Observation . Bandwith Test...strange behavior?


#1

Background:-
Here is the network setup:-

Site A:-
Mikrotik (x86 i3) —> B11 + power brick (CPE)

Site B:-
B11 Powered by Netonix (AP)
Mikrotik CCR 1009
Ubiquiti AF 24 powered by Netonix

Site C:-
Ubiquiti AF 24 powered by Netonix
Mikrotik CCR 1009

=============
Observation:-

Running MT BW Test between Site A & Site B report normal behavior
Running MT BW Test between Site B & Site C report normal behavior

Running MT BW Test between Site A & Site C (this is where we are seeing something very interesting.

#1 (following tests run one at a time)
a) MT BW Test on Site A (tcp / 20 stream / receive from Site C) = approx 500 - 600meg.

b) MT BW Test on Site C (tcp / 20 stream / receive from Site A) = 25 to 30 meg ???

But…
If I run start test (a) above, and then start test (b) we get
Site A/rx from Site C = 400meg
Site C/rx from Site A = 240 meg ???

If I stop the test(a) then the Site C/rx from Site A drops instantly to 25 to 30meg ???

=============================
The B11 link is in Auto Mode Site A = CPE & Site B=AP

If I change the mode from Auto to 50/50
I don’t see this behavior, but I only see about 250-300 meg in each direction only.

Any suggestions, clues as to what may be going on ?

(For Mimosa Techs, I can provide remote access to all of these three sites if needed, and also the B11 link can be seen from the cloud)

Thanks
Faisal


#2

Anybody ? Anybody ?

Any suggestions ?

BTW, when running a duplex test, we are are not seeing full throughput when running a duplex test…
either TCP or UDP

MAC TX shows 520 x 520 getting more like 350/360 in each direction on a bidirectional test…


#3

You have bad SNR probably


#4

SNR is 41/42 on one side and the other side is 46/46


#5

I have seen the same thing on B5 if the SNR is that high. Lowered the power to stay at 28-34 SNR and got better results.
You might try and see if that helps.


#6

Or you have to check for frequency overlaps maybe?


#7

Interesting…The radio setup is on auto… .thus it is managing it’s power … I can see from the dashboard it has reduced it from 20 to 17 automatically…


#8

It;s a licensed link, so the channel is clean or lets say, as clean as it needs to be.


#9

Good Tip in regards to SNR… initial tinkering in reducing the SNR from over 40 to mid 30’ s is showing favorable results.

Thanks.


#10

Good to hear.
I also expected a high SNR to be beneficial, but found that 30-35dB is the sweet spot on my B5 units. I guess a higher SNR is starting to overload the receiver. Mimosa could possibly enhance the AGC function to compensate if it is related to the receiver…
Could also be that the transmitter is operating with some non-linearity (compression) at high level causing the issue, and lower TX power will improve linearity which is very important with this type of modulation.


#11

Faisal, it was nice that we had a chance to take a detailed look at your B11 link. The link data collected over the past week on the Mimosa Cloud looked good, except for a very slight increase(2 to 3%) in PER on one side of the link. The SNR was in the mid-40s, so there is an opportunity to reduce transmit power. We did see higher through-put when we reduced the transmit power. In some cases, less is more. Another thing we found was sign of an interferer on the lower part of channel 1. It is possible that the PER level could be explained by presences of this interferer.


#12

Thanks for the assistance earlier today. I will be doing a bit more tweaking and testing, and will report back here.

BTW, any explanation for the behavior listed in the original post regarding the Auto vs 50/50 setting ?


#13

This really sounds like a flow control issue. The B11 does not have flow control but you really need to set it on the AF24 and the mikrotiks and the netonix.


#14

Auto TDMA adjusts the traffic split dynamically based on load. The bandwidth test consumes available bandwidth (not any already in use). You may get different results based on these two behaviors depending on existing traffic, timing of tests relative to each other, and RF conditions that might cause asymmetry. It is difficult to say what combination of these leads to the observation. We can dive in to investigate if you need - all it takes to email support@mimosa.co to create a ticket.