[net2-wg] Proposal for dissemination structure and interfaces

[Philip Levis]

On Wed, 2005-12-14 at 08:31 -0800, Jonathan Hui wrote:
> On 12/14/05, henri dubois-ferriere <henridf at gmail.com> wrote:
> > Question: Is it possible to take advantage of a multi-basestation
> > setup with this interface? (e.g., originate the same (key, value) from
> > multiple sinks)
> 
> I have the same concern. Sure, the network will be eventually
> consistent, but using the AM address to break ties will kill any
> possible optimizations of using multiple base nodes to speed
> propagation. If we're pushing for efficiency, I'd argue that the
> client must be able to inspect the payload and help make decisions on
> whether data is different or not. If we're pushing for interface
> simplicity, then maybe suffering the performance hit is okay. This
> inteferface only addresses small payloads after all.

I think that you can take advantage of multiple base stations with this
setup. All you need to do is install the same (version, key, value) on
all of them. The address in the key for tie breaking purposes denotes
the originator of the (version, key, value) tuple. Having each base
station independently originate identical values won't lead to identical
keys. So you generate the (version, key, value) pair elsewhere and
install it on each of the base stations. You're going to need some
coordination anyways, if you want to disseminate the same value.

This approach also keeps you to a single protocol: installing a new
(version, key, value) on the base stations is the same as installing it
anywhere else, except that you're doing it through a wired connection.
Otherwise, you get into a situation where you have a separate protocol
for changing the value, which then triggers a base station to change the
version, triggering dissemination.

Ohil