Arris has put out an interesting 20 page white paper calling for new layers of quality of service in WiFi Access Points, where that device is being used with two SSIDs, as a homespot.
It backs this up by explaining in graphic detail the way the “bad apple” problem gets worse when more and more devices are attached to a single Access Point, specifically when there are 5 or 6 devices concurrently connecting in a home, plus one or two visiting devices, from passers-by.
The paper is slightly alarmist but fundamentally true and for any-one trying to get their head around WiFi it’s worth a read. The bad apple problem is when the device receiving the signal which is in the poorest worst condition, due to interference and dis-tance, takes up too much time to receive a stream of data, and compromises the rest of the Access Point clients, simply by de-priving them of delivery time.
It is a fundamental weakness of WiFi, which is a single radio or group of radios in a MIMO chip, dealing sequentially with each client, rather like a multitasking CPU, which does a tiny bit of each workload in turn. If one client cannot receive data at the op-timum speed, because the signal to noise ratio is too low, WiFi automatically changes its Modulation Encoding Scheme (MCS) and sends perhaps in 16QAM or BPSK instead of 64 QAM or 256 QAM. That takes longer and there is less attention remaining for the other clients.
Having extra radios and antennas helps because diversity and beamforming can be used to find the optical way of sending to everyone, but the problem just happens further away when the system reaches similar conditions. Passers-by will really be peo-ple walking down the street outside your house, walking further away, watching YouTube videos on your WiFi, on a second SSID. The fact that it is on a second SSID does not change the physics of the radio signals, merely separate the data streams in terms of security.
This is not the same issue as the broadband backhaul that is be-ing applied to the WiFi. In backhaul cable companies using DOC-SIS, have it over Telcos using ADSL, even vectored VDSL2, in that DOCSIS can add a further channel and use it for the backhaul of the visitor traffic, so that the DOCSIS line will never use up any customer bandwidth for the visitor traffic. Telcos by comparison can only promise that they will not use more than a certain per-centage of the backhaul, say 20%. But with WiFi bad apple prob-lems, the Access Point radio becomes the bottleneck, not the backhaul. Arris also tried it using the WiFi Multimedia Access Categories, which is the 4 layer QoS system that WiFi has, depict-ing all content in the home as the highest priority. The paper con-cluded that in many cases this actually left the home user worse off and did not act as an effective defense against the bad apple (which it won’t).
This problem was first pointed out to us by Israeli chip maker Celeno when it was first heard that Liberty Global planned to use the Celeno chip in the Horizon box in Homespots. Celeno implied a year ago that it was working on a very similar set of changes to those which Arris is calling for here.
Arris has also, in this paper, added detail about what might hap-pen if you have Internet of Things applications running as well, especially live security camera feeds. When someone on a smart phone decides to check in on their home, these feeds have to switch to high quality and go to the cloud in an uplink, and take up even more bandwidth from the camera to the router (although that might include an Ethernet link rather than always rely on WiFi). But where it does rely on WiFi, the uplink to the router is not controlled by the crude 4-level QoS, so traffic becomes even more unpredictable.
In essence Arris wants to limit the total time the second SSID is allowed to use the air interface, and it wants to limit the number of connections to a maximum of 2 on the second SSID, and run a more sophisticated QoS scheme, likely to be one the cloud downloaded Streamboost system that Qualcomm offers. This can take into account the highs and lows of your weekly broadband workload, and allows for changes in the prioritization of traffic from time to time. Arris says, “To provide the expected level of service, service providers must apply specific scheduling algo-rithms coupled with higher level application logic,” which we have taken to mean a new product that Arris is about to come out with in cloud based broadband router management.
The report concludes with, “In addition, an admission control mechanism for allowing or blocking slow client connection to the hotspot SSID is needed.” Now that’s a bit strong. The operator de-ploys all these lovely WiFi devices, which should be able to sup-port plenty of second SSID activity and suddenly this new set of rules can get loaded, possibly by a consumer, onto his or her router and ruin the entire potential.
Let’s think about this a little more clearly. First off, the “home” that Arris cited in the paper, where all the measurements for this paper where taken, is being supported by a 3 x 3 MIMO chip, but talking to 2 x 2 and 1 x 1 antenna devices, and that Access Point could be upgraded to a 4 x 4 MIMO 802.11N chip from Quan-tenna. Then all these problems happen less often and under more extreme conditions
Secondly most clusters of homes will have one house empty of internet traffic as people pass and this should always be selected first. All systems for choosing a WiFi stream select the least busy home with the strongest signal.
In an MDU there will be plenty of signals to choose from in the real world and none of these problems will create much of an issue – initially. In fact the biggest problem in the 802.11AC world is that there are not enough 80 or 160 MHz chunks of channels, to go around and neighbors will begin to interfere with one another. One way around this is to take your own internal user’s bad apple (tablet in the far corner of the furthest room), and put its load onto next door’s wireless router, because it is giving off a health-ier signal in the room the device is in.
That is something that you could organize automatically if you offer sign-on management across neighboring second SSIDs from the same operator.
And even then there are two technical directions we have heard that provide better areas to experiment in, mesh and MU-MIMO
The MU-MIMO direction has it that antennas can break the idea that the processor which controls the radios can only do one thing at a time, and instead it enables simultaneous transmission to groups of clients at a time — up to three or four, making more efficient use of available WiFi bandwidth and speeding up trans-missions.
This means that if there is one bad apple in the group, it only holds up one radio, not all of them. And if there are two, it only slows up two, not three or four.
There are some issues here in that the total amount of radiated power that can come out of a WiFi hotspot is regulated, so multi-ple signals have to add up to the power limit. That means each signal is that much weaker and it needs to use tricks like beam-forming and beamsteering to get enough data through. This is the approach favored by Qualcomm Atheros.
Another approach, favored by AirTies in Turkey, which offers multiple HD video streams across a group of Access Points, is to use mesh. In fact it is still threatening to announce a major mesh deal shortly, and even offers two separate backhaul systems run-ning in parallel, connecting the mesh hubs.
AirTies would say that 5 Ghz signal in an 802.11 N can be used for mesh backhaul, and both what’s left of 5 GHz and 2.4GHz can be used to speak to clients. All the mesh nodes share SSIDs, whether that is one single ID or more than one.
The mesh logic moves a bad apple from an access point that is a long way off, to the nearest access point. Now a really bad apple might crash the capabilities of this access point too, and it may harm its ability to use 5 GHz spectrum for backhaul. So the Access Points also plug into the home power grid, and if the amount of throughput goes below that of powerline, it switches to power-line.
It’s a really elegant solution and one Access Point can be damaged by a single bad apple, but all the other devices in the house can move to one or two other Access Points, and even the remote bad apple is still getting good backhaul even if the chip is saturated by its bad apple requirements, because it reverts to powerline.
These two directions may be a long way off major installations (one or two years max), but Arris, simply explaining the bad ap-ple problem and then without any further explanation, deciding upon its own brute force preferred solution, really has the poten-tial to kill off the performance of homespots, before they even get off the ground.
We wonder perhaps if Arris is in fact reflecting the internal short terms decisions of one of its clients, either AT&T or perhaps even Comcast (we think it uses double SSID devices from Cicso so per-haps not), and simply coming up with an approach which is ap-proved by that operator. But cutting off homespot customers once they are a problem to the home network, seems a short-sighted way to get the best use out of all those homespots out there.