Re: [Roll] updating DAO caches - a proposal

[Richard Kelsey <[e-mail]>]

> Date: Tue, 1 Dec 2009 17:04:54 +0100
> From: "Pascal Thubert (pthubert)" <[e-mail]>
> 
> The term label is encumbered with MPLS, and seems to imply a lot more
> than we do here, like LDP, switching...  Maybe in some future we'll
> integrate more "real" labels, but we're far from doing that here.

If we do use labels, why only use them a little bit?  We
need to maximize the utility of whatever mechanisms we have.
It really seems like we are piling on patch after patch and
losing sight of the big picture.

> Richard>RPL is starting to look schizophrenic.  On one hand we have
> Richard>the DAG mechanism which, especially with local repair, aims
> Richard>to make it as cheap as possible to react to changes in the
> Richard>topology.  On the other hand we have the DAO mechanism which
> Richard>maintains a distributed route database and where a single
> Richard>DAG link change can trigger widespread updates.  How likely
> Richard>is it that topology changes will be common enough that we
> Richard>need fast local DAG repair and also rare enough that the
> Richard>resulting DAOs successfully maintain the downward routes
> Richard>without swamping the network with control messages?  We
> Richard>are in danger of painting ourselves into a corner.
> 
> Good question that applies to DAO in general not just this mechanism :) 

To some extent yes, but it doesn't apply to all DAO
mechanisms equally.  For example, sending plain record
routes, either periodically or on demand, would have much
less interaction with DIOs and DAG changes and would have a
much more predictable behavior.

                              -Richard Kelsey