Fog in Support of Emerging IoT Applications Rodolfo Milito, PhD Cisco Chief Technology and Architecture Office Fog Computing Conference and Expo, November 2014, San José, CA
Agenda o Fog is a vehicle for IoT We examine two IoT spaces: consumer & enterprise Not all sensors are born equal Neither are the associated use cases and apps o Enterprise-based IoT is not a trivial extension of consumer-iot A Canonical example: Smart Traffic Light Systems Windmill Farm Smart & Connected Communities (S+CC) HealthCare When seconds matter: Earthquake detection 2
Whenever possible, Cloud is the answer Cloud Computing efficiencies are unbeatable - Economies of scale (CAPEX, OPEX) - Elasticity - Then, why Fog Computing? Fog Computing complements & interplays with the Cloud to support apps that require low/predictable latency, rapid mobility, and are widely Rodolfo Milito Fog Computing distributed Conference/Expo November 2014 geographically San José CA 3
Closer examination of the Internet of Things required 4
IoT Explosive Growth Billions of Devices 50 40 30 20 10 0 Rapid adoption rate of digital infrastructure 5 x faster than electricity & telephony 6.307 Cross-over Point ~6 things online per person World Population 50 Billion SmartObjects 6.721 6.894 7.347 7.83 2003 2008 2010 2015 2020 Source: Cisco IBSG projections, UN Economic & Social Affairs http://www.un.org/esa/population/publications/longrange2/worldpop2300final.pdf 5
IoT Billions & billions of sensors But not all sensors are born equal 6
Sensors: from wimpy to powerful motes power size Size Power Data rate [Mbps] mote In iinches Runs on AA bateries, or solar DAS Up to 50 Km 600 DAS, DTS Distributed Temperature Sensor (DTS) Distributed Acoustic Sensor (DAS) O&G, seismic monitoring, vehicular traffic monitoring Other dimensions: cost, processing power, memory data rate 7
A Simple & Successful Model (in its own space) Compelling simplicity, but Make things as simple as possible, not simpler endpoint devices 8
Limitations of the model Ignores feedback decision loops (fdl) o Some fdl operate in very slow time scales o Vehicular traffic patterns, power usage, etc. that feed long-term planning decisions o Some fdl require near-real time decisions o Customer behavior in retail o Some fdl demand real-time physical actuation o Elevator control in response to emergency seismic warning, collision prevention systems, smart grid controllers No account for Sensors and actuators organized as systems 9
Missing Arrows Required for feedback decision loops Geodistributed systems interact E-W Near and real-time control loops require proximity to the source 10
Organizing Principles Case in point: Centralization vs. Distribution Mainframes Laptops Cloud Architectural Decisions are constrained and guided by o Physical Laws o Intrinsic characteristics of system under consideration o Technological innovations o Economic considerations Strong interplay o Business & human drivers Ouroboros virtuous circle Innovation enables apps Apps drive innovation 11
IoE, Arguably the Largest Human Deployed System after the Internet Fresh space Built on pre-existing systems SG, S+CC, SCV, Manufacturing Smart Phones Apps Greenfield Assisted living, Smart Homes Industrial/enterpri se based Use Cases Processes and methodologies developed over decades, centuries, millennia 12
IoE, Arguably the Largest Human Deployed System after the Internet User-based IoE Industrial-based IoE Smart Phones Apps Greenfield Assisted living, Smart Homes Industrial/enterpri se based Use Cases Processes and methodologies developed over decades, centuries, millennia 13
Consumer-based IoT models do not carry well into industrial-based IoT Need to incorporate an Organic understanding of IoT Verticals 14
IoT Disruption: Beyond Connectivity Today s Dominant Endpoints Dominant Endpoints in 2025 Transportation and Connected Vehicles Precision Agriculture Industrial Automation Healthcare A person behind every device Intelligent Buildings Smart Grid Devices organized as systems 15
Key Points regarding Industrial-based IoT (I) o System-wide view rather than individual endpoint view required e.g. Smart Traffic Light System (STLS), will talk about it o Machine!= Human behavior M2M chatter o The Internet meets the physical world Deterministic/predictable latency closed-loops New (physical) dimension of threats o Support for rapid mobility Connected Vehicle (CV), Connected Rail (CR) o Managing Large-scale Geo-distributed Systems pipelines, Smart Grid (SG) o From consumer-oriented to enterprise operational technologies Ecology of domain-expert partners needed 16
Key Points regarding Industrial-based IoT (II) o Orchestration and Policy Creation Networking Resource Orchestration Orchestration and Control Laws of Physical Devices o Security (-) Extended attack surface, new vectors (-) Potential for physical damage (Connected Car, insulin pump, etc.) (+) Limited set of interactions built-in behavioral detector o Data Management & Analytics Massive number of geo-distributed sources Move processing to the data Real Time (RT) or NRT analytics Data-in-Motion Interplay with the Cloud 17
Smart Traffic Light System 2013-2014 Cisco and/or its affiliates. All rights reserved. Cisco Confidential 18
Canonical Example: Smart Traffic Light System (STLS) o Traffic Light at Intersection Goal: accident prevention - Detects pedestrians/cyclist crossing - Measures distance & speed of approaching vehicles - Collision likely? - Issues alarms to approaching vehicles, changes from Green to Red, takes photos & Issues ticket o STLS as a system Goal: facilitate traffic flow throughout city/region - Traffic congestion maps route recommendations (no loops!) - Cycle coordination of individual intersections green waves - Faster clearing of traffic accidents (automated dispatch, rerouting, ) Goal: Safety - Green wave for emergency vehicles, emergency evacuation routes Goal: Security - Coordinated surveillance cameras (Amber alerts, etc.) 19
Requirements Derived from STLS Local subsystem (traffic light, sensors, actuators) - Reaction time < 10 msec - Compute/storage capabilities; ruggedized small form factor box Global system - Wide geo-distribution - Middleware orchestration - Multiplicity of agencies running the system (must coordinate control policies) Interaction with Cloud/DC - Efficient traffic management demands Data base of historical records Near real time data on utilities work, street repairs, garbage collection Mobility Geodistribution Enter Fog COMPUTING Low/predictab le latency Multi-agent orchestration Semi-autonomy Big Data & Analytics RT Analytics at the Edge Interplay with Cloud Support for Service Exchange 20
Requirements Derived from STLS Local subsystem (traffic light, sensors, actuators) - Reaction time < 10 msec - Compute/storage capability - Ruggedized & small form factor box Global system - Wide geo-distribution - Middleware orchestration - Multiplicity of agencies running the system (must coordinate control policies) Interaction with Cloud/DC - Efficient traffic management demands large data base of historical records Enter Fog COMPUTING 21
Windmill Farm 2013-2014 Cisco and/or its affiliates. All rights reserved. Cisco Confidential 22
A Perspective on Size Credit: Pao and Johnson 23
Region1: power < losses shut-off Region 2: normal operating Condition Region 3: shut-off for safety Credit: Pao and Johnson 24
Control of an Individual Turbine Credit: Pao and Johnson 25
Characteristics of the Windmill Farm System o o o o Tight interaction between sensors and actuators in closed-loop control loops (local controllers in the turbine) Wide geographical deployment of semi-autonomous modules (turbines) that required coordination (prevention of wind starvation of turbines in the rear) Strong interplay with the Cloud Real-time analytics at the edge Batch analytics in the Cloud Business models in the Cloud Multiplicity of time scales Long-term planning (BI, Cloud) Daily negotiation with Independent System Operators (ISOs) (bidding, commit) Hourly renegotiations to adjust commitments to operating conditions 5 min interval optimization (farm level) Local real time control of individual turbines 26
weather records CLOUD weather data daily weather forecast negotiation with ISO hourly forecast FOG farm optimization turbines output Adjustment/ren egotiation 27
Smart & Connected Communities (S+CC) 2013-2014 Cisco and/or its affiliates. All rights reserved. Cisco Confidential 28
A City as a Living Organism smart buildings emergency services smart grid fire department transportation STLS recycling and garbage collection utility services environmental monitoring police hospitals parks irrigation Multiplicity of agents Interactions & dependencies Missed opportunities for lack of sharing 29
Consolidation of Silos into a Coherent Infrastructure First phase: Reduce of CAPEX, OPEX, and cluttering Second phase: Enable controlled sharing of information between agencies Third phase: Facilitate service creation government agencies, providers of services, citizens 30
Closing Message IoT/IoE is disruptive Yes, the explosion in connectivity is an issue, but note With IoT the Internet meets the physical world (mind the actuators) There are two complementary IoT app spaces: consumer and enterprise In enterprise-based IoT endpoints are organized and managed as integrated system A pure Cloud play does not cut it in many use cases, including the ones outlined Fog is the vehicle for enterprise/industrial-based IoT 31
Thank you.