Hi Richard

> > The flow label helps the router select the instance for a packet.
> 
> "Helps"?  Doesn't the flow label tell the router exactly which
instance to use?

At the moment you're correct, it's one to one. I've heard voices (was
that Kris and Phil?) asking for a more complex stateful mapping at the
RPL network ingress, and a new field in the new RPL option for the
instance. My vote is to leave the instance ID in the flow label where it
is today, since it is end to end application data.  While I agree that
the loop avoidance data  fits very well in a hop by hop option long as
the option stays contained in the RPL network. Which is what Jonathan's
draft does for all I understand so I'm fully happy there.

> 
> The trouble I have with this is that "select a router"
> sounds a lot like routing.  We already have a mechanism for selecting
which
> router to send a packet to: DIOs.
> If a device wants to select between next hops based on routing metrics
then
> it should be a router.
> 
> Why is this such a burden?  Listening to DIOs and making routing
decisions
> based on them does not require sending DIOs or forwarding packets for
> others.

In the host world, there is such a thing as a default router list. The
host builds that list listening to RAs, and applies its best practice to
select the router it will use for a given packet out. A number of
considerations will influence that decision, like the age information,
Mobile IP information, multihoming information, etc...

RFC 4191 is an obvious such consideration that provides guidance to
proactively setup a table that redirects would have setup reactively
anyway. You'll note that RFC 4191 is very voluntarily NOT a routing
protocol thing, and really a router to host to interface. The key here
is that the preference IS NOT a distance. Other hints come from the
mechanics used (RA most usually is router to host though it's been
extended to R2R sometimes like in MIPv6).

In a same fashion as routing information can be transformed into a RFC
4191 RIO, instance information should be made available to the host.

Pascal

> -----Original Message-----
> From: Richard Kelsey [mailto:[email protected]]
> Sent: Friday, May 07, 2010 4:52 PM
> To: Pascal Thubert (pthubert)
> Cc: [email protected]; [email protected];
> [email protected]; [email protected]
> Subject: Re: [Roll] [6lowpan] how does a node get an IP address
> 
> References:
> <[email protected]><4bd58ed9.8070...@gridmerg
> e.com><[email protected]
> .local><[email protected]><0a9b01cae5a3$79d6a320$6d83
> e9...@com><6a2a459175dabe4bb11de2026aa50a5d01bca...@xmb-
> AMS-107.cisco.com><87d3xl0zmn....@kelsey-
> ws.hq.ember.com><[email protected]><87bpd42
> 9wp....@kelsey-
> ws.hq.ember.com><[email protected]
> .fi>  <6a2a459175dabe4bb11de2026aa50a5d01c46...@xmb-ams-
> 107.cisco.com>        <8a977bdc5a7b0e429b0f521e8d6f91ee01f6d...@xmb-
> AMS-103.cisco.com>    <[email protected]>
>       <6a2a459175dabe4bb11de2026aa50a5d01cd9...@xmb-ams-
> 107.cisco.com>        <[email protected]>
>       <6a2a459175dabe4bb11de2026aa50a5d01cd9...@xmb-ams-
> 107.cisco.com> <[email protected]>
> <[email protected]>
> <6a2a459175dabe4bb11de2026aa50a5d01d60...@xmb-ams-
> 107.cisco.com> <[email protected]>
> <6a2a459175dabe4bb11de2026aa50a5d01d60...@xmb-ams-
> 107.cisco.com>
> X-OriginalArrivalTime: 07 May 2010 15:01:59.0538 (UTC)
> FILETIME=[3E3CA520:01CAEDF6]
> Return-Path: [email protected]
> 
> > Date: Fri, 7 May 2010 16:33:41 +0200
> > From: "Pascal Thubert (pthubert)" <[email protected]>
> >
> > The flow label helps the router select the instance for a packet.
> 
> "Helps"?  Doesn't the flow label tell the router exactly which
instance to use?
> 
> > It does not help a host that needs to select a router that serves
that
> > instance.
> 
> The trouble I have with this is that "select a router"
> sounds a lot like routing.  We already have a mechanism for selecting
which
> router to send a packet to: DIOs.
> If a device wants to select between next hops based on routing metrics
then
> it should be a router.
> 
> Why is this such a burden?  Listening to DIOs and making routing
decisions
> based on them does not require sending DIOs or forwarding packets for
> others.
> 
>                               -Richard Kelsey
_______________________________________________
6lowpan mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/6lowpan

Reply via email to