A few people have pointed me at Lee Badman’s blog on backhaul for 802.11ac, and it’s certainly a hot topic of conversation. Although we’re still early on the curve with 802.11ac (the Wi-Fi Alliance certification program hasn’t launched yet), any major network upgrade is being done with an eye squarely on how to make it work better for Wi-Fi.
With 802.11ac, then, the question is what kind of backbone is required to support 802.11ac now, and into the future.
Like 802.11n, 802.11ac will come to market in distinct waves. In the first wave, the capabilities will include 80 MHz channels, 256-QAM, and probably three spatial streams. That gets you to a peak rate of 1.3 Gbps. It’s fast, but not fast enough to outrun the gigabit Ethernet port coming out the back of the AP. Ethernet and Wi-Fi talk about speeds in very different ways; Ethernet speeds are quoted after the overhead of framing and medium contention, and Wi-Fi speeds are the peak data rate for the frame and ignore framing and protocol overhead.
Taking into account Wi-Fi protocol overhead alone pushes the “1.3 Gbps” back down below a gigabit. Early Wi-Fi networks struggled to get 50% of the peak data rate; 802.11n improved substantially but there is still significant overhead.
More importantly, 802.11ac speeds depend on the mix of devices attaching to the network. To get to that 1.3 Gbps number, the client device needs to support 80 MHz channels and three spatial streams. Many laptops will, but all the cool devices that everybody carries around are battery-powered mobile devices, which are nearly exclusively single-stream. Just taking the number back to a single stream drops the speed to 433 Mbps (and that’s with 256-QAM; if you can’t support 256-QAM, the speed is only 325 Mbps). Remember to subtract out protocol overhead, and you’re down to a couple hundred megabits. Fast – especially for a battery powered device! – but not backbone-breakingly fast. You’ll be fine with a single gigabit uplink, even once you have to add in the 2.4 GHz 802.11n radio.
When the second wave of 802.11ac hits, the typical AP is likely to support 160 MHz channels and four spatial streams for a peak data rate of 3.5 Gbps. However, single-stream mobile clients won’t see a huge boost. They stay stuck at a single stream, but may get a bump up to higher speeds by doubling channel bandwidth. It’s a raw speed of 650 Mbps (64-QAM) to 867 Mbps (256-QAM), but only when you can occupy the whole channel. If half of your traffic is from mobile devices at maximum data rates, that means that the rough offered load to the backhaul is around 2 Gbps of wireless traffic. Subtracting out protocol overhead, and you’ll be fine with two gigabit uplinks.
Now add one more wrinkle into the discussion in the previous two paragraphs. I assumed that devices would be able to transmit using the full available channel width. In practice, to reach those speeds requires very clean spectrum so that transmissions will always use the wide channels. Although there’s a lot of spectrum available at 5 GHz, the newer channels are wide enough that adjacent channels will become a reality again. There might even be overlap, so cut those numbers down even further.
The big question: do you need 10-gigabit Ethernet backbones? If you have a lot of single-stream mobile devices (and who doesn’t?), then a pair of 1G uplinks from the AP are likely to suffice even in the second wave. Given that 10-gigabit Ethernet is only available on fiber now and power-over-fiber-Ethernet is a gigantic pain, it’s a good thing. 10-gigabit PoE switch ports aren’t needed for the foreseeable future.
To make a long post short: for first-wave 11ac, make sure you have a gigabit-capable edge. For second-wave 11ac, the gigabit-capable edge is still sufficient, but you’ll need two cables.