Mesh networking in RIOT

[benpicco]

What’s the state of mesh networking in RIOT? It looks like the RPL implementation degrades after a while, faster so the more nodes are in the mesh. Has someone used this successfully?

I’m aware of @jeandudey’s out-of-tree AODVv2 implementation, although for the sensor use case (many nodes sending data to a single sink) the communication overhead of this protocol appears to be significant.

If I remember correctly, @jue89 presented an application that relied on simple broadcast flooding. While this carries the same communication overhead as AODV (and possible more) it appears to be a quite robust solution.

So what’s the current state of the Mesh in RIOT?

[miri64]

@benpicco

What’s the state of mesh networking in RIOT? It looks like the RPL implementation degrades after a while, faster so the more nodes are in the mesh. Has someone used this successfully?

Any idea why? Having non-storing mode might solve the scaling issues.

[benpicco]

From the output of foren6 it looked like that nodes would at some point start to chose nodes with a higher rank as their parent.

This corresponded with a sudden drop in periodic messages received by the root node.

[kirk]

any debugging/understanding of the rpl behaviour is very welcome !

[jue89]

@benpicco

If I remember correctly, @jue89 presented an application that relied on simple broadcast flooding. While this carries the same communication overhead as AODV (and possible more) it appears to be a quite robust solution.

We are just aiming for star typologies. But we allow them to co-exist without any interference. The RIOT nodes are looking for a gateway using ICMP ping on a specific multicast address and select the one with the highest RSSI/LQI. If package loss is too high, the nodes start over searching for a gateway.