Nstream sweetness

RouterOS General
1

Thought you guys would like to hear something positive about Nstream.

Setup:

35 mile shot, small hill to a mountain top.
50% Fresnel zone blockage (another, smaller, mountain is in the way).
5.8Ghz
6ft dishes on each end.
No amps.
Almost no noise.
60ft of 1/2in hardline cable on each side.

We first tried Trango (T-link10), no link at all, even without the hardline.

Then we tried WiLan, got a link, but with a very high BER, >90% packet loss.

We were afraid we would need to get an Orthogon or Redline link, but decided to try Nstream with CM9s first.

And… it works, amazingly well.

The signal level is not good (-85 on each side, we still need to do something about that), but it is stable at 12Mbps over the air data rate.

No appreciable packet loss (lost about 50 out of 100000 1400-byte pings, overnight).

TCP bandwidth-test gives us 5.2Mbps, in both directions, simultainiously (it gives us about 9Mbps either way, with uni-directional tests). Extermely stable speeds, less than 100Kbps in fluctuations.

Latency without the bandwidth-test running is always 1ms or less; with the test running it varies between 27ms and 32ms. Wow!

Simply amazing. I can’t wait to see what this link does when we get the signal levels up to a resonable level.

BTW, the exact same system, with Nstream off, gave us speeds that fluctuated between ~200Kbps and ~2Mbps (in one direction), latency between 5ms and 2500ms (without the bandwidth-test running), and ~5% packet-loss on 1400-byte pings.

Nstream rocks, thank you MikroTik!

2

Nice one!
Impressed …

BTW, what radio cards did you use - CM9?

Regards

3

He used CM9’s (stated somewhere in his post…).

I would be interested in the hardware involved (CPU type, speed, system board, …). Any information on this, Eric?

4

D-oh!
Must learn - read first then write …

Also interested in what HW, but for 6Mbps you could actually use a lower speed CPU, even with Nstreme, probably 266MHz would be fine. No need for a P4 monster …

Regards

5

Agreed on the “non-need” for big hardware here. But would be interesting nontheless…

6

eflanery, there is definitely something wrong. You can’t have -85dB signal on 35mile link with 6ft antennas - I’d expect something like -62dB instead!!

If you resolve this issue, you’ll gain even more performance. Believe me.

7

Yes, there is something wrong, there is a mountain in the way. :slight_smile:

That the link works at all, let alone as well as it does, is the amazing part. We are going to try and get the signal level up, if that means moving one end or the other around, making the towers taller, or amping it (ugh…).

As for radios, we used standard CM9s from wisp-router.

The boxes are SuperMicro 5013C-MBs, with 2.4Ghz Celerons, and 256M ram. Not quite as ‘monster’ as some of our MT boxes, but a bit on the big side.

8

Compact, waterproof tower-mount radios would get rid of all that antenna cable - at 5.8GHz that’ll definitely make a difference.

Need a lower power consumption solution (Pent M) I’m guessing.

Regards

9

What version of mt? I agree with the post about tower top radios. Elimiate that 1/2 inch stuff or go to something like 1 5/8 and you’ll see a big improvement.

10

If you find 1 5/8" cable that is rated for use at 5GHz, please post a link. None of the fat heliax is good past 3GHz. We have used LMR-900, but that’s just 22mm (less than 1")

11

Or, switch to 12" coax, then you won’t need a tower at all :smiley:

Sorry, couldn’t resist.

12

Both boxes are currently running 2.8.27.

Tower mount radios would certainly help with the signal level, but…

The boxes are going to be doing other things as well, including a second Nstream hop off the distant unit, running OSPF, doing PPPoE concentration, and acting as 802.11b APs. So, I don’t think any powerfull-enough tower-mountable units exist (perhaps RB500 will be able to handle it?).

Most importantly though, by putting the radio/router below, it can be serviced by non-tower-climbing techs (which constitute the majority of our techs, the tower-climbers are always busy hanging antennas).

As for cables, we have found 1/2" hardline to be a good compromise, much lower loss than LMR/RG type cable, but still fairly easy to work with, and relatively cheap (compared to ‘the big stuff’).

The boxes had already been planned for these locations, they just wern’t going to be doing the backhaul radio link. Upto this point, all the Nstream links we have done were “sure things” (short shots with 2ft panels, signal levels in the -40s); this is the first time we have really pushed it to the edge, and we are really pleased with how well it works (and how little it cost: ~$500 for CM9s, pigtails, and cable vs. ~$5000 for a Redline set; excluding routers and antennas, which would be there either way.).

–Eric

13

Very interesting comparison, Eric.
FYI there are tower-mountable 1GHz radios, and even 1.5GHz - we build them!
However I can see your point about non-tower-climbing maintenance, that’s a very strong argument for WISP deployment.
Does it make alignment of the antennas harder, or do you do that with other test kit attached?

Regards

14

Stephen, what motherboards do you use in your systems? I’d like a lowish power consumption/low heat system that is powerful enough for nstream, and have been looking into the pentium-m based gear (which I think I saw you mention in another post).

15

Hi there,

We have tested a wide range of standard “off the shelf”, customised and fully custom solutions. The overall findings are:

  • VIA boards (EPIA and the like) consistently “lock up” at high speeds with wireless, within 30 minutes to 1 hour. Not a problem at low speeds, but for backhaul, it’s going to happen. It is a BIOS/chipset problem.
  • VIA V series does not work at all with wireless and ethernet - erratic and very low throughput - BIOS problem.
  • Pentium M boards do not suffer the “lock up” problem, but generate about 22W heat which has to go somewhere. So we’ve spent a lot of time integrating a custom CPU/cooling system/case which end up a complete integrated solution. Just the bits on their own aren’t much use, and without vents or cooling system, the board cooks. Major Downside: Pentium M technology is expensive - blame Intel, not me!
  • A custom VIA board solution, with custom-written BIOS, solves the “lock up” problem. Power consumption a lot lower, about 10W. Again, careful heatsinking to get heat out of the box.
    However, this is not a standard board, it is a CPU submodule, and we’re designing a custom backplane for it, with lots of MiniPCIs etc and POE support.
    (Note: someone asked, can I give them the BIOS: nice idea, but won’t work, the BIOS is specific to each board/vendor. So it would lock the board, or worse. Sorry!)

Summary: it is a minefield out there - very easy to create an unreliable solution. The above has taken many months till we have something that people will be happy with. Ultimately the custom VIA solution will be the best, but interim, Pentium M is a safe bet, if you can get the heat out.
Power supplies need to be looked at carefully, there is no nice off-the-shelf POE solution yet for such high-speed boards.

Send me a mail off-line if anyone is interested, Mikrotik don’t like people advertising on their forum (nor would I if it were mine), -

Regards

16

Actually, we have a tech looking at the signal levels down below (or back at the office), calling out the numbers to the tower climber over 2-way radios. This sucks however, and we plan on getting a pair of signal generators and spectrum analysers.

–Eric

17

I did wonder about this scenario. I think many users have the same problem.
Possibly the right answer is low-speed radio kit, hand carried (backpack!) to the tower, with a module showing onscreen readout of signal level, or still better, audio tones for alignment.
That’s what we do for our laser links, we have an “alignment unit” which has onscreen display, voice circuit across the link, plus audio tones for alignment.
Keep the installers happy - that’s important -

Regards

18

You could use a RouterBoard for that - there is an audio alignment mode in RouterOS, the RouterBoard does have a speaker (although not a 500W boomer :wink: ), it’s fairly lightweight and coudl be operated from a batteryback for a reasonable amount of time (surely long enough to adjust your antenna)…

19

Good suggestion -

And you could temporarily apply +V to the coax at the bottom of the tower (unplug from antenna and wireless card first!) to power the routerboard at the top end, which saves the poor installer carrying the batteries up there … and no “flat battery” experiences …
A real chance of accidently “cooking” some hardware if anyone makes a mistake plugging/unplugging the power though -

20

I wouldn’t suggest using the coax for powering an alignment unit - according to Murphys’ Law you will almost surely fry some equipment some time :smiley:
And to power a RouterBoard for a quite short period of time you don’t have to put a 20kg battery on the climbers back :wink: