[net2-wg] collection stuff

Philip Levis pal at cs.stanford.edu
Thu Apr 20 23:44:33 PDT 2006


On Apr 20, 2006, at 8:31 PM, Rodrigo Fonseca wrote:

> The tought was to avoid having the RoutingEngine as part of the  
> data path.
>
> I have a simpler suggestion:
>  - there is no routing header
>  - the BasicRouting interface becomes
>          error_t command getNextHops(&nextHops, &len)
>          uint8_t command getHopCount()
>
>  All the info necessary for the data path lies in a header filled in
> entirely by the FE.
>
> The interaction between FE and RE is through the BasicRouting, for the
> purposes of forwarding messages.
>
> For address-free collection, I think that works well.
>

Hrm. I see both sides of the issue. I guess the question basically  
boils down to who is specifying the routing topology. The way I saw  
it, the collection header (source, client id) could be completely  
separate from the topology, and therefore the component that  
maintains the topology should be putting in the network headers.

In the general case, the routing topology is a collaborative  
decision; in this specialized case, it's the routing engine.

I think that if we remove the routing header and say that the link  
estimation header has per-hop source and dest, then we can make the  
specialized implementation simple without distancing it so much from  
the general approach that the work done won't be useful. Does that  
seem like a good approach to you?

Phil



More information about the net2-wg mailing list