Beyond the Tweeting Toaster: IoT Analytics with Apache Storm, Kafka and Arduino

Slides from my presentation at Data Day Seattle 2015.

Transcript

1. Beyond the Tweeting Toaster IoT Streaming Analytics with Apache Storm, Kafka and Arduino P. Taylor Goetz, Hortonworks @ptgoetz
2. Credit where credit is due…
3. @mytoaster • Created by Hans Scharler in 2008 • http://nothans.tumblr.com
4. 26 billion IoT devices by 2020 -Gartner http://www.gartner.com/newsroom/id/2636073
5. IPv4 Address Space: 4.6 billion
6. Sensors are everywhere!
7. Is that a sensor in your pocket? • GPS • Proximity Sensor • Ambient Light Sensor • 3-Axis Accelerometer • Magnetometer • Gyroscopic Sensor • Wifi • Camera(s) • UI (senses user interaction) • iBeacon
8. In your car? sensormag.com
9. In your car?
10. On your wrist? Jawbone UP FitBit …
11. 2014 South Napa Earthquake August 24, 3:20 AM (no, the earthquake was not caused by people waking up)
12. Sensors
13. –Wikipedia A sensor is a device that detects events or changes in quantities and provides a corresponding output, generally as an electrical or optical signal; for example, a thermocouple converts temperature to an output voltage.
14. Sensors
15. http://adafruit.com
16. http://adafruit.com
17. IoT Use Cases
18. Detect : Predict : React
19. Detect : Predict : React Detect behavior. Predict behavior. React to behavior.
20. Hotel Room Monitoring and Automation • Why heat/cool an unoccupied room? • Detect occupancy. • Predict occupancy. • React to occupancy.! • Added benefits to customer experience. • Analyze customer behavior.
21. Quikie Auto Lube • Manage inventory in response to demand. • Detect inventory. • Predict customer demand. • React accordingly.
22. Hospital Infection Control • CDC: Hospital acquired infections cost $30B per year and lead to 100K patient deaths. • Inadequate hand washing a big cause.
23. Hospital Infection Control • Ensure proper hygiene of medical staff • Detect staff presence. • Predict hand washing. • React to inadequate hygiene.
24. Hospital Infection Control
25. Auto Insurance • Rethinking traditional risk assessment • Detect (un)safe driving practices. • Predict who is most at risk. • React to risk assessment.
26. Use your imagination.
27. How can you use and combine sensors to better serve your [user’s] needs?
28. What is Arduino?
29. What is Arduino? • Open Source Microcontroller Hardware + Software • Geared toward prototyping • “Physical Computing:” interacting with the environment "Arduino is an open-source electronics platform based on easy-to-use hardware and software. It's intended for anyone making interactive projects.”
30. http://adafruit.com
31. What is Arduino?
32. Programming Arduino • Official cross-platform IDE written in java • C/C++ with some sugar • Program referred to as “Sketch” • Many open source libraries available for various hardware (sensors, etc.)
33. Going Wireless with XBee
34. What is XBee? • Radio modules that support wireless point-to-point communication • Serial communication • Minimal connections required — power, ground, data in, data out (UART) • 2 power options (1 mW/100 mW) • Support for multiple network topologies (Mesh, star, tree, etc.)
35. Architecture/Data Flow Transmit raw sensor data. Receive data, Add timestamp, Publish. Reliable queue. Analytics, persistence, alerting. Sensor (XBee/Arduino) Collector (XBee/R-Pi) Kafka Storm 2.4 GHz TCP/IP
36. Why Apache Storm? • Speed: Process streaming data in realtime • Scalability & Fault Tolerance • Flexibility: Single event + Microbatch/Transactional APIs • Choose the latency/throughput balance that best for your use case. • At-most-once, at-least-once, exactly-once semantics.
37. Why Apache Kafka? • Distributed, Reliable Pub/Sub Event Queue • Allows consumers to rewind to specific points in the queue • Redeploy topologies without data loss • Durability: Provides everything Storm needs for exactly-once and at- least-once guarantees.
38. Architecture/Data Flow Storm Zone 2 Zone 1 Kafka
39. Architecture/Data Flow Analytics Layer (Real-Time/Interactive/Batch) Storm Collection Layer (Device Network) Kafka HDFS/HBase/ Hive etc.
40. Architecture/Data Flow Sensor Ouput Kafka Storm Persist/ETL Analysis Alerting Persist/archive raw/intermediate data for batch/interactive flows and views (e.g. Lambda, etc.) Detect : Predict : React } }
41. Architecture/Data Flow Sensor Ouput Kafka Storm Persist/ETL Analysis Alerting Persist/archive raw/intermediate data for batch/interactive flows (e.g. Lambda, etc.) Detect : Predict : React } } Feed your model
42. Where do I put the smarts? In the IoT Device? Or in the Analytics Layer?
43. Why put smarts near the edge? • Required for User Experience • Device collaboration • Bandwidth Limitations • Storage Limitations
44. Whay not put smarts near the edge? • Updating hardware in the field is HARD! • Updating firmware in the field is almost as hard! • You probably got it wrong in the first place. Now what?
45. Why not put smarts near the edge? • Save all the things! • You will get it wrong. • Weave a Safety Net (i.e. CYA) • Use batch processing to correct errors.
46. Save all the data. Storage is cheap.
47. You can't analyze data you don't have.
48. Demo Twitter: @ptgoetz #ddsea15
52. “Three eyes are better than two!”
53. Arduino Esplora Several sensors included. (no soldering required)
54. Arduino Sketch (Sensor) #include <stdio.h>! #include <Esplora.h>! ! void setup() {! // initialize the serial communication:! Serial.begin(9600);! }! ! void loop() {! // read sensor variables! int loudness = Esplora.readMicrophone();! int light = Esplora.readLightSensor();! int temp = Esplora.readTemperature(DEGREES_F);! int slider = Esplora.readSlider();! int joystickButton = Esplora.readJoystickSwitch();! int xAxis = Esplora.readAccelerometer(X_AXIS);! // … ! ! Serial.print("{");! // Misc. Sensors! printAttribute("temperature", temp, false);! printAttribute("loudness", loudness, false);! // …! Serial.println(“}");! ! delay(1000); ! }! ! Initialize serial communication Read sensor values Dump JSON to serial port (XBee) Include Esplora convenience lib
55. Serial Monitor (Collector) • Read serial data, parse JSON • Add timestamp • Add device/sensor UID (“Sector 7-G”) • Publish to Kafka
56. Radiation Leak Topology Raw Sensor Output (JSON) Extract Req’d Fields Evaluate Threshold Raise Hell! Kafka Spout Parse Bolt Threshold Bolt Alert Bolt Shuffle Grouping Fields Grouping Shuffle Grouping
57. Swag Time Twitter: @ptgoetz #ddsea15
58. Swag Time Be the 5th person to retweet the last tweet from the demo.
59. Resources • Apache Storm
 http://storm.apache.org • Apache Kafka
 http://kafka.apache.org • Arduino
 http://arduino.cc • Adafruit
 https://www.adafruit.com • SparkFun
 https://www.sparkfun.com

60. Seattle Apache Storm & Kafka Meetup http://www.meetup.com/Apache-Storm-Kafka-Users/
61. Thank You! P. Taylor Goetz, Hortonworks @ptgoetz Questions?