In looking through the changelog for 3.21 and 3.22 there is no mention of any changes to fix HWMP, mesh or anything related. Were changes made to fix the routing problems with HWMP? Can we now test with WDS dynamic mesh? Is it worth our effort or should we continue to wait?

Thanks

Thats exactly what im waiting for also, benn checking the site regularly for update cause I think Normis posted somewhere that it would be addressed in 3.21

yes, I think it was. We did send out some test packages to affected people, and they didn't report any further issues.

In looking through the changelog for 3.21 and 3.22 there is no mention of any changes to fix HWMP, mesh or anything related.

Obviously you wasn't looking very hard.
From changelog:

*) fixed some problems in mesh protocol;

The problems with FDB and forwarding in HWMP still exist.

If we set up a mesh using dynamic-WDS-mesh and put the wds interfaces and the wlan interfaces into the mesh interface, the system does not forward properly to more than 3 nodes. The 4th node causes everything to break.

Even at 3 nodes we see significant variations in ping response times, whether we are using mac or tcp ping.

yes, if we set everything up statically, then we can get better performance. But what we are looking for is dynamic mesh where associations are made dynamically, forwarding tables update dynamically, and there is really no user involvement in configuration to add or remove new nodes to the mesh.

Our understanding was that HWMP would improve upon spanning tree so that layer 2 mesh would be stable, but so far our experience has been that it is no better, and may even be worse than RSTP for dynamic mesh applications.

This is not an RF issue. We have been very careful to make sure that we attenuate all links and keep good SNR between all nodes. The issue is in MAC layer.

So, it looks like not much was fixed in 3.22...Normis can you confirm this?

Thanks

please provide us with the support output files from your routwer where it isn't working. Send them to support@mikrotik.com
Could you describe how to make such setup which isn't working?

So the good news is that HWMP seems to be working better in 3.23. Still not perfect but much more stable than before.

Our test bed is 5 routers, all with low power 5GHz radios. They are all within good RF link quality of one another. We are using WDS dynamic mesh and putting all of the wds interfaces, along with ether1 and wlan1 in the HWMP mesh interface.

One of the routers is the mesh portal.

We can ping across all of the links and the latency and throughput are pretty good. Client ARP tables seem to be correct. The FDB is pretty stable.

We start with the mesh network being off our lan. Every thing is fine. We connect the portal node to our lan, and we can still ping across and things seem to work ok. But, if we bring up the portal node with it already connected to the lan , then we start to see the remote mesh nodes showing "larval" entries in their FDB for everything on the lan. This seems to crash the network.

Obviously it is desired to bring the mesh up with the portal node physically connected to the Ethernet network that it is a portal to.

Is there something we are missing in this configuration? Is there a way to make larval entries in FDB refresh more quickly so that the network does not crash. Seems like once you get 5 or 6 larval entries, the network stops forwarding.

But this said, we are really close to having the nirvana of self configuring wireless mesh with 3.23.