802.11ac: 10Gig Uplinks Are Overkill

In my most recent blog on 11ac, I outlined the work required to upgrade WLAN infrastructures to get the full benefit of the pending wireless standard. While that work is significant, it may not require a 10-Gbps backhaul on the wired side as some observers have suggested.

The 802.11ac standard will be phased in in two waves. Wave 1 promises throughput of about 1.3G bps and uses 80-Mhz channels. Several vendors are shipping Wave 1 products. Wave 2, scheduled for 2014, promises maximum throughput of 3.5 Gbps.

The two-phase rollout means IT has several choices: Upgrade to 11ac Wave 1 now to get improved performance over 11n, wait until next year to go all in on Wave 2, or expend capital on two rounds of upgrades. If your environment has new switches and is easy to cable, two rounds of upgrades to adopt 11ac now and then get to its top end in the next couple of years may not be a big deal. Bigger environments (my own WLAN is pushing 4,000 access points) with cabling challenges and a mix of switch technologies have more to ponder.

Back to the numbers. As mentioned in my last piece, emerging industry guidance recommends two data cables per 11ac access point. Access points are already shipping with two Ethernet ports (like Aruba's new 220 series and Motorola's 8200s), though there are variations. Some early APs, like the 220, use an Etherchannel for bandwidth, while others use the second port for power or module-based feature capability. Nothing is simple with 11ac.

Another issue of contention is whether you need a 10-Gig uplink to support the max throughput of 802.11ac. 10 Gig sounds like a good idea for a standard that promises dual-band connectivity and data rates of 3.5 Gbps in the 5-GHz band, and in some cases simultaneous 600 Mbps in the 2.4-GHz band.

But maybe not. Wireless is a shared-medium, half-duplex technology with lots of overhead. As a rule of thumb, you can take any stated wireless data rate and cut it roughly in half to get a sense of what real throughput--the part that makes it to the wired uplink--will be.

This is why a single Gigabit uplink is sufficient for the expected 1.3 Gbps rates of Wave 1. Even when paired with throughput on the 2.4-GHz side of a dual-band AP, first-wave 11ac in 5 GHz won't fill the 1-Gbps uplink. So far, so good, right?

Remember, 11ac promises rates "up to" 3.5 Gbps in Wave 2, just as 11n promised rates up to 600 Mbps. In practice, 11n peters out at 450 Mbps. It's not unreasonable to think that most environments won't see 11ac's maximum rate, either, but there is just no way to know yet. (The 11ac standard also promises 6.9-Gbps maximum throughput at some point, but that's beyond the first two waves.)

If you do the "cut it in half" math on that number, Wave 2's best throughput is still under 2 Gbps, which means a 10-Gbps uplink isn't necessary.

For a better description on the nuances of the 11ac rates, see my colleague Matthew Gast's excellent blog, but be warned: I challenge whether even 2-Gbps backhaul will be needed for many 11ac cells.

One reason is that the explosive growth in mobile devices works against the ability to fully exercise 11ac's high rates because most smartphones operate at what amounts to fast 11n rates. To date, there have been no public hints that faster mobile silicon is on the horizon. Tablets don't fare much better, and laptops (which have the best chances of being equipped with the most robust 11ac adapters) are statistically in decline.

Also, regardless of the advertised top rate, clients need to be very close to the AP to achieve it. Even in dense deployments, only a fraction of clients will be close enough to get the max rates, and that only happens if they are on a laptop that supports the high rate.

We also have to consider that 11ac works exclusively in 5 GHz, and that as the new standard gets as popular as it should, 11ac's wide channels may eventually make the 5-GHz spectrum almost as cluttered as 2.4 GHz is now. This also puts up speed bumps against the fastest rates as RF works against itself. And where my 11ac network bumps up against yours, built-in courtesy mechanisms will slow us both down.

In most cases, I think WLANs will do well with 1G bps to each 11ac access point, unless you have a room full of 11ac laptops all doing HD video.

The bottom line is that 11ac is complicated. It's easy for the WLAN industry to simply tell us all to run new Etherchannels for each new AP, but that puts a lot of cost and burden on the customer for promises that have yet to be proved.