Max TCP speed for 5 GHz point to point link with turbo

[FlemmingFrandsen]

I'm thinking about re-doing my backhaul using powerstation 5's, what is the maximum real world speed I can expect with and without turbo (20 vs. 40 MHz)?

IOW: Does the CPU max out before the radio?

Does the firmware restrict on which channels turbo mode can be used on, if so, which channels are available?

[UBNT-Mike.Ford]

Hello Flemming,

1. The max throughput is 25Mbps TCP/IP in 20mhz and 32Mbps in 40Mhz.

2. This is CPU limited.

I will post a pic of the channel list for you, but for witch country?

Thanks,

Mike

[Nasp]

I have one link, 800m with LOS:

1 PS2 ( AP ) >>>> 1 NS2 ( Client ) - 2.0 a 2.2m/bs on FTP.

Very near of the limit 25Mbps, but very far away 54mbps Wireless G.

[FlemmingFrandsen]

I have one link, 800m with LOS:

1 PS2 ( AP ) >>>> 1 NS2 ( Client ) - 2.0 a 2.2m/bs on FTP.

Very near of the limit 25Mbps, but very far away 54mbps Wireless G.

For some reason the overhead in 802.11 means that the 54 Mb/s signaling rate comes out to a payload rate of about 25 Mb/s.

The interesting bit is that with 40 MHz channels it ought to be possible to hit 50 Mb/s, but the Integrated Ubiquity products are based on a weedy 180 MHz MIPS CPU that can't push bits around that fast.

[FlemmingFrandsen]

Hello Flemming,

1. The max throughput is 25Mbps TCP/IP in 20mhz and 32Mbps in 40Mhz.

2. This is CPU limited.

I will post a pic of the channel list for you, but for witch country?

Thanks,

Mike

Are there any plans for NS or PS products with beefier CPUs?

I'm in Denmark, but I don't mind setting the country value to Elbonia if that let's me use the channels that local law allows in addition to the ones defined by ETSI.

[UBNT-Mike.Ford]

I've tried this in a "bench" testing environment and I could never get the PS5 units to push more than 15-ish meg.

This was using 20Mhz channels and trying every combination of throughput enhancements (fastframes etc) and ack timing adjustments and I just couldn't push more than that.

Also another question, I was using an external tester testing bandwidth through the radios, not using the radio test tools themselves. I am curious also as to how well the internal test tools work, do they cause some CPU overhead as well when running them or are they a pretty good judgment of the speed?

Perhaps there was something I was missing in the setup but I can't understand what it was. CCQ was at 100% the entire time and noise floor was perfect, signals weren't too hot, I just couldn't get it to go any faster.

This sorta put a damper on it because I normally can squeeze 24meg out of a Mikrotik with XR5 cards, but I would much rather have a single integrated unit like the PS5s as I'm not worrying about coax connectors or other external factors that can go wrong. Any ideas of what I'm doing to not make this work as it should?

Hey Kewl,

The internal speed test is only good for about 15-16 Mbps due to the CPU overhead involved in running the test. For testing purposes I would recommend using a program such as NetIQ Chariot.

NASP,

You can never achieve a true 54Mbps data rates due to wireless overhead involved. Most of the time the actual throughput is about half of the over the air data rate. This applies to ALL radios, not just ours.

Thanks,

Mike

[CDX825]

Only way to go more is to go full duplex.

[UBNT-Mike.Ford]

Only way to go more is to go full duplex.

Or use our SR71 radios and get 140Mbps TCP/IP

Mike

[DrNutbush]

Which board/software will get us such speeds with SR71 Mike?

[FlemmingFrandsen]

Only way to go more is to go full duplex.

Or use our SR71 radios and get 140Mbps TCP/IP

Mike

I'd be quite interested in moving our backhaul to SR71-5 radios on routerstation boards once we outgrow our current 25 Mb/s link, but is the routerstation CPU beefy enough to support the bandwidth?

How about the antennas for the 802.11n point-to-point link, is it just a matter of placing two differently polarized antennas, one horizontal, one vertical?

[WHT]

How about the antennas for the 802.11n point-to-point link, is it just a matter of placing two differently polarized antennas, one horizontal, one vertical?
Nope...You're trying to use "N" for something it was never designed to do.

I can't find the posts I've made addressing the flaw in that logic, so I'll briefly cover the points.

802.11n uses a modified implementation of spacing diversity. Traditional space diversity (which was well covered in another post) requires spacing on the order of 20 to 30 feet for WiFi. Type N uses a very close spacing and relies on phase diversity where phase difference is detected by slightly offset antennas. That slight offset when used for an AP covering several miles would have to be on the order of tens of feet (that also was covered in another post).

Type N would work, in theory, for what you are trying to do if you spaced your antenna several tens of feet apart and with the same polarity. But in practice, theory often falls apart.

[FlemmingFrandsen]

Type N would work, in theory, for what you are trying to do if you spaced your antenna several tens of feet apart and with the same polarity. But in practice, theory often falls apart.

Right, that was what I gathered from various sources as well, but I've also seen references to using 802.11n for point-to-point links.

[WHT]

Type "N" is advertised as faster speed and longer ranges. But they don't tell you those are mutually exclusive...choose one or the other, but not both. Nor do they tell you its effective up to a few tens of feet.

Since MIMO technology avails itself on alternative signal paths. One path might be poor and another better, so it grabs the data from the best path.

While you may have seen references to using it, have you seen field tested before and after comparisons? I'll bet you see something like, "We tested a high priced XZY brand of antenna that promises amazing performance and it works great!"...never mind the fact a more traditional (and lower cost) antenna would have preformed the same

For MIMO to work, there has to be a significant difference in the signal path. This is similar to antenna spacing diversity. If you are shooting twenty miles, towards two antennas spaced only a few feet apart - both antennas are pretty much seeing the same path. But if you space them thirty feet apart, then one will see a different path then the other.

[FlemmingFrandsen]

If you are shooting twenty miles, towards two antennas spaced only a few feet apart - both antennas are pretty much seeing the same path. But if you space them thirty feet apart, then one will see a different path then the other.

Yeah, I guess the only way we're going to see any information about 802.11n for point-to-point links is to do the experiments ourselves once the hardware is available...

[rconaway]

So what Ubiquiti product using a 40MHz channel and 5GHz is the fastest in the Ubiquti line during an ftp transfer?

I have stats on a lot of products doing this and not to brag but we got 26.1 Mbps out of an NS2 once as long as you only do ftp transfers for everything. In reality, about 25 Mbps was about average for an ftp transfer and we were running SNMP. SNMP totall choked at 40MHz though when the radio hit 33Mbps.

As a comparison, Airstream hit 44Mbps in that same environment and they cost about $4K retail.

[UBNT-Mike.Ford]

Hey Guys,

I have personally been in the field and witnessed the speeds. We will post some whitepapers with the theory, instructions ect on how to achieve these speeds and the way around the multi path problem with 802.11n

Thanks,

Mike

[Ron]

I've tried this in a "bench" testing environment and I could never get the PS5 units to push more than 15-ish meg.

Same here, 15-16 Mbps is the TCP max of the PSx AFAIK.
I think those who claim 25 Mbps are doing UDP tests or some atypical traffic.

[UBNT-Mike.Ford]

I've tried this in a "bench" testing environment and I could never get the PS5 units to push more than 15-ish meg.

Same here, 15-16 Mbps is the TCP max of the PSx AFAIK.
I think those who claim 25 Mbps are doing UDP tests or some atypical traffic.

Hello Ron,

Are you using the internal speed test ? Do you have SuperA features enabled?

Thanks,

Mike

[rconaway]

Ron, we compared all our numbers with real-time SNMP readings (except in turbo mode where SNMP was not updated real-time due to processor overhead) and they were all extremely accurate. I do a lot of UDP traffic with video cameras but you are kind of wrong on this one. I'm betting installations of several hundred thousand dollars and Homeland Security projects that I'm right so I really don't have the luxury of being wrong. As they say in the movies, "If I'm wrong, people die". Kind of an exaggeration at this point but who knows.

We use equipment from many manufacturers so we can smell B.S. a mile away. We have called many manufacturers to the carpet with let's just say, optimistic marketing material that doesn't really tell the whole story. The 2's transfer about 20Mbps and the 5's will do 25Mbps in non-turbo and 33-34Mbps in tubo. We think there was a slight amount left because we were running SNMP and had extra reporting turned on but I don't have time ro rerun the tests. These were optimal numbers based on the radios in the same room also but we don't expect any less performance in the field assuming we set them up correctly.