Two Radios Are Better Than One (Unless They're Both 5 GHz)
If there's one thing we've learned about WiFi over the years, it's that problems get fixed. Site surveys used to be sooo annoying. So we got controllers with auto-RF. Guests used to complain and complain that we didn't have WiFi for them. So we got captive portals. And so the worl--
Wait a minute. Auto-RF really doesn't work and we still need site surveys. And captive portals annoy our friends in desktop support as much as they annoy our guests, plus they drag down overall WiFi performance. Hmmm... Let's start over:
If there's one thing that we've learned about WiFi over the years, it's that sometimes what seems like a good solution to a problem only ends up making things worse.
Which brings us to our latest solution-that-only-makes-things-worse: Dual-radio APs with band-selectable radios.
A little background:
Last week, I was sitting in an office, minding my own business when I got a call. A friend of mine who works in WiFi had a question for me: "What do you know about these band-selectable APs from Xirrus?" I answered that I hadn't been paying close attention to Xirrus's product line, and thus I was unaware that such APs even existed.
My friend was quite negative on the idea of band-selectable APs. He was worried about interference. He is a big RF guy (probably one of the few WiFi folks I know who is more of an RF person and I am, in fact) and he just did not buy the idea that two radios in the same AP (or, "box" as he called it) could transmit or receive at the same time in the same frequency band.
I, being the staunch contrarian that I am, was more sympathetic to the idea of band-selectable AP radios. I was trying to imagine a real world situation where band-selectable radios would help. I thought of convention halls and football stadiums; places where such a high number of users congregate in one area that you might need as many 5 GHz radios as possible. If a space -- a convention hall, for example -- had six band-static APs (that's what I'll call APs that one radio each for the 2.4 GHz and the 5 GHz bands) installed, then three of the 2.4 GHz radios should be disabled. That would make for nine total AP radios to service the high-density space. With band-selectable APs, all twelve of the AP radios in the six installed APs would be available. The wireless LAN could be installed with three traditional 2.4/5 GHz APs and three non-traditional dual 5 GHz APs.
I even mapped out what such an installation might look like:
"See," I said to my friend. "I would be able to space out the channels when setting both radios to 5 GHz on the same AP. It would be better to have both AP radios enabled, even if interference makes it so that a two-radio AP does not give two-times the performance of a single radio AP." And the throughput tests (and, yes, I hate throughput tests but I also know that other people want to know what they say) looked better (approximately 90 Mbps of total single-stream, 802.11n throughput with two 5 GHz radios enabled on a single AP, compared to around 70 Mbps when only one 5 GHz radio was enabled). Plus there's the extra associations (typically enterprise APs have per-radio association limits) when both AP radios are enabled.
And that's not even touching on the real world interference vs. theoretical interference. "Sure," I said, "in theory two radios in the same box on the same band would cause interference. But look at this:"
That's a screenshot of Chanalyzer taken during my testing. I placed my WiSpy less than three feet from the Xirrus 620 access point, and there was essentially no sign of interference on my Chanalyzer screen.
So, there I had it. There are real-world cases where band-selectable APs are better than band-static APs.
"Check the Retries."
...or are there?
That little question (and, in the interest of full disclosure, I must admit that my friend was not present while I tested. "Check the Retries" was part of my internal monologue) and the path it led me on made me change my entire opinion.
As long as there was only one AP radio on the 5 GHz band, the AP performed well. Uplink retry rates (counting bytes retried, not packets retried) were around 3% and downlink retries were around 16% (the lower uplink Retry percentage is because I was testing using an iPad, and iOS devices use RTS/CTS to improve performance in congested areas). Those numbers stayed consistent between tests where one AP radio was disabled and where one AP radio was using the 2.4 GHz band.
When both AP radios were configured for the 5 GHz band, the AP melted down. It didn't literally melt down, of course, and even the throughput numbers didn't melt down (as mentioned earlier, throughput did increase by almost 30% when a second 5 GHz radio was enabled on the AP). But the Retry percentages jumped so dramatically (to around 20% on the uplink and 34% on the downlink) that a typical ENTERPRISE WIRELESS NETWORK (which, honestly, is what I care about. I couldn't care less what Chariot tells me in a lab) would face CERTAIN MELTDOWN if it became densely populated with users.
In fact, the jumps in Retry percentages were so dramatic that I will go out on a limb and say that in the vast majority of enterprise wireless network scenarios (again, the stuff that I care about) the WiFi will be better if the second AP radio is DISABLED than if the second AP radio is set to the same 5 GHz band as the first AP radio.
In graphical terms, this:
will work better than the other Ben's Convention Hall idea that was illustrated earlier on.
I have one final thought after experimenting with the Xirrus 620 AP: We need to stop these things.
Many of us know and remember the damage that previous well-intended, wrongheaded WiFi solutions caused. We know how many problems were caused when people started accepting auto-RF in lieu of a site survey. We are still dealing with the legions of poor-performing wireless networks that have been hamstrung by the captive portal. Let's try to nip this in the bud. Band-selectable, dual-radio APs will look good in a throughput test and they'll be tempting to folks who have yet to learn about WiFi. But we should know better. We should know that, in this case, one AP radio is better than two.
Wait a minute. Auto-RF really doesn't work and we still need site surveys. And captive portals annoy our friends in desktop support as much as they annoy our guests, plus they drag down overall WiFi performance. Hmmm... Let's start over:
If there's one thing that we've learned about WiFi over the years, it's that sometimes what seems like a good solution to a problem only ends up making things worse.
Which brings us to our latest solution-that-only-makes-things-worse: Dual-radio APs with band-selectable radios.
A little background:
Last week, I was sitting in an office, minding my own business when I got a call. A friend of mine who works in WiFi had a question for me: "What do you know about these band-selectable APs from Xirrus?" I answered that I hadn't been paying close attention to Xirrus's product line, and thus I was unaware that such APs even existed.
My friend was quite negative on the idea of band-selectable APs. He was worried about interference. He is a big RF guy (probably one of the few WiFi folks I know who is more of an RF person and I am, in fact) and he just did not buy the idea that two radios in the same AP (or, "box" as he called it) could transmit or receive at the same time in the same frequency band.
I, being the staunch contrarian that I am, was more sympathetic to the idea of band-selectable AP radios. I was trying to imagine a real world situation where band-selectable radios would help. I thought of convention halls and football stadiums; places where such a high number of users congregate in one area that you might need as many 5 GHz radios as possible. If a space -- a convention hall, for example -- had six band-static APs (that's what I'll call APs that one radio each for the 2.4 GHz and the 5 GHz bands) installed, then three of the 2.4 GHz radios should be disabled. That would make for nine total AP radios to service the high-density space. With band-selectable APs, all twelve of the AP radios in the six installed APs would be available. The wireless LAN could be installed with three traditional 2.4/5 GHz APs and three non-traditional dual 5 GHz APs.
I even mapped out what such an installation might look like:
"See," I said to my friend. "I would be able to space out the channels when setting both radios to 5 GHz on the same AP. It would be better to have both AP radios enabled, even if interference makes it so that a two-radio AP does not give two-times the performance of a single radio AP." And the throughput tests (and, yes, I hate throughput tests but I also know that other people want to know what they say) looked better (approximately 90 Mbps of total single-stream, 802.11n throughput with two 5 GHz radios enabled on a single AP, compared to around 70 Mbps when only one 5 GHz radio was enabled). Plus there's the extra associations (typically enterprise APs have per-radio association limits) when both AP radios are enabled.
And that's not even touching on the real world interference vs. theoretical interference. "Sure," I said, "in theory two radios in the same box on the same band would cause interference. But look at this:"
That's a screenshot of Chanalyzer taken during my testing. I placed my WiSpy less than three feet from the Xirrus 620 access point, and there was essentially no sign of interference on my Chanalyzer screen.
So, there I had it. There are real-world cases where band-selectable APs are better than band-static APs.
"Check the Retries."
...or are there?
That little question (and, in the interest of full disclosure, I must admit that my friend was not present while I tested. "Check the Retries" was part of my internal monologue) and the path it led me on made me change my entire opinion.
As long as there was only one AP radio on the 5 GHz band, the AP performed well. Uplink retry rates (counting bytes retried, not packets retried) were around 3% and downlink retries were around 16% (the lower uplink Retry percentage is because I was testing using an iPad, and iOS devices use RTS/CTS to improve performance in congested areas). Those numbers stayed consistent between tests where one AP radio was disabled and where one AP radio was using the 2.4 GHz band.
When both AP radios were configured for the 5 GHz band, the AP melted down. It didn't literally melt down, of course, and even the throughput numbers didn't melt down (as mentioned earlier, throughput did increase by almost 30% when a second 5 GHz radio was enabled on the AP). But the Retry percentages jumped so dramatically (to around 20% on the uplink and 34% on the downlink) that a typical ENTERPRISE WIRELESS NETWORK (which, honestly, is what I care about. I couldn't care less what Chariot tells me in a lab) would face CERTAIN MELTDOWN if it became densely populated with users.
In fact, the jumps in Retry percentages were so dramatic that I will go out on a limb and say that in the vast majority of enterprise wireless network scenarios (again, the stuff that I care about) the WiFi will be better if the second AP radio is DISABLED than if the second AP radio is set to the same 5 GHz band as the first AP radio.
In graphical terms, this:
will work better than the other Ben's Convention Hall idea that was illustrated earlier on.
I have one final thought after experimenting with the Xirrus 620 AP: We need to stop these things.
Many of us know and remember the damage that previous well-intended, wrongheaded WiFi solutions caused. We know how many problems were caused when people started accepting auto-RF in lieu of a site survey. We are still dealing with the legions of poor-performing wireless networks that have been hamstrung by the captive portal. Let's try to nip this in the bud. Band-selectable, dual-radio APs will look good in a throughput test and they'll be tempting to folks who have yet to learn about WiFi. But we should know better. We should know that, in this case, one AP radio is better than two.
***
If you like my blog, you can support it by shopping through my Amazon link or donating Bitcoin to 1N8m1o9phSkFXpa9VUrMVHx4LJWfratseU
Hi Ben, I recently strayed into the strange and scary world of WiFi from LTE while doing some studies on unlicensed LTE (yes, some strange & scary stuff). I have done some studies on WiFi and your findings puzzle me somewhat. If WiFi is CSMA/AC with RTS/CTS and the two 5GHz radios are clearly in different frequencies as shown in spectrum analyzer, what caused the spike in retries? This is something that may come when LTE-U/LAA goes into 5GHz and we have a dual band WiFi in the "same box" (without disabling 5GHz WiFi).
ReplyDeleteSpike in retries is because even though the spec an antenna can't detect interference from 3 feet away, there is actually massive interference INSIDE the AP.
ReplyDeleteA dual radio, LTE-U/WiFi (5 GHz) AP is also going to be a disaster, unless the frequencies are hard coded into the AP. The only way it's going to work well in a high density environment (which is the whole point of LTE/U, right?) is if you can somehow get an AP maker to manufacture dozens of different AP models, all with hard coded frequency assignments for the two radios.
Interesting read.. During a design session with my Wi-Fi survey/design provider (LEVER here in the UK) we quickly came to a conclusion that we really will be wasting money on redundant 2.4GHz radios as we move into an 11ac dominated world. As such, I took away the action to start looking to see if vendors would do APs with 2 software configurable radios to do just as you've described; allowing me to turn redundant 2.4GHz radios into extra-capacity-providing 5GHz radios.
ReplyDeleteBased on your findings I hope that instead, we see a selection of single band APs re-emerge.
Meru has had band selectable radios for a while, but you are only supposed to use both radios on the same band if you have external antennas and they are separated by a recommended 6 feet. Would this same recommendations allow Xirrus' solution to perform better on dual 5GHz operation?
ReplyDeleteGustavo: 6 feet might be enough. I suppose I would have to see it in action to make any firm recommendations.
ReplyDeleteJon: Single band AP radios are definitely better from what I've seen.
I see cisco 3800 is similar to the products mentioned with band selectable radios. They tout a dual 5 GHz AP supporting crazy (2.6Gbps *2 = 5.2 Gbps) metrics for "over-the-air speed". I think these metrics are not sustainable in a common business environment because the channel width required but are being used to drive switch port sales. I would be interested in your thoughts on this.
ReplyDelete...and now there is the Aerohive AP-250. I'm curious if it suffers from the same "meltdown" results?
ReplyDeleteI'll be surprised if the Cisco 3800 and Aerohive 250 don't melt down under stress.
DeleteTwo separate transceivers close to each other or dual integrated devices could work, but they need to be designed well. You can't make a blanket statement that it won't work at all. The problem usually lies with the device's receiver front end. Even if the signals are far enough apart from each other in the band, a receiver with insufficient filtering and/or dynamic range can suffer desense from a nearby strong signal, sometimes completely outside of the intended band. Well designed receivers can effectively reject nearby signals. Less than stellar designs suffer.
ReplyDeleteMultiband devices very often have inferior filtering on the receiver because they must accept signals from a much wider frequency range. Oscillators are reasonably easy to make frequency agile, but RF filtering is much harder. The designer of a multi-band receiver / transceiver essentially has three options. Build switchable, discreet filtering for each band of operation ( complicated, physically cumbersome), build frequency agile, adjustable filtering (difficult to engineer this kind of filter to the same performance as discreet filtering and considerably more expensive) or just build one big bandpass that accepts all frequency ranges that will be used (cheap, easy to design, but doesn't reject unwanted signals inside that big bandpass range).
ReplyDelete