AF5x connected to Netonix blows its cookies on the RF Fabric

[mhoppes]

For 4-5 years I've been chasing a bizarre issue in our network on one segment and I only just, this morning, figured out what was going on.



Now, some of it makes sense and makes me wonder if I shouldn't be making this change on the switch ports on all of our links -- the other part of me feels like there is some sort of flaw in the software of either the airFiber5x radios or the Netonix switches.



Here's the scenario:

Router--Netonix--AF5x~~AF5x--Netonix--AF5x~~AF5x--Netonix--Router



For the longest time we were chasing a situation where all traffic flowing through the entire chain would stop for about 2 minutes, then just start flowing again. When this would happen you could reach the first AF5x and everything showed fine, but you couldn't reach the second AF5x on that first link or anything beyond.



I believe there was 180megabits available on AF5x link 1 and 125megabits available on AF5x link 2.



We tried numerous things (flow control on/off, port isolation). The only thing that hasn't been done yet is putting a router between the two links -- though I'm planning to do that soon.



Slowly, the issue seemed to go away and would only happen once in a blue moon -- which was acceptable to be filed away on the "we'll revisit this much later" board.


Over time as we've added customer that 180megabit link is no longer providing adequate service levels and the S45-23 small dish was replaced with a S45-30 dish to improve signal and modulation levels.



This change made the first AF5x link go from 180megabits to 265megabits. Almost instantly the issue returned with a vengeance. The path began dropping once, somtimes twice or three times an hours... worst during prime time... but all day long it would drop.



It finally dawned on me this was a buffer issue of some sort. The 125megabit connection wasn't being saturated since there was only 180megabits feeding two different sites. Now with a larger pipe feeding (as back in the beginning when there were fewer customer) the second AF5x link could become saturated.



I watched and performed some testing. Finally, on the Netonix on both ends of the second af5x link I put in place a WRR QoS Rule of 110megabits out (and 25 megabits back on the far netonix). This, in theory, polices the traffic going out the second AF5x link to 110megabits before it hits the radio, and allows a 15megabit headroom buffer.



VIOLA! The issue has all but gone away. I think I counted it happening maybe three times in total today for a period of 5 pings.



MY QUESTIONS: Would it be good practice to start policing RF links on the switch ports prior to trying to cross the RF fabric as a general rule?



Would putting a router at the middle site 100% clear this up.



Who's fault is this (Ubiquiti or Netonix) in that when more traffic is shoved down the second AF5x link the whole fabric falls apart and the RF fabric on the first AF5x stops passing traffic?

[mhoppes]

I see this never got an answer.....