Re: RFC4941, temp-addr only, etc. (Fwd: New Version Notification for draft-gont-6man-non-stable-iids-00.txt)

[Mark Smith <[e-mail]>]

On 27 May 2016 12:51 PM, "Lorenzo Colitti" <[e-mail]> wrote:
>
> On Fri, May 27, 2016 at 12:00 PM, Mark Smith <[e-mail]>
wrote:
>>>
>>> A simple algorithm would seem to be:
>>> If you don't support, or don't want to use, DNA, then every link is a
new link and you start off with no IP addresses. So you form a link-local
address, send an RS, etc. etc. starting from scratch.
>>
>>
>> That is significantly reducing the robustness of e.g. TCP connections,
as they won't survive that event because their address will change.
>
>
> Most mobile devices close all TCP connections every time a network
disconnects, anyway.

My laptop running general purpose Linux doesn't. Servers don't.

I think that is the insight. I don't think you're considering any other
types of device in this discussion, meaning you won't be seeing where
stable addresses that have increased privacy properties (compared to
EUI-64s, but not in comparison with RFC4941s - which don't meet the
stability requirement) are beneficial.

>The reason for this is that if the device switches to a different network,
the TCP connections are dead anyway (because the IP address changes) and
apps just get stuck. Keeping the connection alive is only useful if you
think that there's a very high chance of reconnecting to the same network
soon, and you either have no other network in the meantime, or you think
you're going to reconnect so soon that it doesn't make sense to
re-establish a connection on another network.
>

>>>
>>> If you support and want to use DNA, then you use DNA with the same MAC
address as you had before.
>>
>>
>> The issue with DNA is that it seems to be relying on the MAC address and
therefore the Link-Local address of the on-link routers being stable for as
the link remains operating. If routers were changing their MAC addresses
periodically too, then DNA will fail. It seems you've been arguing for no
stable L2 and/or L3 addresses on any device, so that would have to mean
that routers don't have them either.
>
>
> Actually, I don't expect that routers will need random MAC addresses.

I think it depends on the type of router. Service provider or
corporate/enterprise router I'd agree.

Residential router, I'd disagree, unless there are any significant
drawbacks.

First thought is new random MACs on boot. The privacy enhancement would be
both periodically changing MAC addresses in WiFi SSID announcements, and no
disclosure of the router NIC manufacturer.

I'll look to find time to do this on my residential router in the next few
weeks.

Regards,
Mark.