Fwd: [net2-wg] Collection TEP draft

[Rodrigo Fonseca]

On 2/17/06, Omprakash Gnawali <gnawali at usc.edu> wrote:
>
> > Minor questions/comments:
> >
> > 1. "RootControl.setRoot() MUST make the current node a root of
> > the tree specified during instantiation."
> > "MTreeRoutingTreeP SHOULD
> > allow a node to be configured as a root or a non-root node
> > dynamically."
> > Shouldn't these both be SHOULD or both be MUST?
>
> Yes, both should be SHOULD or MUST. I would like to say MUST here if
> it seems reasonable to other people.

I think MUST is ok.

>
>
> > 2. This is a departure from the single-sink 1.x model, so it might be
> > nice to explain a little more about having a tree with multiple roots,
> > and the anycast semantics we discussed in last week's call.
> >
>
> I agree. There is one source of confusion about terminology. People
> who take definitions and terminology seriously might not like the fact
> that there are multiple "roots" in a tree. Perhaps we should call
> those nodes "sink" in the TEP and stick with this terminology in our
> discussions.
>
> I am also curious about the tree_id. I think sometimes it is a forest
> id. For example, if I configure bunch of nodes as a sink node in a
> network, each sink will form its own tree, overlapping or not
> overlapping with other trees in the network. In that case, we would
> now be specifying the forest id rather than a tree id -- it is the
> underlying system that determines for you which tree you are on. So,
> maybe we should call this tree_id something else.

So should we just change the name to collection_id? A packet sent to a
collection_id
will reach one of the sinks that share that collection_id. Each sink
is the root of a tree.

>
> > 3. Can getNextHops return nexthops to different roots?
>
> I think the current understanding is it can. Rodrigo ?
>

It can. The way I thought about it is that even though you maintain a
parent (which primarily serves to determine your hopcount), when
routing a packet you may choose any neighbor with a lower hopcount
than you to make progress. The idea is that nodes maintain the
hopcount in the collection_id, but not which specific tree they are
at. Maybe there's a clearer way to explain this.

> _______________________________________________
> net2-wg mailing list
> net2-wg at millennium.berkeley.edu
> https://mail.millennium.berkeley.edu/cgi-bin/mailman/listinfo/net2-wg
>