Enabling third-party apps in home gateways for and - PowerPoint PPT Presentation

Enabling third-party apps in home gateways for and (through virtualization, edge computing, SDN, and RF management) Suman Banerjee (suman@cs.wisc.edu) 1

IoT and smart home apps

IoT and smart home apps Should the cloud manage your home thermostats?

IoT and smart home apps Should the cloud manage your home thermostats?

IoT and smart home apps Should your home security feed be uploaded to the cloud?

IoT and smart home apps Should your home security feed be uploaded to the cloud?

IoT and smart home apps Can we watch 4K streaming movies?

IoT and smart home apps Can we watch 4K streaming movies? Network variability

What is ParaDrop?

What is ParaDrop? Edge computing in the extreme Move the cloud right into your home

What is ParaDrop? Edge computing in the extreme Providing a cloud-like abstraction in your home WiFi router

WiFi router has low latency to home devices, and is always-on 3 rd party apps/services drop into your home WiFi router on-demand

How it works? SeeCam cloud ParaDrop service cloud manager SeeCam

How it works? SeeCam EnvSense cloud cloud ParaDrop service service cloud manager SeeCam EnvSense

ParaDrop examples • Transcode video to adapt to wireless channel conditions • Cache movies in router from head of instant queue • Vehicular: Passenger tracking and engagement • Home vent control

More examples … • Kids router • Framily router • TOR in your router • Monitor your ISP (a la BisMark) • <Your idea here …>

Let us know if you need one! (We can ship you one or more) Check out www.paradrop.io

ParaDrop smart router • A programmable substrate • Virtualization framework • “Chutes” • Isolated and proprietary • Multiple wireless interfaces • WiFi, Bluetooth, ZigBee • Cloud-based RF management

Overall architecture Cloud-based Virtualization Wireless and Management edge computing SDN Commodity AP Hardware 19

Overall architecture Cloud-based Virtualization Wireless and Management edge computing SDN Commodity AP Hardware 20

Problem statement Wireless users Expectation: wire-like performance Experience : “seems to work well mostly” “flaky”, “connection keeps dropping” [thread from user mailing list]: “Is anybody else finding the network to be slow/flaky ? I want to sanity check before I tell lab admin ” “It was flaky for me yesterday . Restarting the laptop seemed to solve the problem”

Problem statement What is the cause of interference?

Problem statement Network administrators [email from wireless network admin]: “We continued to hear reports of problems, but could not reproduce the problems "on demand" nor do we have enough details to identify the real issues.” Problem: issues are transient, lack of details and tools “Is my wireless network operating well enough?”

A “blood test” for the network (Comment made by Pravin Bhagwat: CTO, Airtight Networks)

Basic question Access points WLAN score L6 L1 L3 L5 L2 BAD GOOD L4 Can we provide a real-time, overall health score for the Clients entire network?

Basic question Impact: 0.7 WLAN score Impact: 0.3 L1 Impact: 0.1 L2 BAD GOOD Impact: 0.2 L3 L6 Links L4 weak signal L5 Identify why certain L6 links are BAD GOOD under-performing Goodness of the link

RF Management goals Cloud controllers Key questions: • Detect: Who are the real interferers? • Quantify: How much interference do they cause? • Locate: How do we locate these non-WiFi devices? • Mitigate: What configuration needed to mitigate them?

Our solution Real-time Interference estimators WiFi to WiFi Interference Non-WiFi to WiFi Interference (use WiFi-only hardware!) - Airshark - COLLIE (Collision Inferencing Engine) 1. detect non-WiFi devices - WiFiNet - PIE (Passive Interference Estimator) 1. quantify interference impact 2. pin point device location

Spectrum at a university cafe High powered non-WiFi devices share the spectrum with WiFi devices 29

Is non-WiFi interference a real problem? How severe is the impact? More than 50% throughput drop and, in some cases, throughput drops to zero! How prevalent are the devices? Across locations, many devices frequently appeared with a high signal strength 30

Non-WiFi interferers ZigBee Analog Cordless Phone Video Camera (Jennic JN 5121) (Uniden EXP4540) (Pyrus Electronics) Microwave Ovens High duty devices (Whirlpool, Daewoo, Sunbeam) Broadband interferers Frequency hoppers Bluetooth ACL/SCO FHSS Cordless Phone Game controllers Audio Transmitter (iPhone, iPod touch, (Panasonic KT-TG2343) (GoGroove PurePlay) (Wii, PS3) Jabra headsets)

Use a coarse-grained WiFi lens Coarse-grained freq. resolution Coarse-grained time resolution 312 kHz Sub-carriers O(10 3 ) to O(10 4 ) samples/sec Time Frequency 22 MHz RSSI/power per sub-carrier 32 Limited spectrum view

Airshark: how it works Device Detection Pipeline Spectral Pulse Feature DT-classifiers Sampling Detection Extraction Extract Features (ON) (OFF) Training / Classification Decision Tree Classifier 33

Detection Accuracy > 98% accuracy at signal strengths >= -80 dBm 34

Many trials Including integrating with a commercial off-the-shelf AP platform 35

Works with • We built Linux drivers for spectral sampling for the AR92xx and AR93xx NICs • Open sourced and publicly available • Have also released an Android version – For Samsung Nexus devices Check wisense.io Available from Google Play 36

Cloud-based RF Management COAP: Coordination framework for Open Access Points 37

Cloud-based RF Management COAP: Coordination framework for Open Access Points 38

Cloud-based RF Management COAP: Coordination framework for Open Access Points Analytics and mitigation techniques Interference estimation 39

Cloud-based RF Management COAP: Coordination framework for Open Access Points Analytics and mitigation techniques Interference estimation Commands Management Spectrum 40

Cloud-based RF Management COAP: Coordination framework for Open Access Points Analytics and mitigation techniques Interference estimation Commands Management Spectrum 41

Field Trials • OpenWRT based APs – ALIX 2d2 platform: (500 MHz AMD Geocode CPU, 256 DDR RAM, flash storage) • 30+ APs deployed in homes & apartment complexes for 3+ years Madison, WI • Cloud controller hosted in MDU Deployment 1 off-the-shelf Linux server MDU Deployment 2 14 APs 42

Microware cuts throughput 43

30.1 25.5 16 13 100% improvement most of the time Data is sorted

Enabling third-party apps in home gateways for and Suman Banerjee (suman@cs.wisc.edu) 45