Single Channel Architecture - 802.11n/ac in 5GHz

[sfx2000]

Just thinking about 11ac and the limitations of channel management - not just in the enterprise, but also in the consumer space.

One of the challenges in 5GHz with 11ac is that we basically have 2 wide 80MHz channels - yes there is the middle band, but that can be a challenge as well, and there's certain limits in the upper bands depending on regions - we're in NA, so the lower and upper channels are pretty much clear, with DFS/TPC being in the middle.

That being said, when looking at adjacent networks, what are your thoughts and experiences?

 

[coxhaus]

I think 11ac 80MHz may be slow to take over in the enterprise area because who wants to allocate that much bandwidth to wireless devices. Especially if there are lots of them. There may be key areas which 11ac are used in but I bet most will stay with a more 802.11 restricted mode. You can handle more devices with less strain on the network.

My thinking this is more aimed at a consumer market.

 

[sfx2000]

This was somewhat of a loaded question

At the office, we did a small study to evaluate this very problem in the enterprise space - prior to 11ac, we had a very well designed 11n network, but with AP's that were end-of-life, and approaching end-of-support, so it was time to review options. Knowing that we had the reuse limitations, we knew that multi-channel was going to be an issue, so we set up enterprise class AP's (vendors "A", as incumbent, and vendor "C" as a possible change). Also should note that these are managed AP's with a dedicated controller.

All AP's have common SSID (two SSID's - CorpWIFI-test, and CorpGUEST-test) on both bands, running narrow channels in 2.4 (11n), and 80Mhz channels in 5GHz (as 11ac). In 20Mhz, F3 reuse pattern (as per "best practices"), and 11ac, F2 reuse initially, and then the F1 reuse test case...

F2 reuse looked good, pretty much as expected - we cannot use the DFS channels as luck would have it, we're in a place where DFS gets kicked quite often, and the attendant drops and timers kick in - so we basically can't use those channels..

So, after discussion with the vendors on the channel issue, they both suggested looking at an F1 scheme for 5Ghz - there wasn't much consensus around the table, but we agreed to run a limited test - using 3 AP's, and a good mix of clients (laptops, handsets, tablets, mix of 11n and 11ac, with a couple of older Cisco 11a cards dropped into the mix).

The results were a bit odd, and some folks were surprised - the AP's started coordinating with each other to ensure that all STA's received appropriate airtime on the channel.

Going back to the bench, and breaking out the test boxes, including some very expensive 802.11 tracetools, we started digging into what was going on - and confirmed that as part of an extended BSS set, there was coordination happening as the BSS, and adjacent BSS nodes, and this was happening at the MAC/PHY layers... in other words, at the chipset level, below the vendors operating system in the AP's...

Surprisingly enough, both vendors performed pretty much the same - both "A" and "C" - so having this info, we did a small trial deployment - two suites on one end of the building, and turned up the debug with real-world users. Suite 1 had Vendor A's gear, Suite 2 Vendor C...

And everything was fine.. we did eventually deploy into the entire building - 5 floors, and alternated into an F2 scheme of sorts - everything on the single floor was deployed as F1, and then we alternated high-low from floor to floor...

FWIW - we did end up sticking with the incumbent vendor, but it was a good change to see what the competition had to offer...

 

[sfx2000]

But wait - there's more... again, we felt that there was still some "special sauce" going on, so a couple of the engineers went down to the local BestBuy, and picked up some consumer AP's (vendors A, A1, N, L, and T) - it was like Noah's Ark, two of each - with the intention of getting different chipset vendors - ended up with Broadcom, Atheros, and Marvell chipsets across different models. Mostly to see if this was something at the chipset level...

Ran the same test - and guess what - same results - with single and cross-chipset vendors...

As long as the AP's are part of the same ESS, and common SSID, the BSS's did coordinate to share airtime - breaking up the ESS and going with unique SSID's per AP, then things fell apart, mostly due to after analysis, overlapping BSS interference..

The takeaway here - and we need to go into some high-level basics...

SSID - this is the network name - if all AP's have the same SSID, then they're sharing the distribution network (e.g. the Ethernet LAN, with common routing, IP subnets, DHCP, etc...)

BSSID - this is the MAC address of the AP

BSS - this is the clients associated with a single AP, including that AP

ESS - this is the extended service set - and includes all BSS's within the Extended Set, all sharing a common SSID, and common distribution network.

Again, the takeaway - with modern AP's, single frequency can work... best practices with common SSID's are still valid in an ESS, and that 11ac is remarkably tolerant of mix-mode 11a/11n/11ac.

Can't really go into much more detail, as we're actually writing a whitepaper on our finding, ans while it's pretty much done, since it's on corporate time and letterhead, it's pending legal and business review.

It was a fun project, one of those "20 percent" efforts outside of our normal day jobs - having some pretty sophisticated test and analysis tools is what helped confirm most of the findings...

Now, I'm not saying everyone needs to jump to a F1 plan, but it's worthwhile to see for yourself - might be an option to consider...

 

[thiggins]

Thanks for sharing that, sfx. Wonder if MediaTek and Realtek would also behave nicely.

 

[sfx2000]

I would suspect that MediaTek and Realtek probably would be similar...

 

[sfx2000]

I think 11ac 80MHz may be slow to take over in the enterprise area because who wants to allocate that much bandwidth to wireless devices. Especially if there are lots of them. There may be key areas which 11ac are used in but I bet most will stay with a more 802.11 restricted mode. You can handle more devices with less strain on the network.

My thinking this is more aimed at a consumer market.

In the enterprise space, one consideration is to deploy 802.11ac in VHT40 mode, if single frequency isn't practical with VHT80 - the upside to deploying VHT40 is that one can use the same frequency plan as HT40 in 802.11n - and one still gets the benefits of 11ac, just not as much...

Should also note - during our 11ac testing, pretty much confirmed Tim's article here on SNB about 11n clients seeing improved performance - esp. as our older 11n AP's were N300 in 5Ghz, whereas with the new AP's, the N clients see the AP as an N450 class (three radio's/3 streams)...

 

[coxhaus]

Neat test. Do you have a good link for the F1 scheme? How was ESS implemented?

 

[sfx2000]

Neat test. Do you have a good link for the F1 scheme? How was ESS implemented?

You'll have to wait for the White Paper

But putting things into Lay Terms - AP's do check for other BSS - either to coordinate, or to use protect in the case of the BSS not being a peer... so multiple AP's will self-organize into an lightweight ESS set if their all in the same SSID (network name)...

If each AP uses a unique SSID - eg "Thelma" for AP1, and "Louise" for AP2, then Thelma and Lousie mark each other as OBSS, and implement protection..

 

[coxhaus]

What's your target date for the white paper?

 

[sfx2000]

What's your target date for the white paper?

Don't know - it's in the business/legal layer... part of my frustration is that it's good info to share... it might die, so what you have above is what might remain..

There are a number of vendor pubs that say this should work - our efforts were to prove it..

sfx