Today I finished some extensive testing on a 4 km backhaul between a Netmetal and 911G-5HPacD

I started with v.6.40.5, tested v.6.40.8 and ended with latest 'current' v.6.41.1

I tried both 'n' protocol as 'ac' for all three working modes. nstreme, 802.11 and NV2.

First in v.6.40.5 I tried to find the best setting for each of the working modes by altering some of the fine tune settings (preamble, guard interval, Hw retries)
All tests done from behind the link connected routers (SXT-ac and Sextand) towards a rb750G r3 upstream.
All tests done with a single stream and only the download (from rb750 towards the SXT/Sextant)
Each test ran at least for 2 full minutes and done over a 'live' link that also has some customers traffic.
The link runs in a region with plenty of other 5Ghz usage so nowhere a real 'quiet spot' is found in the spectrum. The one we used we found after days of trying for the best...

2 major conclusions and some minor to be made after some 30 tests (4 hours! of work) are:

1. Upgrading (both link units) from 6.40.5 via 6.40.8 towards 6.41.1 does not bring spectacular improvements. Latency in general goes down a bit, but speed not. In fact the highest consistent throughput is measured in nstreme v.6.40.5!

2. Shocking is that in all tests nstreme outperforms NV2 by 200%! And 802.11 is a good follow-up to nstreme but slightly slower. (Some 5-10%). The only thing NV2 is good in is latency. That is consistently around 3-5 ms where nstreme is around 25 - 35 and 802.11 is around 30-45 ms.

3. Difference between 'n' and 'ac' is marginal. 'ac' is slightly better in throughput but more variable then in 'n'. The ping though is slightly more stable where in 'n' it is more variable.

4. The 'nstreme' protocol can take up to a minute to connect and looks like it needs a 'learning curve' before it becomes stable. NV2 and 802.11 make almost immediate connection (station has frequency set in the scan list). NV2 builds the highest modulation sightly faster then 802.11.

5. 'nstreme' seems to have more issues in reaching, and keeping, the highest modulation rate, but overall the throughput, even when modulation drops for a little while, is still higher then NV2.

I have done this test after several others claimed NV2 wasn't as good as it could be. Also after I myself on several links saw the same effect that 802.11 or nstreme is better.

This is still a test that doesn't means it is representing for all networks, although I think my network is pretty typical.

Imho Mikrotik needs to send their NV2 (for backhauls, this is a backhaul link. It might well be on P2MP the result is different. Although also there I found 802.11 outperforming NV2 in some of my AP's) technology back to the draw board and do a hell lot more of testing.

I made a spreadsheet with the findings of the test. In fact much more variables could be tested too. But the matrix over 3 different software versions, 3 different protocols and 3 different IEEE standards ('n' or 'ac') together with the several variables for each of the protocols that can be set has force me to set some variables fixed for all tests or only test them in some occasions.

If we think we have at least 5 (or more) variables per protocol (3), times 2 standards ('n' and 'ac'), times 3 ROS versions then we already talk about 5 x 3 x 2 x 3 = 90 different setups........
It's just too much to test these all........ (And every link is different again!)

This is the spreadsheet I made. If you have any questions, I'll explain or comment.

Hi WirelessRudy,

If you have any questions, I'll explain or comment.

Interesting. Out of curiosity why are you running nv2-mode=fixed-downlink ? It doesn't seem like a fair comparison to the other modes which will essentially have twice the airtime available to them?

Hi WirelessRudy,

If you have any questions, I'll explain or comment.

Interesting. Out of curiosity why are you running nv2-mode=fixed-downlink ? It doesn't seem like a fair comparison to the other modes which will essentially have twice the airtime available to them?

The downlink rate is set to 50 so meaning it is exactly the same as in the other modes. But I did try both variable and fixed downlink with a ratio of 70 but it didn't bring any better performance. In fact, in this link (and I have seen others too) 'variable-downlink' makes things worse.... Link becomes very unstable in regards to throughput.

Well done for taking the time to conduct these tests, I am currently disabling unused system packages (Hotspot, Ipv6,mpls,routing) on the CPE's to see if there any improvements,
as ROS is almost all based on software and recall what another user posted way back and I paraphrase "Just because you can do it in software does not mean its the best way to do it!"

Well done for taking the time to conduct these tests, I am currently disabling unused system packages (Hotspot, Ipv6,mpls,routing) on the CPE's to see if there any improvements,
as ROS is almost all based on software and recall what another user posted way back and I paraphrase "Just because you can do it in software does not mean its the best way to do it!"

We do that already for years. We only enable those packages needed. We also have no firewall rules, mangle or nothing in the CPE's. Only nat and PPPoE. The less processing in the CPEthe better imho.

Today a new discovery, well "discovery..."

We have one Omnitik 'n' serving some 20 clients, mainly SXT's.
Today we needed to test a LHG that couldn't connect in the field. So I set it up to connect to this Omnitik from my house.. no problem whatsover.
The Omnitik runs some 20-25Mbps towards clients in NV2 protocol and v.6.40.3 running...

Out of curiosity after my experiences of the tests from yesterday and before I'd tried 802.11 now here as well. Bang, immediately I double the throughput on several devices!
Now suddenly 50, 60 Mbps to the same units that never ran that in NV2.

I am starting to get serious doubts about the whole NV2 protocol of Mikrotik. It seems to be sick and the more I look to it the more sick it looks.....

TI am starting to get serious doubts about the whole NV2 protocol of Mikrotik. It seems to be sick and the more I look to it the more sick it looks.....

Yes 100% confirm!
If you connect two Mimosa C5C (no GPS, no Sync etc) with a 20 MHz Channel you will see about 130-140MBit TCP throughput
and a latency of 1-3ms.
Try this with Mikrotik TDMA implementation

TI am starting to get serious doubts about the whole NV2 protocol of Mikrotik. It seems to be sick and the more I look to it the more sick it looks.....

Yes 100% confirm!
If you connect two Mimosa C5C (no GPS, no Sync etc) with a 20 MHz Channel you will see about 130-140MBit TCP throughput
and a latency of 1-3ms.
Try this with Mikrotik TDMA implementation

Is that throughput result from using Mimosa Wi-Fi Interop or Mimosa GPS Sync (TDMA) !

...............
Out of curiosity after my experiences of the tests from yesterday and before I'd tried 802.11 now here as well. Bang, immediately I double the throughput on several devices!
Now suddenly 50, 60 Mbps to the same units that never ran that in NV2.

....

Has the latency increased with 802.11 ?

...............
Out of curiosity after my experiences of the tests from yesterday and before I'd tried 802.11 now here as well. Bang, immediately I double the throughput on several devices!
Now suddenly 50, 60 Mbps to the same units that never ran that in NV2.

....

Has the latency increased with 802.11 ?

As I already saw in the extensive test, latency is the only good thing of NV2. 2-4ms on a P2MP setup where both nstreme and 802.11 are more in the 20-30ms region.
Just done again a new test on yet again another Omnitik somewhere in the field. Clients are both at 500 meters as at 2,5km connected. Again the same result. Swapping to 802.11 doubles the throughput on a single client. I did the test on this client while a made a 8Mbps download running and setup up a 200ms timeout ping test. 24-35ms in about 5 mins running, but mostly around 30-32 and pretty stable.
Latency in itself is 'the lower, the better' but jitter is probably even more important for voice and video conversations. So yes, NV2 is better but at the expense of halving the throughput?

I'd wish some MT guy would have a say about these findings... I am afraid they are still in the 'denial fase'....