This ones a bit of a head scratcher
Here's the fast and dirty - Customers are getting approx 30mbit/s, though radio's can do at least 180mbit/s, main uplink is capable of 500mbit/s+ and no CPU or ethernet bottlenecks anywhere
The network topology for any given customer is...
DistRouter -> SwitchA -> AP -> SM -> SwitchB -> CustomerRouter
The AP serves 8x LHG60 radio's. Everything is using port-isolation/split horizon and STP is off. Hub-and-spoke topology with no layer2 traffic bleeding between any customers
First thing I did was check SwitchB to the Router, this gets approx 180mbit/s
SM to DistRouter = 180mbit/s
AP to DistRouter = 960mbit/s
CustomerRouter to DistRouter = 30mbit/s
Ok so it must be between the customer and the switch..... nope
CustomerRouter to SwitchB = 960mbit/s
CustomerRouter to SM = 960mbit/s
CustomerRouter to AP = ~50mbit/s
So for whatever reason the 60ghz link is ultimately giving about 1/5th the throughput capacity to the end user
As I start to kick LHG radio's off, dropping the SM count down to 7,6,5,4 etc the speeds progressively get better across the board. But its still at that roughly 1/5th ratio, and with 4x SM's connected then the customer is getting about 150mbit/s and the SM doing about 600-800mbit/s
I am using PPPoE between Customer and Router, but did temporarily setup IP to allow it to test directly to Switch/SM/AP/Router as well as all the way out to the internet (not really any difference)
Have confirmed every cable is doing 1gbit's, no traffic bleed or sub-optimal L2 switching, no CPU bottlenecks anywhere
I've had this weird sort of issue before with RB3011's specifically, where for whatever reason traffic 2+ switch hops away runs appallingly bad 'through' it. LHG and Cube60 also use the ARM chipset so perhaps the same sort of thing is happening here. Has anyone else experienced this? Better yet anyone have knowledge of how to fix this? All radio's progressively updated to 7.15rc2 and no difference
I can't say with 100% certainty but I don't believe it was this bad when the Cube60SA was instead a LHG60 (also serving the 8x LHG SM's) so maybe some sort of bug mixing 802.11ay and 802.11ad?