Chuck (and all), This topic really has me thinking. :-) My point was probably missed because I put it in the wrong section of my statement. When I wrote that I was thinking about Ethernet throughput. My fault. The absolute fastest situation possible with Wi-Fi is one AP and one client of the same Wi-Fi standard and capabilities. Anything more and efficiency is reduced. But as you said, even with MU-MIMO its well below Gbit speeds.
I started to write a bunch of stuff and then I realized I wrote what I wanted to say in a blog 18 months ago. I promise I’m not trying to pimp a blog that I write in every 12-18 months. :-) https://gthill.wordpress.com/2015/03/15/random-thoughts-on-wave-2-mu-mimo When I wrote that I worked for Ruckus. I no longer work for a Wi-Fi manufacturer. Otherwise Lee would have kicked my butt by now. :-) Suffice it to say, it is my opinion that MU-MIMO will not be looked upon favorably years from now. It won’t be looked upon as bad, but it isn’t revolutionary. I’m not saying you should avoid an AP with MU-MIMO. If its the latest and greatest and you can afford it, then fine. But the efficiency gains are still unknown. And of any Wi-Fi technology to date, its the one that has to have everything just right to be useful at all. Here’s a list of what has to be right in order to take advantage of MU-MIMO Plenty of MU-MIMO compliant devices on the same AP radio. Off the cuff, I’m saying 30+. This is because of MU group selection and the factors below. Multiple MU-MIMO clients have inbound data at the exact (within 150 microseconds – Ok I made that up but I think its close) in time. If they aren’t within that time, then the AP will just Tx in normal SU-MIMO The RSSI of all clients in the group need to be similar or you get efficiency loss. The client devices with inbound data aren’t physically close to each other. (requires spatial separation) The client devices that require data at that moment in time can’t be physically close to each other. This distance isn’t easily known since it depends on the characteristics of the room. So of any environment, large classrooms sound like the ideal use case for MU-MIMO. And it is. However, in many of your high density environments, how many clients per radio do you have? Enough for good MU group selection? What are the chances that at least 2 of them need a frame within a VERY short time of each other AND are physically separated enough to get spatial separation AND have similar RSSI? Let’s just say, I’m not optimistic. :-) GT Hill From: The EDUCAUSE Wireless Issues Constituent Group Listserv <[email protected]> on behalf of Chuck Enfield <[email protected]> Reply-To: Chuck Enfield <[email protected]> Date: Friday, August 5, 2016 at 1:41 PM To: <[email protected]> Subject: Re: [WIRELESS-LAN] Outsourced ResNet Thanks GT, I definitely agree with your overall point, but I have to take issue with the following: MU-MIMO just takes the same number of streams and distributes them to multiple clients. For example, 3 MU streams has no greater Eth load than a 3x3:3 client on a 3x3:3 AP. This statement is technically correct but incorrectly applied. MU-MIMO doesn’t increase the max theoretical throughput of an AP, but it will significantly increase the real-world throughput in some situations. The ability to talk to multiple devices simultaneously effectively reduces contention. Reduced contention will mean higher throughput in contentious environments. If you have lots of contention from single-stream devices and really high channel duty cycles, then you can reasonably expect a 2X throughput increase over the roughly 200Mb/s you’re probably seeing on wave-1 APs in that environment. That’s still well below Gbit speeds, but it’s nothing to sneeze at. If your duty cycle is low now, contention isn’t your limiting factor and wave-2 won’t affect throughput, but it could affect latency for improved real-time protocols. If you have high duty cycle from a small number of 3-stream laptops doing large file transfers you won’t see much benefit either. So what is the res-hall environment? It probably falls into a mix. A few devices on each AP are generating 90% of the traffic, but there are enough devices on an AP to see some performance benefits from MU-MIMO. My expectation would be a 30% to 50% throughput increase in a busy res hall network, but those are based on shorthand calculations rather than real-world measurements. Chuck Enfield Manager, Wireless Engineering Telecommunications & Networking Services The Pennsylvania State University 110H, USB2, UP, PA 16802 ph: 814.863.8715 fx: 814.865.3988 From: The EDUCAUSE Wireless Issues Constituent Group Listserv [mailto:[email protected]] On Behalf Of GT Hill Sent: Friday, August 05, 2016 1:09 PM To: [email protected] Subject: Re: [WIRELESS-LAN] Outsourced ResNet Hello all… Just a few thoughts on this topic. Wave 2 isn’t any faster than wave 1 so it doesn’t need two Eth ports etc. Now, by true specification, yes it CAN be faster but that’s only because of 160 MHz channelization. MU-MIMO just takes the same number of streams and distributes them to multiple clients. For example, 3 MU streams has no greater Eth load than a 3x3:3 client on a 3x3:3 AP. However, new 11ac APs are 4x4:4. So technically they can be faster. But, the only way that will have any effect whatsoever is if you have a 4 spatial stream client device. And while those will come out (if not already) most devices on campus are mobile, so 2 spatial stream max. MU-MIMO would then be able to send two, two stream transmissions. However, keep in mind that each MU-MIMO stream will be lowering its data rate vs. a single device. (longer discussion) One single 1 Gbps port will take you through to 11ax. Wi-Fi is half duplex and Eth is full. I used to work for a Wi-Fi manufacturer and in any test we could throw at it, we couldn’t get 1 Gbps ethernet to be our bottleneck except is completely unrealistic environments (single direction traffic only, 160 MHz channelization, 4x4:4 client etc) Wave 1 to Wave 2 is a VERY small upgrade in the grand scheme of things. 11g to 11n was revolutionary. MU-MIMO hasn’t been proven except in a lab. Yes, in perfect scenarios it can provide some improvement. But there is a lot of cost (overhead) in making MU-MIMO work. Dollar for dollar, I would only consider MU-MIMO APs in my most highly dense areas. And even for that I may not be convinced… Look at individual features on wave 2 APs. There ARE sacrifices in new technology for sake of getting it to market. Often times you will see better performance from an older generation (I use generation loosely with 11ac W1 to W2) APs. Look to make sure that all performance features (ATF, band steering etc) are there are newer APs. Oddly enough, some features are dropped b/c programming those into a new chipset takes TIME. Random thoughts I am not saying don’t buy W2 APs. I’m saying that you shouldn’t expect the features in W2 to have that much of an improvement New chipsets are almost always better at PHY level stuff vs. older chips EVEN with the same specs (3x3:3, 4x4:4 etc). Chip manufacturers just get better at what they do. Don’t forget about 11ax. Its here in two years and it should have significant improvement for high-density (not overall, single device throughput) applications. Client devices will of course take some time but as someone mentioned, higher-ed has the fastest client adoption turnover in any vertical. Sorry that was such a long response. GT Hill From: The EDUCAUSE Wireless Issues Constituent Group Listserv <[email protected]> on behalf of Philippe Hanset <[email protected]> Reply-To: The EDUCAUSE Wireless Issues Constituent Group Listserv <[email protected]> Date: Friday, August 5, 2016 at 11:34 AM To: <[email protected]> Subject: Re: [WIRELESS-LAN] Outsourced ResNet Brian, Food for thoughts... How is the over-subscription to the commodity Internet keeping up with Wi-Fi these days? Most services are in the cloud and it seems that Internet Commodity could be the limiting factor rather than wave1 or wave2 or even staying with 802.11n. Is it worth worrying about 802.11ac wave 1 or wave 2 when your Wi-Fi is so much more capable than your campus uplink? (or is it?) When we talked about 802.11g VS 802.11n there were huge differences between the two. Is it still the case between wave 1 and wave 2? Software support lifecycle seems to be the main determining factor in Wi-Fi infrastructure upgrades. So, rather than Wave1 VS Wave2, we should maybe consider vendors with longer software lifecycle support. Also, many of us upgraded from 802.11n to 802.11ac building-wide and even campus-wide because n and ac didn’t play well together. How do Wave1 and Wave 2 play together? Philippe Philippe Hanset www.eduroam.us www.anyroam.net On Aug 5, 2016, at 12:01 PM, Jeffrey D. Sessler <[email protected]> wrote: There are few problems I see with this line of thinking. a) This is the same argument people made when 802.11n arrived i.e. Stick with 802.11g as it’s less expensive, proven, and there are hardly any 11n clients. For those of us who jumped on the cutting edge, we road an explosive wave of 11n clients and all the benefits of being prepared for it. Others that stuck to 11g no doubt regretted their decision. b) If there is a cost difference between Wave 1 and 2 it’s because the manufacture knows Wave 1 is dead, and they are more than happy to get that inventory cleared out. You’ve just purchased on the declining edge of that technology’s life-cycle. c) Life-cycle. If your AP life-cycle is say five years (or longer), a Wave 1 AP is already a couple of years into its eventual EOS/EOL with the vendor. This means you could get four years out and it’s no longer supported by current controller code. By purchasing at the leading-edge, you’re many more years from having to deal with that scenario. Jeff From: "[email protected]" <[email protected]> on behalf of James Andrewartha <[email protected]> Right now I would still buy mid-range Wave 1 APs, because the pricing is significantly cheaper, and there’s hardly any MU-MIMO clients yet, Apple devices in particular. ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/. On Aug 5, 2016, at 12:01 PM, Jeffrey D. Sessler <[email protected]> wrote: There are few problems I see with this line of thinking. a) This is the same argument people made when 802.11n arrived i.e. Stick with 802.11g as it’s less expensive, proven, and there are hardly any 11n clients. For those of us who jumped on the cutting edge, we road an explosive wave of 11n clients and all the benefits of being prepared for it. Others that stuck to 11g no doubt regretted their decision. b) If there is a cost difference between Wave 1 and 2 it’s because the manufacture knows Wave 1 is dead, and they are more than happy to get that inventory cleared out. You’ve just purchased on the declining edge of that technology’s life-cycle. c) Life-cycle. If your AP life-cycle is say five years (or longer), a Wave 1 AP is already a couple of years into its eventual EOS/EOL with the vendor. This means you could get four years out and it’s no longer supported by current controller code. By purchasing at the leading-edge, you’re many more years from having to deal with that scenario. Jeff From: "[email protected]" <[email protected]> on behalf of James Andrewartha <[email protected]> Right now I would still buy mid-range Wave 1 APs, because the pricing is significantly cheaper, and there’s hardly any MU-MIMO clients yet, Apple devices in particular. ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/. ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/. ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/. ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/. ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/.
