Introducing the web of things and some work on open source implementations

Transcription

1 Introducing the web of things and some work on open source implementations Dave Raggett, W3C 北京, 9 July 2016

2 Massive Potential for the Internet of Things Smart Homes Wearables Healthcare Power & Environment Smart Cities Manufacturing 2/30

3 IoT Landscape 3/30

4 Lots of standards organisations 4/30

5 IoT is being held back by fragmentation The IoT is still very immature with Lots of incompatible platforms Lots of standards Lots of IoT technologies Lack of interoperability at the application level This creates confusion and holds back market opportunities Fragmentation increases costs for developers Fragmentation increases risks for suppliers and customers Fragmentation reduces the market size 5/30

6 Analogy with early days of networking Before the Internet there were many incompatible networking technologies IP made it simple to connect different network technologies Standard sockets API independent of the underlying networks Simplifying application development Network services took off with the introduction of IP Bigger markets and Lower costs The network effect The Web and exponential growth 6/30

7 Living with a mix of platforms and protocols Variation in requirements across application domains No single platform or protocol will suffice How can we enable services across a mix of platforms? The solution for interoperability is metadata that describes: The interface exposed to applications for interacting with things Communications and security requirements for access to things Lightweight semantic models and domain constraints Roles and relationships between things Validation to ensure interoperability Search across a web of services Composition of services 7/30

8 The Web of Things Things denoting physical or abstract entities Things with properties, actions and events Building upon success of event driven programming Every thing is named with a URI This name acts as a link to the thing s description The description tells platforms how to access the thing A scalable framework for data and metadata W3C Resource Description Framework (RDF) Web developer friendly representation with JSON 8/30

9 Scripting things at a distance... Thing descriptions can be used to create proxies for a thing, allowing scripts to interact with a local proxy for a remote entity Scripts can run on servers or as part of Web pages in Web browser for human machine interface Thing topologies Peer to Peer, Peer to Peer via Cloud, Star, Device to Cloud, Star to Cloud Uses Thing Publishes Thing 9/30

10 Communications Stack with a clean separation of concerns Application Developer (WoT focus) Platform Developer (IoT focus) Application Things Transfer Transport Network Scripts that define thing behaviour in terms of their properties, actions and events, using APIs for control of sensor and actuator hardware Software objects that hold their state Abstract thing to thing messages Semantics and Metadata, Data models and Data Bindings of abstract messages to mechanisms provided by each protocol, including choice of communication pattern, e.g. pull, push, pub-sub, peer to peer, etc. REST based protocols, e.g. HTTP, CoAP Pub-Sub protocols, e.g. MQTT, XMPP Others, including non IP transports, e.g. Bluetooth Underlying communication technology with support for exchange of simple messages (packets) Many technologies designed for different requirements 10/30

11 From the network edge to the clouds 11/30

12 Architectural Patterns Browser for HMI Gateway (apps) Cloud based Services Powered, multi-protocol Ambient or battery operated IoT devices Firewall 12/30

13 Web of Things Gateways A gateway can support more than one protocol TCP, CoAP over UDP, HTTP, WebSockets*, Bluetooth, A gateway can support low end devices IoT device discovers gateway or vice versa Gateway identifies the device and gets description from cloud Gateway publishes things locally, e.g. over WiFi Gateway publishes things remotely via cloud server A gateway can act as an application server for apps Flash based storage for data logging, apps, etc. Invitation to install app when visiting vendor s website Apps coded in JavaScript, Java, native code, etc. * HTTP and WebSockets needed to support direct access by Web browser 13/30

14 Proxy Chains* A software object in one device can act as a proxy for a software object running in another device The platform creates these objects based upon the thing s description These objects can form proxy chains 1. Initial object created on IoT device 2. Proxy for (1) on an IoT gateway 3. Proxy for (2) on a cloud server 4. Proxy for (3) in web page script executing in a browser Platform manages the messaging up and down the proxy chain The application code is unware of this Different links in the chain may use different protocols * Actually these can form trees since each server may have multiple proxies 14/30

15 Some observations The Web of Things is asynchronous It takes time for messages to cross the Internet This depends on the protocols and communication patterns Different communication patterns are appropriate in different contexts Push, pull, pub-sub, peer to peer Single sensor readings or actuator updates Multiplexed and buffered readings Streams for data logging, display and analysis Many different and evolving IoT technologies IP based, e.g. HTTP, CoAP, MQTT, XMPP WebSockets, Wireless: WiFi, sensor networks with Thread,... Non IP, e.g. Bluetooth, ZigBee, Z-Wave,... LPWAN with LTE-M, LoRa, SigFox, Weightless,... 15/30

16 Data types for application scripting Decoupling data types from metadata formats and communication protocols Things with properties, actions and events Properties (may have sub-properties) Actions with asynchronous responses (zero, one or more) Events signalling something s happened, e.g. proximity alert Data types that can be assigned to properties, or passed with actions and events null, boolean, number, string, array, object (name value) things (passed as URIs for thing descriptions) streams (interface for time sequence of data tuples) Early and late bound types Sometimes you only have partial knowledge in advance Integrity constraints Min and max values for numbers Enumerated values e.g. for strings Cardinality constraints (e.g. on how many things are expected) Constraints that span multiple properties (expressions over paths) 16/30

17 Lightweight semantic models Interoperability depends on sharing the same meaning as well as the data Semantic models define what it means for a thing to be of a certain kind A temperature sensor, for example, must provide a numeric value with the physical units from an agreed set, e.g. Celsius, Fahrenheit or Kelvin What kinds of metadata a thing must or may have Modular to support easy composition of semantics Versioning to support old and new definitions Cross domain terms vs domain specific terms How to support agile processes for standardising vocabularies of terms at different stages of maturity? Experimental, limited commercial deployment, massively deployed Need to encourage participation of many different communities Encouraging re-use of existing vocabularies where practical Schema.org as a useful precedent Need to allow for different formats for representing metadata JSON, Turtle/N3, XML, 17/30

18 Data Security Essential to retain trust and realise the opportunities Breaches of privacy, cybercrime, physical safety, threats to national infrastructure and looming risks of cyberwar Challenges for securing embedded devices, and countering threats based upon gaining physical access to devices Tamper resistant hardware Strong mutual authentication Secure transmission and storage of data Important of secure software upgradability Challenges with provisioning and managing devices and services Initial set up, change of owner, change of service provider Boot strapping trust, and relationship to trust delegation Risk of user experience becoming the weakest link Necessity of keeping up with best security practices 18/30

19 Privacy and the IoT The IoT has the potential to provide huge and unprecedented amounts of personal information This information may last indefinitely Risk of abuse by individuals, criminals, companies and governments Sense of intrusion into your personal space Fear of harm due to disclosure of personal information Strongly identifying information Your address, data of birth, sexual orientation, Principle of data minimisation high cost to companies for handling personal data securely Privacy policies determining what purposes data can be used for, and for how long Weakly identifying information When sufficient such data is combined this can uniquely characterise you Companies need to provide privacy policies on how they handle such data Need for adhering to best practices to avoid reputational damage to companies Including regulatory requirements 19/30

20 My Open Source Projects NodeJS based project For relatively powerful devices, e.g. Raspberry Pi running Linux Work at an early stage C++ for low-end microcontrollers Initially for Arduino Uno and Wiznet based Ethernet Shield Highly constrained low cost microcontroller Arduino has lots of nice hardware add-ons C++ for gateways on more powerful microcontrollers 32 bit ARM based devices from under 100 KB to MB RAM Compiles and runs on Linux and OS X for ease of development Why am I doing this? Nothing like working code for gaining a deeper understanding Huge potential market for low-end IoT devices Role of Maker communities for popularising the Web of things 20/30

21 Required Hardware Arduino Ethernet Shield Native support for IP MicroSD card slot Cost: 4.75 GBP on ebay Arduino Uno with ATmega328P MCU 2 KB RAM 1 KB EEPROM* 32 KB FLASH Lots of I/O pins Cost: 2.33 GBP on ebay * EEPROM useful for storing settings and error logs, etc. 21/30

22 Required Software Arduino IDE Free from Serial driver for CH340 USB to serial chip Required for cheap Arduino clones A bit of a pain to find and set up L The Web of Things library Your Arduino sketch 22/30

23 Examples Proximity sensor US-015 ultrasonic distance measurement Arduino sketch Control via GPIO pins digitalwrite(), pulsein() Events via CoAP protocol Accelerometer & Gyro GY-521 board with MPU axes over I2C bus Arduino sketch Read data via I2C bus Stream data over TCP connection Block of six 16 bit numbers 23

24 Interrupts and Events Hardware interrupts are valuable for handling sensors and actuators etc. But the interrupt service routine (ISR) must be quick to avoid blocking other interrupts Event driven behaviour is very convenient, but event handlers can take time to execute Solution: ISR pushes event onto queue, which is dispatched from within Arduino loop() method Same single threading execution model as for Web browsers event handlers are executed sequentially Saves developers from having to deal with complex synchronisation issues 24/30

25 Wiznet W5100 Ethernet Controller Native Internet Protocol support TCP, UDP, multicast, ARP, IGMP protocols 1 to 4 sockets with cyclic receive and transmit buffers Allocated from 16 Kbytes of embedded RAM Accessible over SPI bus Polling or hardware interrupt The default Arduino library is large and leaves little space for sketches I wrote my own W5100 library from scratch using the data sheet and vendor code samples UDP and TCP DHCP for auto configuration of client network parameters Multicast DNS for discovery of services 25/30

26 Constrained Application Protocol (CoAP) Specified by IETF CORE Working Group as RFC 7252, RFC 7641 Loosely speaking equivalent to HTTP over UDP datagrams Binary headers for compact messages HEAD, GET, PUT, POST, and DELETE request methods Response has numeric status code, content type and payload Asynchronous responses to GET with observe option CoAPserver sends new response whenever resource changes value Reliable delivery via setting CON flag Repeatedly send message until matching ACK is received Secure connections with DTLS (datagram transport layer security) Challenging to fully implement on low end microcontrollers Requires UDP support Buffer memory management that minimises RAM Uses considerably more RAM and code than for TCP with W /30

27 IoT Device Network Configuration Use static IP address or DHCP for dynamic address Uses mdns to discover Web of Things gateway Service type: _wot._tcp.local (unregistered) My MacBook s firewall blocks all other multicast sockets Registers remote proxy on gateway for local thing on the Arduino Uses TCP for messaging with gateway Less code due to native IP support on W5100 chip Very similar to my Web Socket implementation for NodeJS web of things server CoAP would take more space and be less capable Note the choice of protocol depends on the choice of hardware, which in this case is an Ethernet controller 27/30

28 Software architecture Avoid malloc and free for robust operation Fragmentation and long term instability Static node pool JSON for numbers, booleans, objects, arrays etc. Average Length Binary trees (objects and arrays) Mark/Sweep garbage collector In place defragmentation algorithm 6 bytes per node on the ATmega328P JSON object property names replaced by symbols Efficient binary message encoding/decoding Relies on deterministic assignment of symbols Designed with short packet technologies in mind e.g. Nordic nrf24l01+ with 32 byte payloads 28/30

29 Life after Death Eventually something bad will happen Memory issues Running out of RAM for the program stack Running out of free nodes for use with JSON Overflow of critical queues Program bugs You can never be 100% confident that there are no bugs Bad data Violation of design assumptions Software restart on detection of fatal errors Hardware watchdog timer forces restart if not reset in time This allows the device to recover after being stuck in a loop Gateways and other devices must be resilient to such restarts 29/30

30 More information The Web of Things is in a pre-standardisation phase W3C Web of Things Interest Group We expecting to launch additional groups later this year W3C Business Group to study business and policy requirements W3C Working Group to standardise APIs and metadata vocabularies We want people to explore the ideas experimentally and help us to build a shared understanding of what is mature enough to move into standardisation Further reading WoT IG Current Practices for Plugfests Some other ideas for types and use of JSON for metadata Pointers to related implementation work 30/30