>>>
>>>>> So the next logical thing to do would be to have the router default to
>>>>> unicast Router Advertisements, measure the rate of received Router
>>>>> Solicitations, and switch to multicast RA mode past a certain
>>>>> threshold to cover this sort of situation. Once the number of RSes
>>>>> falls, it switches back to unicast RA mode.
>>>>>
>>>>> That would get rid of the configuration knob proposed in this ID, and
>>>>> is behaviour that I think could be universal for all link types,
>>>>> rather than just for the case of wireless ones with mobile devices.
>>>>
>>>> If it were me implementing it, I think I would go about this in a little different way, hopefully simpler. I would want to send at most one (e.g., either zero or one) RA per some interval (a second?). In the normal case, that is sent unicast. However, having sent a unicast RA at time t, if I now receive another RS before t+1, I send the next one (at time t+1) as a multicast.
>>>
>>> values of T less than 3 seconds would make performance worse than today.
>>>
>>> There are many things to optimize for:
>>>
>>> - wireless airtime
>>> - bandwidth used by RAs
>>> - CPU usage sending unicast RAs by router
>>> - energy consumption on devices for more than one medium
>>
>> and:
>>
>> - fast handovers: on some links the more frequent the multicast RAs the
>> faster the handovers are, i.e. less packet loss for end nodes, less
>> glitches in the video conference stream.
>>
>
> If the network sends the periodic RAs frequently enough to avoid the
> glitches of the video stream, this is explicitly *NOT* the network we
> are concerned about in this draft.

But mentioning 'mobile' for the nodes, makes think so.  Because fast 
address auto-configuration is a characteristic of mobile nodes.  One 
wants to spend as little time as possible waiting for RAs.  For that 
reason the multicast RAs period was reduced to milliseconds.

> Also, we wanted to be very focused what we talk about in this
> document, to be able to ship it quicker.

I agree.

> We can say "does not apply to 802.11p, applies to 802.11(a|b|g|n|ac).
> Applicability to other networks is left at readers' discretion".

I agree.  It should also say it does not apply to LTE.  BEcause in LTE 
an RA is sent on only one ptp link, there is no risk of waking up somebody.

> Or something along these lines of that - feel free to propose the text
> if the above is not what you had in mind.

On a scale MUST NOT, MAY, SHOULD - SHOULD is the right term.

"SHOULD [...] unless the network is of type LTE or 802.11p".

Or:

"SHOULD [...] unless there is an interest in as numerous RAs as possible 
to help in fast address auto-configuration to large number of mobiles".

or

"SHOULD [...] unless the Hosts are allowed to explicitely express 
interest in the multicast group dst of RA".

or

"SHOULD [...] unless RFC-MLD is updated to require Hosts to express 
interest in the multicast group dst of RA".

Alex

>
> --a
>
>> Alex
>>
>>>
>>> Some of them are orthogonal, some of them contradict each other, some of them align. People may want to optimize differently.
>>>
>>> I'd opt for simplicity - and use the text Erik Kline posted in another email.
>>>
>>> This would allow the developers to adapt for individual use cases on the spot.
>>>
>>> --a
>>>
>>>> _______________________________________________
>>>> v6ops mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/v6ops
>>>
>>> _______________________________________________
>>> v6ops mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/v6ops
>>>
>>
>> _______________________________________________
>> v6ops mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/v6ops
>