Chapter 62 – Ubiquiti versus Mimosa Networks PTP– Battle Royale?

Well, not really a Battle Royale more like a thumb wrestling match. It wasn’t what we wanted but the weather Gods just made it tough to do much more so it at least gives us a foundation to start with. At 112 degrees Fahrenheit outside and the temperatures on the roof and tower being much higher, it was determined that the basic information we needed could be garnered very quickly. We have 2 of the four radios, the Ubiquiti AF5x and the Mimosa Networks, the B5c, already deployed and have been running them for a couple years. In the case of the Ubiquiti Prism and the Powerbeam 620’s, we have had those deployed for a couple months in PTMP and PTP environments. The Mimosa Networks C5c PTP radios (same C5c as the multipoint but different firmware) got here today. We have a lot of expansion plans coming up and this link needed to be upgraded so . . .

Due to time, temperatures that would destroy most switches, techs that were imagining new ways to get back at me for making them go out and do this at the hottest time of the day, it turned into a quick test so empirical data is going to be minimal. Since this isn’t the laboratory, we figured that it was good enough and we couldn’t wait any longer for it to cool down. Based on our experience with Ubiquiti and Mimosa, we are confident the numbers on the links are pretty consistent with what we see today with the only wild card being the C5c. On the Prisms or any of the Ubiquiti AC radios we are running now with 8.3 firmware, we see internal speedtests at 70-80% of the PHY speeds in the upper corner on the dashboard. On the Mimosas, their internet speed test it’s pretty much what the Mac Layer shows.

The antennas were a Ubiquiti RD-34 and a Jirous 29dBi deep dish. We also only used 20MHz channels since at 15 miles with the noise in the area, we didn’t think there was any way 40MHz was even going to be usable. We had been running the Prisms in 10Mhz and 20MHz there for a few weeks but due to the interference, 20MHz channels had about the same PHY rates. It was partially optimized to the best channel as we had been weeding them out when we saw modulation and interference issues.

We didn’t use split channels on either the B5c or the AF5x to keep the comparison as close as possible. That and we didn’t want to spend the afternoon trying to find the best 2 channels on the A5x while my techs broiled in the sunshine. I like 10MHz channels and definitely the 1MHz channel granularity. In the case of 1MHz granularity though, there is no way to take advantage of it in a dynamic environment. We didn’t have hours and days to optimize it, especially if we had to get 250lb passed out tech off a roof (so inconvenient) and if a competitor changes channels at whim, which they do, I’m starting the process all over again.

With the A5x, it’s like having to decide between a manual transmission and an automatic. I’m a car guy who grew up with manual transmissions. I still have 2 cars, one a race car, both with with manual transmissions, because both cars are from the past when manual transmissions ruled. If you had a muscle car back them with an automatic, your manhood got questioned. The reason is that in the past automatics were inefficient, slow, and had the nickname slushboxes, which they truly deserved. Cut to today and automatics are fast, efficient, and have as many or more gears than my 10-speed bicycle. Throw in dual-clutches and technology and 0-60 times didn’t get down in the 2 second range because Billy Bob learned to shift faster. Processors and sensors measure everything from wheel-spin to shift rates to acceleration, and so on to squeeze out the most performance. In 2017, processor are smarter than all of us in analyzing wireless environments in real-time, but the AF5x still thinks it’s 2003 and because of that, it’s not practical in our environments if we are trying to maximize throughput. Even the Prism beats it with a simple frequency scan at reboot and the ability to lock out 5MHz channel increments if it makes a mistake.

First observation, AF5x in a high-interference 802.11 dynamic environment, isn’t happening. We couldn’t keep link up on first 3 channel choices on AirView that showed as the most clean. Got connected, dropped, connected, and so forth. This has been our experience with the AF5x and why we also stopped deploying it in our environment. We eventually gave up after a few tries and then went to the Mimosa C5c and let it find us good channel. I know the AF5x was on 3.2 firmware but there were no radio improvements in 3.23 and we didn’t catch that until the end anyway. At that point it was too hot to mess around with it much more. We had already spent more time on the AF5x than any other radio with all the channel changes and resets. The AF5x seems to be the firmware that time forgot.

We started the C5 on auto and it quickly jumped on 5735 as a good channel. It was okay but not great. Shortly after that, it jumped to 5745GHz which was a good improvement with better modulation. We let it settle for a few more minutes and it didn’t move frequency. Since we now had a decent channel to start with, we grabbed a screen shot and moved on. It ended up at 116Mbps down and 87Mbps up on the PHY Layer and 70/52 on the MAC layer.

The B5c was next. Set it to 5745GHz and let it settle down. In all fairness, we didn’t use auto-everything since it probably would have split the channels. The PHY rate was 144/87, slightly higher than the C5c but the MAC speeds were 50/30, lower than the C5c. We didn’t let things settle long enough on the B5c on the MAC side due to the heat. We just assumed that the higher PHY rate B5c vs the C5C was due to the slightly higher power and the rates would settle in at about the same. We also know we would be running split channels on the B5c since we already had several running in that area and based on our experience, would result in higher number anyway. Good enough. Next.

Back to the A5x. Set it to 5745GHz and even spent some time tweaking the power/antenna EIRP settings again to try and coax more out of it. It settled on 2x, 30bps down, 8Mbps up as the maximum. Could it have done better, who knows? I do know in the same channel in this environment on this link, the 802.11ac radios all beat it. We didn’t cherry pick this link so don’t even start. It’s a production link. This was just the most convenient one to run this test on since it’s an area we are expanding in, we had the dishes already installed, and we needed more capacity. We started with the Powerbeam 620acs and ran them for a month with multiple setting changes so we were pretty familiar with the environment.

Put the Prism back up. It went to 42Mbps on the PHY Layer both ways. Our first upgrade from the Powerbeam 620’s to the Jirous/Ubiquiti/Prism combination really didn’t result in much more speed. It was a little more stable but moving between 10 and 20MHz channels with quite a bit of time on reboots and locking out channels never got us above this speed. It was about as optimized as it was going to get and pretty much what we had already seen.

Unfortunately we simply didn’t have time to tweak and adjust every radio to its maximum but this test also demonstrates that we shouldn’t have to. Auto-channel select, at least at boot up, is the way to go for most high-interference environments, especially with multiple competitors. I read AirView on the A5x and I wasn’t even close to the right channel or its alternates. In low interference environments, it is an efficient and solid radio. In this environment, it was overshadowed by all the 802.11ac radios with far less effort.

The C5c in PTP mode is a formidable little radio. Our quick test showed it almost 3 times faster in one direction than the AirPrism and twice as fast in the other direction. In our experience, internal speed tests for both radios show the Mimosa’s in general match their MAC speeds and the Ubiquiti AC radios with 8.3 firmware hit about 70-80% of their rated PHY speeds on the main screens. So in this test, real world results were that the Mimosa C5c was slightly more than twice as fast one way and about 60% faster the other way. Again, since we didn’t have time to do an iPerf test, this is our best estimate. If someone has a different experience set than this, let me know. We’ve got enough of all this equipment up and running right now to confirm our extrapolations of this data except for the C5c so we are confident in the general results and the trends. This was our first link with the C5c but we are now planning more because of this result.

So, what is our conclusion? Those of you who think 802.11ac with the latest chipsets will never be as good as a proprietary PHY layer, I hate to be the one to destroy your fantasy as the products are delivered today. There is theory and there is reality. Could it be better, maybe with a lot more development time on the firmware but who can monitor it all the time? If the AF5x hardware is truly better, I’ll never find out because I plan on retiring in 10 years which is how long it will take to find a good channel in 1MHz increments. Does it work better in non-dynamic environments remote environments? Sure it does because it has higher QAM rates than 802.11ac and if the PHY layer doesn’t have to battle 802.11 airspace or even older Canopy radios, it’s a great radio. Keep in mind that it’s big feature, improved filtering, only works if the interference is in adjacent channels, not in the same channel or covered spectrum and you don’t have to change channels very often or tweak it. In that scenario, its lower latency and higher spectral efficiency make it a good choice for some links. But we also know what happens when an 802.11 signal invades its space, the AF5x rolls over and plays dead. Every radio has its strengths so apply accordingly.

The reality is that 802.11ac research and development dollars far exceed the budgets or development of proprietary PHY devices. Do you still have the overhead that makes it less efficient spectrally, yes. But 802.11ac’s (or even 802.11n which is what all the ePMP Cambium users have been saying) ability to adapt to noise in band is still superior to anything proprietary and it ends up with far more throughput. And if you follow where 802.11ac is going for its next generation of chipsets and if it gets into our industry, trust me, the battle is so far over, that it’s not even worth attempting to continue that development.

The other thing we got out of this is that the C5c is the real deal for a budget PTP radio. I liked the Prisms and really, really liked the PowerBeam 620ac mechanicals. But the C5c at 60% less cost than the Prism and only 30% of the cost of the AF5x in a high-interference environment had double the throughput. Even against its older and more capable brother, the B5c, you have to really look hard to justify 4 times the price unless your noise is simply too insane for a single channel. The little radio that could just set a new low-cost standard in price/performance for us.