RB333 Works Great but Quirky

RouterOS General
1

After reading the specs for the new RB333 boards, we were very impressed. Our StarOS WAR boards (IXP425) were beginning to see high latency and dropped packets at peak hours, and we were starting to shop around for options.

Our WAR link was running in turbo mode, using Ubiquiti XR5 radio cards and Super A/G mode compression/fast frames. We would see roughly 20 mbit combined up/down across the link at peak hours, and these seemed to be the limit of the boards. They were doing static routing only (StarOS does not properly support OSPF, another reason for the switch), and no connection tracking.

We started by replacing the AP, which did not talk very nicely to the StarOS SU (high latency) - so we promptly replaced the SU. The RB333’s were preconfigured for a fairly standard AP/SU setup - compression off, NSTREME off, turbo mode but otherwise standard settings.

The first issue we ran into was on the SU, we have a roughly 200ft cable run (armoured STP) - and it appears with an 18V power supply, the RB333 would not run two Ubiquiti SR5’s properly. Removing one of the cards worked, and we later installed a proper industrial-grade regulated 28V power supply, which cleared up any power issues.

In a normal AP/SU scenario between the two RB333’s, we were seeing hardly any packet loss - but very high latency, roughly 250-300ms on average. The available bandwidth seemed to be very low. We tried changing almost every setting (compression, preamble, frequencies, ACK timings, etc.), but to no avail. We are pushing about 20-30mbit and 5500-8000 packets per second over the link.

Eventually, we turned on “NSTREME”, “Polling”, and “Framer Size: Exact Fit” - and this immediately solved the problem. The link is now 9-10ms latency solid, and has absolutely no issue handling the load at peak times. We are quite amazed at the performance of these little boards, and are looking at replacing the majority of our backhauls with them. We are also looking forward to trying NSTREME-DUAL when this link starts to run out of capacity as it is.

Overall, very pleased with the performance of the boards - once they are up and running. Getting them to run was a bit of a challenge, and we’re chalking that up to the beta firmware. Perhaps there are some quirks in the standard 802.11 stack, that are overridden when NSTREME is turned on.

The link is roughly 12km long, and has roughly -52 signal strength (on two 4ft dishes).

2

The OSPF problems of StarOS are not the only issues, if you take a WRAP board and you install StarOS and do a performance test and then install RouterOS afterwards and do the same again, you’ll always see an substancial increase in bandwidth.

We’re getting quite good performance of the routerboards (mostly RB532 or x86 base), but just to give you an example, of what you can expect of RouterOS:

Platform: P3 1 GHz, Router 2.9.x, 2 x 23 dBi Elborxrf panels
Radio: 2x Wistron CM9 cards (std. 100mW) running 5 GHz-turbo mode, EoIP-tunnels bonded to get faster one way speed instead of full-duplex as you would get with Nstreme2. Nstreme-polling enabled, exact size framer policy.
Distance: 17 km

That setup on both sides obviously. The general througput we’re getting on a link like that is 33 mbit/s tcp (one way). Average latency is < 3ms.

On StarOS you wouldn’t even have dreamt about getting results like that.

/Martin

3

We did not mention anything about OSPF. We have not even tried that yet.

And also we are comparing a Star-OS WAR 566mhz board to the RB333. So far the RB333 is faster.

We also had to set one end of the link’s ethernet to half duplex or we would see packet loss on the ethernet. It is about a 200ft run on armoured cat5 that ran fine on other hardware.

On Star-os WAR/566 we get about 29mbit, on RB333 about 44mbit.

Ya, WRAPs are crap for speed but no one makes or sells them anymore.

4

I am amazed! 5GHz-Turbo, Panels for 17Km links XR cards and speeds that are close to what you should get with a non-turbo link and proper parabolic dishes!
Glad I am not in your neighborhood… :laughing:

5

Skaught is using XR cards and 4 ft dishes on 12 km, which i agree is a bit overkill. We’re using standard cards and 1 1/2 ft panels, nothing major, for the 17 km.

Sorry, but even in a lab with 2 RB532’s you won’t pass more than 20 mbit/s tcp throughput on a regular 2.4 or 5GHz OFDM link (without nstreme enabled). CPU’s won’t handle more. Also, check the overhead, that’s on 802.11(x). 802.11b at 11 mbit will not pass more than 5 mbit/s tcp throughput.

So what are you talking about ?

/Martin

6

It is the opposite really , the 4ft dishes won’t spray RF around the place as much as other antennas since they have a tighter beam.
If you were to use 10ft dishes with collars they would waste less again, which is why you see such animals hanging from communcation towers sometimes.

7

Correct me if I’m wrong, but won’t testing OFDM equipment in a lab or ‘on the bench’ yield inconsistent results compared to testing it out in “the field?” OFDM was designed to deal with multipath issues… which are difficult to reproduce in a lab environment. “In the field,” radio signals bounce off buildings, trees and other objects as they travel between the two antennas. This bouncing effect produces multiple “echoes” or “images” of the signal. As a result, the original signal and the individual echoes each arrive at the receiver antenna at slightly different times causing the echoes to interfere with one another thus degrading signal quality. OFDM, by design, overcomes these issues when present, but I’m not sure what happens when they’re not. I would think that you would get less impressive results. Anyone?

8

It’s not the size of the antenna, I’m worried about, it’s the combination of the gain of the antenna + the rf power of the radio. Truely, there is less spray, the bigger the antenna, then again that depends on the design, too.

The polish 1 1/2 ft 23 db panels are by far narrower than the 2 ft 24 db PacWireless panels or for example 2 1/2 ft. Stella Mesh at 28 db. Size is not all that matters. If the Stella was designed to chanelize the power, where it belongs, and wouldn’t leak all over the place, it wouldn’t be outperformed by far smaller antennas.

/Martin