Re: New Version Notification for draft-dykim-6man-sid6-00.txt

[DY Kim <[e-mail]>]

in-line.
---
DY

> On 4 Sep 2018, at 05:45, Ole Troan <[e-mail]> wrote:
> 
> From a quick glance it looks like you have re-invented multilink subnet routing.
> Which of one flavous is described in https://tools.ietf.org/html/draft-ietf-ipv6-multilink-subnets-00.

If a site consists of a single subnet, SID6 (my model) should basically reduce to the multi-link subnet (MLS?) model. A difference might be that SID6 explicitly talks about deprecation of the subnet ID whereas MLS is silent about that.

To retain that difference, SID might have to successfully justify the additional benefit of having the subnet ID deprecated.

> That particular proposal didn’t go anywhere, for the reasons described in RFC4903.

RFC4930 comes quite discouraging.

> The basic idea was to share a /64 across multiple IPv6 links.
> Steve never wrote down the details, but today, we’d probably do a combination of ND with ARO and an IGP.
> Possibly with some modifications.

Is it possible that we get more elaboration of your updated idea, please? Is there a possibility for your updated idea to converge with SID6 in some way?

> Detecting movement and whenever someone leaves the link is the tricky part.

As for a node emerging on a link, I’ve ‘naively?’ assumed that such a ‘link-up’ detection is already part of IGP (OSPF or IS-IS) protocols.

As for a node leaving a link, it might be tricky to detect as you point out. I’m not quite sure whether this ‘leave’ detection is a necessity. Until all router host routes are updated by the ‘link-up’ detection that should ensue in a short while, the knowledge of a node having left a specific link might not be of any significant help in retaining the e2e transport connection. Lost packets in the intermittent ‘dark’ period should be recovered by a usual transport operation.

On the other hand, one can also argue that prompt knowledge of a node having left a link should still help if the router implements an additional function of storing the packets destined to the very node until its location be identified by a later link-state update by the router observing the link-up by a visiting node. However, this additional functionality on the part of routers might be too far stretched.

Anyway, before going into any more details, we might first come to a consensus whether tackling again on the issue of multi-link subnet in a form similar to SID6 or any other should be a worthwhile 6man WG work.

Thanks.