We have posted several times trying to get to the bottom of using HWMP in a mobile mesh architecture. Like what other systems like Firetide or Strix does using ad hoc mode and OLSR or AODV.
The HWMP protocol looks like it should work, but we can get it to scale beyond 4 mesh nodes before it fails. This is the same issue we see if we just use dynamic WDS with RSTP.
While we would like to set up this mesh using single radio 2.4GHz, the diagrams show dual radio 2.4/5GHz systems, where 5GHz is used for mesh backhaul and 2.4 is for client access.
The mesh needs to be self healing, self configuring. If obstructions move around or links are lost, the paths need to reconfigure and traffic should not be lost.
There would be one mesh portal with Internet access. This would be a NAT server to the mesh. It would be the single DHCP server to the network, including all of the client devices. Clients should be able to communicate across the mesh as well.
Has anyone been successful in implementing HWMP over dynamic WDS links? What was the trick?
Our target range would be 500-1000 meters with a nominal datarate between nodes of 2Mbps and no more than a total of 6Mbps of total traffic on the mesh. Latency needs to be less than 80ms between any two nodes and not too variable. If network reconfiguration is required, it should take less than 1000ms.
So far we can get a network like this to work with 4 nodes using WDS and RSTP, but add a 5th node and everything stops.
We want it to be simple so that users don’t have any setup, assuming that there is no generalized interference in the area.
Today I started reconfiguring a 15 node hotspot from WDS-bridge-RSTP to HMWP mesh. We have managed to restore all the links but will need more time for testing. I will keep you posted.
So far so good. I’ve replaced all the bridges with mesh interfaces and once the routers have learnt all the routes from their peers - seems to take a few minutes - it seems very stable and possibly more responsive.
Haven’t managed to get dynamic-mesh mode to work so I am using wds-mode=dynamic wds-default-bridge=mesh1
Thanks Uldis, we are anxious to see the improvements.
The issue seems to come up when there are multiple paths across the mesh, a change occurs, then the FDB does not seem to be refreshed properly and paths just don’t seem to work.
Kind of looks like the same kind of issue we see with STP convergence times. That the network takes a long time to reconfigure itself and during that time traffic is frustrated.
We will wait to see what the 3.21 release looks like.
Just checking to see if auto tx power leveling and auto channel selection is important in your deployments. If so have you figured outt a method to achieve this? We have looked very hard at Cisco and others that have a method of adjusting tx power up and down as well as choosing the best channel to transmitton.
The mesh looks fine in the Dude. I can reach every node and ping from node to node across the mesh. But when I connect a client PC, it gets a DHCP address from the gateway router, but I can ping no further than the actual node I’m associated to.
This seems like a pretty fundamental bug, and has cost us time and revenue. Can I be confident 3.21 will fix this or do I go back to a bridged network as before?
