Like many other wireless architects and admins are probably doing right now, I spend a fair amount of thinking time these days on matters surrounding the fast-coming 802.11ac wireless standard. To the WLAN Industry's credit, I don't think the 11ac hype machine has been quite as foolish as was 11n's a few years back, but there is still a lot to sort out here. I recently wrote about the challenge of knowing when to get into the 11ac game, depending on your budget and sense of adventure, but recent conversations with a few industry bigwigs have me scratching my head even more.
Two points to consider beyond just when you move to the next great thing in WLAN:
- Many of us have invested hundreds or thousands of cable runs in our current wireless networks, each supporting up to 1 Gbps. Where asbestos abatement, new conduit and cable tray, and building permits are required, running cable for an AP can often cost more than the AP itself did- even for market leading Cadillac-grade access points. How drastic does our "investment" in our installed cable change with 11ac? Surely, a single 1 Gig uplink won't be enough for an 11ac AP.
- There is no free lunch for any technology. To get to the promised data rates of 6.9 Gbps for the second-wave of 11ac, we'll need to use crazy wide channels and reduce the 5 GHz spectrum to potentially the same state of overuse as 2.4 GHz suffers now (even if the FCC burps up more spectrum). And if your ultra-wide 11ac channels happen to be physically near my legacy 11a/n deployment, you'll probably get booted from my list of fake friends on Facebook.
Let's talk about each a bit more.
On the cable thing, if you are familiar with the likes of the TIA-568 cable standards and know folks that run cable for a living, you know that an installed UTP run is a "component". It's the bedrock of the OSI or 5-layer model, and it gets installed methodically and to exacting standards (and often at great expense).
If you had the foresight to provision a robust cable design as part of your 11a/g dense deployment back in the day, you likely didn't have to do much with cable when it came time for 11n. You were able to leverage your investment, and replace an old AP with a new one for the most part, because both a/g and 11n do well on Cat 5E, 6, and 6a cable and Gig uplinks.
Alas, unless we're all being sold snake-oil by the WLAN industry, dual band-11ac APs will need more than 1 Gbps of connectivity (we'll save the related PoE discussion for another time). But how much will we need? A 2 Gbps Ether-channel? A 10 Gbps uplink? Either way, we should be concerned. Most of us can't afford to rebuild our wireless cable plants from scratch. To me, the lofty claims of what 11ac will be able to deliver point to 10 Gbps uplinks, but I'm hearing from a growing cadre of industry voices that the expectation is two 1 Gbps will do OK. Please, someone explain the math behind this, if what we're hearing from 11ac marketers isn't a recipe made of 2 parts truth and 3 parts pure county horse manure.
Regardless of whether we have to run an additional cable and burn an extra switch port per 11ac access point, or get into some sort of 10 Gig PHY that likely doesn't yet exist, the physical layer aspects of second-wave 11ac need to be watched as closely as anything.
On the wide-channel thing: again, we're being told to expect to use 160 MHz channels to get to data rates of 6.9 GHz. Yes, we don't HAVE to use them, and can settle on 80 or 40 MHz channels and lesser data rates. But if we buy into the promise of 11ac, we need to be ready to pull the trigger on 160 at some point. And where we do, each one of our wide channels will be quite unfriendly to any of our 5 GHz neighbors' APs that run in 5 GHz using 20 or 40 MHz channels from 11a or 11n deployments. In multi-tenant environments where lots of companies can see each others' signals strongly, there is a lot to contemplate here. Will we be good neighbors?
Takeaways: I don't have answers. But I do know these issues have to be considered along with all of the huge promises of performance gain that 11ac is supposed to provide that are being bandied about. Between the Wave 1 versus Wave 2 quandaries, and the conversational recklessness of the highest-end claims of 11ac's capabilities that are being spit out by marketers, never have we had to be more on our guards about what a new wireless technology will REALLY mean for large WLANs.
I've actually been pondering these same issues, especially the uplink question, and I think I have an explanation of where the vendors think that a pair of 1Gpbs will be adequate.
ReplyDeleteThere are two sizing factors involved. The first is efficiency. With a 1Gbps link, you can expect to get 95% or better throughput on the link (assuming the machines at each end are capable), whereas with wireless, this is typically far lower. As more clients come on, contention and retransmissions cut into bandwidth. Let's be generous, and say that after the airtime is all sorted out, the uplink only ends up having to handle 50% of the maximum wifi capacity.
Second is the simple fact that the 1Gbps links are full duplex - 1Gbps up, 1Gbps down - while the wifi capacity is shared between both directions. Assuming you have a fairly even up and down spread (optimistic, I know) this means you have to apply another 50% factor when comparing marketing sheets.
So out of that theoretical max 7Gbps of wifi bandwidth, it's likely that only 25%, or 1.75Gbps, will have to get handled by the uplink, easily handled by a 2Gbps trunk. Knock yourself down to 80MHz or 40MHz channels, and you've got plenty of headroom. This isn't anything I've officially heard from the vendors, but it's what the tea leaves say to me today.
You should really read Matthew Gast's book on 802.11ac.(http://shop.oreilly.com/product/0636920027768.do) It goes into detail on both the question of physical links and how the increased bandwidth interference problem will be handled. He specifically says that for the first wave, gigabit ethernet is enough, but that dual-radio, 4 stream APs will require 10-GB ethernet. Also in the presence of multiple people using 160MHz channels, the AP will downshift on a frame by frame basis to make sure there isn't interference. Lots of good other information in this too that goes way beyond marketing and gives you the practical details you need to know.
ReplyDeleteExcellent, Chris. Looks like time for a new read for many of us, me included!
ReplyDeleteFrank, though I follow and could say it sounds plausible, your calculations (and thanks for sharing) don't line up with Gast's new 11ac book. strangely, we're not hearing much from vendors on clear physical layer future for W2ac.
ReplyDeleteThanks - I actually just got a copy of his book, but haven't had time to do more than skim though it. Given the cost incurred of rolling out that density of 10Gbit (there aren't even that many 10GbaseT switches on the market yet, just SPF+), as well as the difficulties in getting 10Gbit certified cabling, my bet would be that the vendors just oversubscribe and do their best to stick with dual 1Gbit for a while. I'd certainly love to be proved wrong, though ;-)
ReplyDeleteFrank beat me to the answers I'd type in. Ethernet is full-duplex, and 802.11 is half-duplex. Plus, there's the question of client mix. With so many clients being single-stream devices (phones/tablets/etc), you won't be using the full multi-stream speed of 11ac because a significant portion of the traffic will be using single-stream rates. Even with 256-QAM, those don't get significantly faster than 11n. In the first wave of 11ac products, the peak rate is "only" 1.3 Gbps, but 802.11 speeds are always given as peak PHY rate and don't include overhead. (How much did you get out of your "54 Mbps" 802.11a/g network?)
ReplyDeleteAnd as Frank points out, 802.11ac selects a channel width for transmission on a frame-by-frame basis, so a network adjacent to yours that wants to use 80 MHz or 160 MHz frames will only do so when your network isn't transmitting. There's a small book excerpt on dynamic per-frame bandwidth here: http://blogs.aerohive.com/blog/the-wi-fi-security-blog/80211ac-channel-bandwidth-sharing
The characterization of second wave 11ac as 7 Gbps is probably a bit optimistic. My guess on second wave 11ac is 4-stream clients, possibly supporting 160 MHz channels. That gives you 3.5 Gbps, but only when you're using a 4-stream client. If you have a significant fraction of single-stream mobile devices, you'll place way less demand on the backhaul. As much as I'd love to see an 8-stream device (required for 7 Gbps) in the second wave, I won't hold my breath.
Now we're getting to the good stuff. Thanks, Matt (and everyone) for jumping in. The technical reasons behind why 11ac isn't won't be quite what it's being hyped to be are important, and it's great to see them spelled out. We really would benefit from clear "and this is what you'll need for cabling as you contemplate 11ac" from the vendors now- but that message really does work against the Adventures in Marketing currently going on.
ReplyDeleteMost 3x3 enterprise AP's fashion two Gigabit ports and if you bundled them thats 2G up and 2G down. Usually you would get high bandwidth coming out of the AP when you are in a high density environment -- and in such a case there will be several flows - and therefore pretty good load balancing across a 2 Gig Etherchannel. Assuming 75% efficiency -- that would mean 1.5G Up and 1.5G Down.
ReplyDeleteThe more interesting aspect is the need for 10G uplinks and higher PoE on the access switches. And 10G uplink on access switches translate to 10G aggregation switches and so on. Given that 10G costs are coming down rapidly 802.11ac will help transform Campus and Mid-size branches into 10G networks as well - something that many data centers, but only few campuses have adopted widely.
What intrigues me is how/if 11ac changes default/typical channel plans and whether smaller cells of higher capacity/high density will be considered a standard-design-template?
[…] Read more at: So- How Much Interference And How Many Wires Would You Like With Your 802.11ac? […]
ReplyDelete