Bullet M not beating Mikrotik RB433 + R5H in Speed :(

[wispwest]

First off I want to let you all know I was REALLY excited on trying these Bullet M's out on a backhaul link. The results were surprisingly horrible. Here is the current setup.

- Mikrotik RB433 + R5H
- Transmitting on frequency 5600 (ZERO noise floor)
- Signal is -64db, 54Mbps TX 54Mbps RX, TURBO, and NSTREAM enabled

Actual TCP Throughput at the tower is around 40-50Mbps RX, 20-30Mbps TX

So here's what I did. COMPLETELY disconnecting the Microtik Radios and power, and plugging in the Bullets at each end, configured as Access Point WDS and Station WDS. Status on the bullets were:

-TX power 25db, signal -52db!
-TX Rate 150Mbps RX Rate 125Mbps
-Frequency 5600 (same as before)

Actual TCP Throughput = 8-10MBps RX, 1-4Mbps TX =wtf!!

As everyone will ask, YES, I've tried EVERY which setting possible, I've tried

-Changing the channel width, 40/20/10/5 (speeds got worse when lowering the channel width, almost no speed at all on 10Mhz)

-Lowering the TX power (didn't do anything)

-Yes, AUTO ACK IS ON, the link is only a little over a mile!

-Tried changing frequencies, no difference, there is no 5.4-5.6 being used where this is anyway (I've run a spectrum analysis from our canopy 5.4 SM)

- And yes, I tried forcing the Data Rates step-by-step all the way down when I saw the same consistant TCP speed until I reached MCS0, when it finally dropped to 4-5Mbps RX 1-2Mbps TX.

Every which way possible, the maximum throughput I saw peaked at 15Mbps RX and 4Mbps TX and that was with AIRMAX on, it seemed to give me a few more Mbps, but still nothing satisfactory.

As of now, we have pullet the bullet M's off the link and swtiched back to Mikrotik, and will be looking into the Mikrotik R5N cards instead. After messing around with the Bullet M's for 1/2 day trying to get decent throughput at just 1.2 miles, I don't think we'll ever touch these again. They may be fine indoors or something, I was able to push around 80-90Mbps in the office with them 20 feet apart, but just too risky for the outdoor backhaul link.

[UBNT-Robert]

WispWest --

What are you ACK settings?

Are you using autoack?

[lastmilesolutions]

-Yes, AUTO ACK IS ON, the link is only a little over a mile!

[wispwest]

yes, as I mentioned, I am using auto-ack

[SPQ]

I have bad expiriance whit TDD links, only FDD microwacve link have preformance like in specification. A use some Exalt TDD profesional link at 5 Ghz a pay over 7k$ and performace its not like in specification aslo i test radwin, same story. Aslo i use Mikrotik routerbord 411AH and UBNT XR5 and I have very good performace for thet money.
I whot to bay Bullit 5 M to test but i see thet performace its poore.

I whont to UBIQUITI ing. make real test of Airmax product and put on forum, becouse as future customer we must know real performance. Aslo to make some chart list whit other similar product.

[karyal]

You should have both links much better than this.. RB433AH with 5GHZ turbo + nstreme should give you at least 50+ mbps half duplex with that signal level, at -55 you should get quite easily at least 72mbps half duplex.
We usually prefer XR5 over R5H on backhauls but at such a distance (you talking about 1KM) you should have no problem at all. If your rate is 54Mbps check RX/TX CCQ, if it's not 100% you are either getting interferences or one you cards is damaged.
Try fiddling with NSTREME settings too (framing policy and size) as you will get different perfomances with different settings, and each link is pretty a story by itself. If you're using a plain 433 (not 433AH) expect quite less througput, but if you get no interferences you should get at least the same throughput on TX and RX test.

About the Bullet M, we just tested it today, both in lab test and production test, over a 3 Km link... didn't have enough time to test longer links at the moment.
We got the same results in both situations, even if outdoor test was performed on a production enviroment with at least other 3 WISP in the area, and the RF in the area is pretty crowded.
Performances have been quite like expected by Bullet datasheet, signal level was in the -60 area, tcp throughput was about 98Mbps half duplex with airmax off, about 85/92 mbps with airmax on (jperf test, 50 concurrent sessions, tcp).
We also tried a full duplex jperf test, reaching 50/54 mbps full duplex, tcp, airmax off, jperf 100 concurrent sessions.
CCQ 100%, Airmax quality reported 95-98%, Airmax capacity 43/45%,
TX/RX rate 150/135
I do have a screenshot of both the airmax test and connection status during the tests, but it doesn't seem to be possible to attach it to the message :?

Note: if you test on single session you will get from 6mbps to about 30 mbps (at least that's my experience) with the Bullet, you really have to use multiple data streams to put it under pressure
Bye,
Ricky

[SPQ]

Its good performance.
Tell me latency.

[karyal]

Its good performance.
Tell me latency.

At full link usage latency goes up to 50/70 ms at 40 mhz channels, over 100 if you use 20 mhz channels or less.. didn't test too much though, as i expect latency to grow quite a little bit when you saturate the link (and the whole point of the test was saturating the link)... i think it would be much more intersting to see how latency behaves on daily use, when link is under pressure, but not saturated. With little (less thank 50mbps hdx) or no traffic latency was stable in the 1 to 3 ms area (but again, it's no news, you can expect this from any link running almos unloaded).
We plan to put the first unit in production next week, we just have to find some backhauls on the nextwork loaded enough to be worth to test, and backed up enough so that if something goes wrong we can switch easily without affecting customers.

[wispwest]

I wasn't complaining about the Mikrotik links speed, just the Bullet M. Maybe I'll hold on to these bullets until the new V5.1 firmware is released, or is it?

[WHT]

Turn off Auto ACK, it hoses my 22 mile link.

[karyal]

I wasn't complaining about the Mikrotik links speed, just the Bullet M.

I didn't mean to compare the two :D i mentioned what you could get from MT just because i feel you're not getting what you should expect from the MT link too ->>> you may have some RF interference problems in the are that do not allow you to get all you can from either the Bullet AND the MT :D
Bye,
Ricky

[lncommunications]

wispwest...how can you achieve 40~50mbit TCP on a 54mbit link, even with nstream and tweaks?

[InoX]

wispwest...how can you achieve 40~50mbit TCP on a 54mbit link, even with nstream and tweaks?
like this http://img31.imageshack.us/img31/363/20090527114832.jpg

[netsplice]

wispwest...how can you achieve 40~50mbit TCP on a 54mbit link, even with nstream and tweaks?

he was using turbo 40mhz channel size which means 108 over the air throughput.

wisp west... this link that was running -52 did you have clear LOS? and what were you using to test bandwidth on the link?

[Headbang]

2 RB433AH/XR5/nstreame 10km very bad freznel, 1 end pointing into a spruce tree, 29db grids. I get a fluctuating link of 28-39mb/s @ 20mhz.

2 nanostation5 LOS good freznel 1km, I get 44mb/s solid @ 40Mhz

2 PS5's 2km 6 houses in the way lots of power lines a few branches lots of noise. I get 44mb/s solid @ 40mhz

All tests are done with filezilla ftp server and filezilla ftp clients 1 stream, passive.

I plan to change first link to bulletM's to test difference. And later want to try dish & rocket on same link. I expect to see 40-50mb/s on bulletM and would love it if the rocket and dish would take it to 50-70mb/s. And the PS5 link is my primary feed, so would love to double it's throughput with rocket and dish.

[WHT]

Have you tried it with Auto ACK turned off ? ? ? ?

Even with AirOS V5, it drags my connection down.

I may have inadvertently suggested in another post that it was all right to have Auto ACK on only one side. It might have been with an earlier firmware, but even with AirOS 5, it still rips apart my 22 mile link.

[InoX]

2 nanostation5 LOS good freznel 1km, I get 44mb/s solid @ 40Mhz
2 PS5's 2km 6 houses in the way lots of power lines a few branches lots of noise. I get 44mb/s solid @ 40mhz

you are the first on this forum who claim more than 35mbps with ubnt products.

[UBNT-Mike.Ford]

First off I want to let you all know I was REALLY excited on trying these Bullet M's out on a backhaul link. The results were surprisingly horrible. Here is the current setup.

- Mikrotik RB433 + R5H
- Transmitting on frequency 5600 (ZERO noise floor)
- Signal is -64db, 54Mbps TX 54Mbps RX, TURBO, and NSTREAM enabled

Actual TCP Throughput at the tower is around 40-50Mbps RX, 20-30Mbps TX

So here's what I did. COMPLETELY disconnecting the Microtik Radios and power, and plugging in the Bullets at each end, configured as Access Point WDS and Station WDS. Status on the bullets were:

-TX power 25db, signal -52db!
-TX Rate 150Mbps RX Rate 125Mbps
-Frequency 5600 (same as before)

Actual TCP Throughput = 8-10MBps RX, 1-4Mbps TX =wtf!!

As everyone will ask, YES, I've tried EVERY which setting possible, I've tried

-Changing the channel width, 40/20/10/5 (speeds got worse when lowering the channel width, almost no speed at all on 10Mhz)

-Lowering the TX power (didn't do anything)

-Yes, AUTO ACK IS ON, the link is only a little over a mile!

-Tried changing frequencies, no difference, there is no 5.4-5.6 being used where this is anyway (I've run a spectrum analysis from our canopy 5.4 SM)

- And yes, I tried forcing the Data Rates step-by-step all the way down when I saw the same consistant TCP speed until I reached MCS0, when it finally dropped to 4-5Mbps RX 1-2Mbps TX.

Every which way possible, the maximum throughput I saw peaked at 15Mbps RX and 4Mbps TX and that was with AIRMAX on, it seemed to give me a few more Mbps, but still nothing satisfactory.

As of now, we have pullet the bullet M's off the link and swtiched back to Mikrotik, and will be looking into the Mikrotik R5N cards instead. After messing around with the Bullet M's for 1/2 day trying to get decent throughput at just 1.2 miles, I don't think we'll ever touch these again. They may be fine indoors or something, I was able to push around 80-90Mbps in the office with them 20 feet apart, but just too risky for the outdoor backhaul link.

Hey Wisp,

How are you measuring the throughput?

Thanks,

Mik

[UBNT-Mike.Ford]

2 nanostation5 LOS good freznel 1km, I get 44mb/s solid @ 40Mhz
2 PS5's 2km 6 houses in the way lots of power lines a few branches lots of noise. I get 44mb/s solid @ 40mhz

you are the first on this forum who claim more than 35mbps with ubnt products.

I havent seen more then 35Mbps TCP with Nano or PS.

Thanks,

Mike

[UBNT-Mike.Ford]

Also WISP,

Please post your configuration files from both sides of the link.

Thanks,

Mike

[UBNT-Edmundas]

wispwest,

aggregation is on on both bullets?

Please post configuration files and describe how you did measured performance.

Thanks!

-Edmundas

[rconaway]

Turn off AutoACK. There is a bug in it right now. We have 2 links up, 20MHz each. Both have problems with AutoACK. Turn it off, they run great. Guess about 5% higher than your distance but you may tweak it during testing for better results. I'm sure this will be resolved in the next firmware version.

[lukic]

Turn off AutoACK. There is a bug in it right now. We have 2 links up, 20MHz each. Both have problems with AutoACK. Turn it off, they run great. Guess about 5% higher than your distance but you may tweak it during testing for better results. I'm sure this will be resolved in the next firmware version.

If there is a problem with AutoACK, how it will be working in AirMax mode with "up to 300" CPE in different distances from AirMax AP? :-) (just wondering how fast is something on market, even if software is still not working properly)

[wispwest]

Sorry for the delay guys. I don't have the config anymore because I took the bullets down and can't log into them, but overall, like I said, I tried EVERY possible mix of settings, execpt turning Auto ACK off, because everyone on the forum besides this thread was saying to make sure Auto ACK is ON, and its a big no-no to turn it off, so it didn't even cross my mind to mess with that. Bandwidth tests, I do the same as i did with the Mikrotik R5H's. We have our own speedtest mini site at our core. Also did TCP up/down test through the Mikrotiks.

[UBNT-Mike.Ford]

Sorry for the delay guys. I don't have the config anymore because I took the bullets down and can't log into them, but overall, like I said, I tried EVERY possible mix of settings, execpt turning Auto ACK off, because everyone on the forum besides this thread was saying to make sure Auto ACK is ON, and its a big no-no to turn it off, so it didn't even cross my mind to mess with that. Bandwidth tests, I do the same as i did with the Mikrotik R5H's. We have our own speedtest mini site at our core. Also did TCP up/down test through the Mikrotiks.

Hello,

The standard MikroTik single stream testing will not stress the link. You need to follow the speed testing suggestions at the top of the forum to see the true speed of the link. Single stream data (not typical for backhauls) can result in speeds 50% lower then what is exepcted.

Thanks,

Mike

[wispwest]

Mike,

I was able to push 70-85Mbps in the office with this set of bullets with a mikrotik on each end using TCP. Whats the difference if I do the same at our tower 1.2 Miles away? I know you say to use a different speedtest method, but it worked in the office 15ft apart fine! The speedtest seems to work fine.

[george]

Important to note that the newer MikroTik versions all use a default of 20 streams for TCP testing. My tests indicate that results are pretty close to Jperf. I can probably come up with exact comparisons at some point.

Once we get these Bullets out in the wild I can always find a couple of MikroTiks to test between, but it isn't always easy to find a couple of PCs. If we can get the correlation of MikroTik to Jperf nailed it will help.

I suspect there are a few of us in the same situation.

George

[UBNT-Mike.Ford]

Important to note that the newer MikroTik versions all use a default of 20 streams for TCP testing. My tests indicate that results are pretty close to Jperf. I can probably come up with exact comparisons at some point.

Once we get these Bullets out in the wild I can always find a couple of MikroTiks to test between, but it isn't always easy to find a couple of PCs. If we can get the correlation of MikroTik to Jperf nailed it will help.

I suspect there are a few of us in the same situation.

George

Hey George,

I just downloaded Btest from the MikroTik website and it only has single stream.

And WISPWEST,

Please post your config files from both ends.

Thanks,

Mike

[george]

Hey George,

I just downloaded Btest from the MikroTik website and it only has single stream.

Yes, they haven't updated the Windows version yet. All the RouterBoard-based bandwidth testers since about 3.25 have been multi-stream. See attached pic.

George