dslreports logo
site
 
    All Forums Hot Topics Gallery
spc

spacer




how-to block ads


Search Topic:
uniqs
5575
share rss forum feed

rconaway8

join:2005-11-10
Phoenix, AZ

1 recommendation

reply to mkaishar

Re: UBNT Rocket 5M 1/4 Mile Performance Numbers

Numbers are hard to pin down without all the details between manufacturers. I covered this in my articles, "Tales from the Towers" but here is the summary as it relates to Ubiquiti.

The Atheros chips are more efficient with multiple streams going through. For example, assuming you were doing a single ftp transfer from a Nanostation 5M to another 5M and your modululation rate was MCS(15) with zero errors. You will see about 35Mbps. Run 2 parallel transfers one way and you will see 70Mbps. Add a 3rd and it starts peaking out around 85Mbps. Now starts doing the same thing the other way and the aggregate bandwidth keeps going up until somewhere between 140Mbps and 150Mbps. Now this assumes a 1500 byte packet. Reduce the packet size and then you find the real limit is actually 19,000pps. Proxim is no different in terms of total bandwidth although it can handle, and I'm estimating here, around 35,000pps. I suspect a Motorola PTP600 is probably around 50,000pps but I'm pulling numbers out thin air based on throughput that I've seen.

So, there really is no reason for a 1Gbps when your pps limit is 19Kpps anyway. This is what you should design your links around. If it's not fast enough, run parallel radios and a pair of Microtik routers and set them up with bonding. In reality, the next major jump over this is a set of Proxim's at $4K for 20% more performance or start looking at 24GHz or licensed links with full-duplex.


38632383

join:2009-09-25
Houston, TX
said by rconaway8:

... the aggregate bandwidth keeps going up until somewhere between 140Mbps and 150Mbps. Now this assumes a 1500 byte packet. Reduce the packet size and then you find the real limit is actually 19,000pps.
Why are you comparing bandwidth to pps ? What chart did you use to correlate ?

Does reducing the packet size improve the "real limit" of bandwidth as well as improving the "pps" ?


38632383

join:2009-09-25
Houston, TX
reply to DaDawgs
said by DaDawgs:

Rather than debug his results, tell us what you did?
That is off topic. I have discussed my own experiences in my own thread.

rconaway8

join:2005-11-10
Phoenix, AZ
reply to mkaishar
PPS are basically fixed. If you have smaller packets, you have less bandwidth. think back to the old X-Modem, Y-Modem throughput differential. That's why a lot of guys ask for larger packet support to improve the bandwidth.

mkaishar

join:2000-12-20
united state
Reviews:
·Verizon FiOS
reply to mkaishar
Hello everyone...don't get your panties in a bunch

We run Bridgewave 60Ghz GigE full duplex because we need it and this is our primary link to connect our 2 buildings together that are 1/4 apart. This link has been up and running without any issues for over a year now.

UBNT's are the backup link in case the bridgewaves go offline

I just wanted to show that the UBNT is actually very good in comparison relative to price.

Obviously every situation is different depending on perspective and the area, our environment is very easy to diagnose, troubleshoot and resolve because a 1/4 mile wireless link is so short.

I can do more tests...easily and post results...the links are still in testing mode and have not been integrated into production.

-Mark


WHT

join:2010-03-26
Rosston, TX
kudos:5
reply to raytaylor
said by raytaylor:

I believe that the ubnt 150mbit claims are when they are doing a 2 way transfer over the air.
»ubnt.com/forum/showthread.php?t=18672
Why claim 150Mb throughput?

mkaishar

join:2000-12-20
united state
Reviews:
·Verizon FiOS

1 edit
reply to mkaishar
Click for full size
Click for full size
OK, so I have some numbers after more tests today for simplicity and ease of understanding we will call each building location Warehouse1 (W1) and Warehouse2 (W2)

--------------------------------------------------------
First test with AirMax
iperf -c 10.1.1.111 -P 10 -i 1 -p 5001 -f m -t 30
(10 streams running concurrent on each side)

W1W2
[SUM] 0.0-30.1 sec 148 MBytes 41.2 Mbits/sec

W2W1
[SUM] 0.0-30.1 sec 148 MBytes 41.2 Mbits/sec

--------------------------------------------------------
Second test without AirMax
iperf -c 10.1.1.111 -P 10 -i 1 -p 5001 -f m -t 30
(10 streams running concurrent on each side)

W1W2
[SUM] 0.0-30.1 sec 178 MBytes 49.5 Mbits/sec

W2W1
[SUM] 0.0-30.1 sec 178 MBytes 49.5 Mbits/sec

--------------------------------------------------------
Same test above with larger tcp window size and max 1500 byte for 60 seconds test

iperf -c 10.1.1.111 -P 10 -i 1 -p 5001 -w 10.0M -M 1500.0B -f m -t 60
(10 streams running concurrent on each side)

W1W2
[SUM] 0.0-73.6 sec 576 MBytes 65.7 Mbits/sec

W2W1
[SUM] 0.0-73.5 sec 576 MBytes 65.2 Mbits/sec

--------------------------------------------------------
This test is 10 streams but only from one side
iperf -c 10.1.1.110 -P 10 -i 1 -p 7700 -w 10.0M -M 1500.0B -f m -t 60

W2W1
[SUM] 0.0-68.9 sec 775 MBytes 94.3 Mbits/sec

--------------------------------------------------------

I've attached a screen shot of UBNT's speedtest, now I ran this test while running iperf tests

I would like to know how:

1. UBNT's speed test is run?
2. what program is being used?
3. how did they come up with those results as compared to my end results using iperf?
4. how come their speed test takes less than 10 seconds to run and provide a result as compared to my iperf tests?
5. are they truly measuring bandwidth usage while running the test or are they taking a snapshot of an estimate based on the TX/RX rate?

Conclusion:

1. AirMax does affect performance in a PTP...thanks guys...it will be turned off.

2. The Ethernet Full Duplex capability does help with performance as in one of the tests we can push 65Mbits concurrent on each side

3. The Ethernet 100Mbit limitation will not allow to push max than that ports technical/theoretical speed of 100Mbits/sec give or take a few Mbits based on TCP over head, packet size, interference, etc...

Conclusion...no matter how much you market the product and claim 150Mbits throughput aggregate over the air, udp throughput, etc, etc, etc, you will NOT be able to push more than the slowest port in the equation and that is OK with me

-Mark

Additional notes:
The product is good, it works well, it is priced very well, you just have to bypass all the marketing fluff (this is with any product) and accept engineering/technical limits.



38632383

join:2009-09-25
Houston, TX
said by mkaishar:

...

Conclusion...no matter how much you market the product and claim 150Mbits throughput aggregate over the air, udp throughput, etc, etc, etc, you will NOT be able to push more than the slowest port in the equation and that is OK with me ...
With your 130 Mbps aggregate bandwidth you did not miss the marketing hype by very much. Not too bad for your first UBNT link !

mkaishar

join:2000-12-20
united state
Reviews:
·Verizon FiOS
With your 130 Mbps aggregate bandwidth you did not miss the marketing hype by very much. Not too bad for your first UBNT link !
I am not complaining...and you are right the aggregate is close to their marketing claims in a best case scenario

The real world test will be when the fail-over test is done, how smooth that will NOT work initially and then feedback from users with the much slower connection as compared to our expensive but well worth it BridgeWaves.

rconaway8

join:2005-11-10
Phoenix, AZ
reply to mkaishar
Your connection numbers are very good but you can't go by the built in speed test. The radios are faster with AirMax than without, even in a PTP link. Use external computers for that test and try it again. Also use the latest firmware.


superdog
I Need A Drink
Premium,MVM
join:2001-07-13
Lebanon, PA
said by mkaishar:

Conclusion:

1. AirMax does affect performance in a PTP...thanks guys...it will be turned off.
said by rconaway8:

Your connection numbers are very good but you can't go by the built in speed test. The radios are faster with AirMax than without, even in a PTP link. Use external computers for that test and try it again. Also use the latest firmware.
OK, Now I am confused?. The majority of users say you shouldn't run Airmax and now rconaway8 See Profile says you should. Which @!#$%*! one is it, LOL!
--
»www.wavecrazy.net

rconaway8

join:2005-11-10
Phoenix, AZ
reply to mkaishar
I can only go by my own experience, testing, and recommendations from Ubiquiti. Where AirMax is on, the AP won't acknowledge other radios in there area that are not AirMax compatible. There is also a setting in Airmax to prioritize the traffic per radio. Set it to high and then retest using computers and test software on both sides, not the internal test program.


Mad Dawg
Mad Dawg
Premium
join:2006-03-19
reply to superdog
Well rory and i may differ on the airmax yes or no ptp topic but thats ok ..we can still respect each other

My experiance has been that AM on a PTP link actually degrades throughput unless it is a high interfearance area or a long link that you disable ack on

it also could be that he is using RC and I am sticking with B2
--
Best Regards

MD

wolfcreek

join:2003-12-02
Pagosa Springs, CO
reply to mkaishar
I do not have a picture of the results as this was for an emergency so no time for testing but we had to use a pair of M5 rockets to backup a licensed link for maintenance. Distance was 11 miles. We were pushing about 80 mbs for the duration about 3 days. Ping times were about 2ms versus 1 ms of the licensed and most importantly the customers did not notice. I do know this is a mix of traffic from web, voip, etc so many types of traffic real world and we were quite impressed. This was 40 mhz channel with AirMax enabled.

We have left the link in as a backup incase the need ever arises. No they cant do 1 million pps like the licensed link but we rarely see levels over 10,000 pps on this link.

mkaishar

join:2000-12-20
united state
Reviews:
·Verizon FiOS

2 edits
reply to mkaishar
Reran the tests again...

AirMax DISABLED
Signal Strength -48 dBm
Chain 0 / Chain 1: -51/-50 dBm
Noise Floor: -94 dbm
CCQ: 100%
TX/RX Rate: Auto 270Mbps/270Mbps
ACK/Distance: 27 / 0 Miles
TX/RX chain: 2x2
channel/freq: 36/5180 MHz
channel width: auto 40MHz (upper)

antenna gain: 30 dBi
cable loss: 0 dB
output power: 17 dBm
max tx rate: mcs 15 auto
obey regulatory rules

Aggregation: enabled
32 frames / 50000 bytes
Multicast data: allow all

iperf 10 parallel streams (simultaneous from each side)
60 seconds transmit
tcp windows size: 10MB
max segment size: 1500 bytes

w1-w2
[SUM] 0.0-73.3 sec 522 MBytes 59.8 Mbits/sec

w2-w1
[SUM] 0.0-73.7 sec 522 MBytes 59.4 Mbits/sec

--------------------------------------------------------

AirMax Enabled
AirMax Priority: High
Signal Strength -48 dBm
Chain 0 / Chain 1: -51/-50 dBm
Noise Floor: -95 dbm
CCQ: 100%
AirMax Quality: 83%
AirMax Capacity: 58%
TX/RX Rate: Auto 240Mbps/240Mbps
ACK/Distance: 27 / 0 Miles
TX/RX chain: 2x2
channel/freq: 36/5180 MHz
channel width: auto 40MHz (upper)

Aggregation: enabled
32 frames / 50000 bytes
Multicast data: allow all

iperf 10 parallel streams (simultaneous from each side)
60 seconds transmit
tcp windows size: 10MB
max segment size: 1500 bytes

w1-w2
[SUM] 0.0-73.9 sec 533 MBytes 60.5 Mbits/sec

w2-w1
[SUM] 0.0-75.6 sec 546 MBytes 60.6 Mbits/sec

----------------------------------------------------------

OK, so now I am even more confused, tests before show that with AirMAX enabled performance degrades, ran the tests again and there is no performance difference between non-airmax and airmax enabled.

So...which one is better???
At this point I am changing my previous response from disabling airmax on this PTP to just enabling it.

-Mark


WHT

join:2010-03-26
Rosston, TX
kudos:5

1 edit
On my 22 mile beta link that went in April of 2009 with Bullet M5 and 28 dBi grids.
said by UBNT Developer :
Hi WHT,

Just completed tests on the 5.2.1-rc on your links:

AIRMAX-OFF

AP->STA : 13.5 Mbps
STA->AP : 12.9 Mbps

AIRMAX-ON

AP->STA: 20.3 Mbps
STA->AP : 16.2 Mbps

Raw test results:
quote:
Also the behavior with airmax is same as that of the Jan 29th 2010 test I did on your link. It clearly gives you more throughput. I didn't measure latency today but its always been lower with airmax when you have traffic.
quote:
Just tested on 5.2

With Airmax-OFF the data rate goes as low as 39 when transferring data. I see that most of the time the link does not carry any traffic. So when airmax is OFF the CCQ numbers can be mis-leading without traffic.

When Airmax is ON there are still control packets being sent between the link even when there is no traffic.


38632383

join:2009-09-25
Houston, TX

2 edits
reply to mkaishar
said by mkaishar:

So...which one is better???
As rconaway8 mentioned, AirMax will help mitigate some on-channel interference. If you have little or no interference you will not see much difference between AirMAX on/off. In light of what WHT posted about the AirMAX control packets during idle times, I would leave AirMAX disabled until tests show it with an advantage for your link.

With the signal strengths you have you might want to turn off the auto link rate selection, and lock the link rate to 300X300. I see that several of your tests were made at slower link rates than 300X300 which is what you should have. With the strong -48 dB signal you have, the last thing you need is to have the radio dynamically shifting air link rates.

mkaishar

join:2000-12-20
united state
Reviews:
·Verizon FiOS

1 edit
said by 38632383:

With the signal strengths you have you might want to turn off the auto link rate selection, and lock the link rate to 300X300. I see that several of your tests were made at slower link rates than 300X300 which is what you should have. With the strong -48 dB signal you have, the last thing you need is to have the radio dynamically shifting air link rates.
When I disable auto link rate then ping increases from 1ms-2ms and bounces as high as 200ms, some packets are even dropped too, with or without airmax


38632383

join:2009-09-25
Houston, TX
said by mkaishar:

When I disable auto link rate then ping increases from 1ms-2ms and bounces as high as 200ms, some packets are even dropped too, with or without airmax
Sounds like you must have some interference you didn't know about. Might want to try different channels.


WHT

join:2010-03-26
Rosston, TX
kudos:5
reply to mkaishar
The more I jump into 2.4 WISP projects with FW 5.3, I'm coming to the conclusion one shoe does not fit all.

AirMax MIMO can be a magic bullet *in_some_applications* with marginal LOS /NLOS or noisy conditions, but understand it will usually rely on only one chain to at least get a resellable signal.

AirMax 5.3 has its own unique advantage that helps even more in a crowded band. It doesn't make the noise disappear, but at least you can have a working system.

rconaway8

join:2005-11-10
Phoenix, AZ
reply to mkaishar
Did I mention that running 40MHz channels is almost a waste of time. The processor can easily max out under load with 20MHz channels. Unless you need that last 5%, I would run 20MHz channels. That would also help with your interference issue.


38632383

join:2009-09-25
Houston, TX

1 edit
said by rconaway8:

Unless you need that last 5%, I would run 20MHz channels. That would also help with your interference issue.
5% data delivery difference between 40 Mhz channels and 20 Mhz channels ?

How are you measuring the CPU load on these 400 Mhz CPUs ? Is there a meter in the GUI that I missed or some method by telnet ?

rconaway8

join:2005-11-10
Phoenix, AZ

1 edit
You can watch it in telnet. Certain types of traffic will push the processors pretty hard. FTP transfers, video traffic, etc... can run the processor at 65% or more on a one way transfer with 20MHz channels. Push it to 40MHz and SNMP data starts getting delayed, latency goes up, and the processor will peg out at 100% easily. Remember the 19,000pps limit. This isn't unique to Ubiquiti as most vendors have similar issues. Rarely does a 40MHz channel produce double the throughput over 20MHz for that reason.