1 Fog Computing: A Platform for Internet of Things and Analytics
Flavio Bonomi, Rodolfo Milito, Preethi Natarajan, and Jiang Zhu Presented by Andrew Levandoski
2 Motivation Use Cases and Requirements Fog Architecture New Research
Outline Motivation Use Cases and Requirements Fog Architecture New Research
4 Cloud Computing Cloud is efficient, inexpensive, and frees the enterprise and end user from the specification of details. However, latency-sensitive applications like IoT deployments require location awareness and mobility.
5 Why Fog? Fog was conceived to address applications and services that do not fit with Cloud. IoT motivates the need for a hierarchical platform that extends from the edge to the core of the network. This is useful for: Applications that require low and predictable latency Geo-distributed applications Fast mobile applications Large-scale distributed control systems
6 Fog + Cloud Same resources networking, computation, storage
Shared mechanisms virtualization, multi-tenancy Fog does not substitute the Cloud, it compliments it!
7 Business Intelligence HMI
Days to months Business Intelligence HMI Key performance indicators, dashboards, reports Very high latency Business data repository Cloud Enterprise Technical Minutes to Days Transactional analytics HMI, M2M Visualization, reporting, systems, and processes Historical data Seconds to sub-minutes Medium speed/medium latency real-time analytics HMI, M2M Visualization systems and processes Operational and non-operation data Fog Milliseconds to sub-seconds High speed/low latency real-time analytics M2M Protection and control systems Very low latency Grid sensors and devices HMI = Human-Machine Interaction M2M = Machine to Machine
8 Geo-Distribution Analyzing data close to the device that collected the data can make the difference between averting disaster and a cascading system failure. The main requirements to support IoT devices using Fog are to: Minimize latency Conserve network bandwidth Address security concerns Operate reliably Collect and secure data across a wide geographic area with different environmental conditions Move data to the best place for processing
10 Smart Traffic Light System (STLS)
Sensors Measure distance and speed of approaching vehicles Detect presence of pedestrians and cyclists at crossings Goals Prevent accidents Maintain a steady flow of traffic Collect relevant data to evaluate and improve the system Smart Traffic Light System (STLS)
11 STLS Requirements Local subsystem latency
Middleware orchestration platform Networking Infrastructure Interplay with the Cloud Consistency of a distributed system Multi-tenancy Multiplicity of providers Decision Maker Federated message bus
12 Wind Farm Characteristics
Tight interaction between sensors and actuators in closed control loops Controllers to tune yaw and pitch of windmill blades Goals Improve wind power capture and power quality Reduce structural loading Forecast wind accurately Wind Farm
13 Wind Farm Requirements
Networking Infrastructure Global Controller Middleware orchestration platform Data Analytics
15 Heterogeneous Physical Resources
The Fog architecture should facilitate seamless resource management across a diverse set of platforms. Fog Nodes can be high-end servers, edge routers, access points, set-top boxes, vehicles, sensors, mobile phones, etc. All of these platforms have varying levels of resources and run different operating systems and applications.
16 Fog Abstraction Layer The abstraction layer provides generic APIs for monitoring, provisioning, and controlling CPU, memory, network, and energy resources. The following multi-tenancy features must be supported: Data and resource isolation A single, consistent model across physical machines Ability to expose the physical and logical network to administrators
17 Fog Service Orchestration Layer
The service orchestration layer provides dynamic, policy-based life cycle management of Fog services. Management is achieved with the following components: A software agent capable of bearing the orchestration functionality Distributed storage to store policies and resource metadata capable of supporting high transaction rates A scalable messaging bus A distributed policy engine with a single global view and local enforcement
18 Policy Manager Service Directory Policy DB No Life Cycle Manager
Retrieve relevant policies: Performance Security Capability Do service instances satisfy policy constraints? Policy DB No Life Cycle Manager Capability Engine Capabilities DB Provision service Find platforms that are capable of offering service
20 Application Provisioning
Paper Contribution Hong et al. Propose a high-level programming model for IoT applications provisioning. Giang et al. Propose a DDF programming model for IoT applications provisioning in which the application topology is expressed as a directed graph. Yangui et al. Propose a layer-based architecture for IoT application provisioning that spans Cloud and Fog.
21 Resource Management Paper Contribution Bittencourt et al.
Propose a layer-based architecture that support VM migration between fog nodes. Agarwal et al. Propose an architecture for resource allocation that includes an algorithm that distributes the workload between the Cloud and Fog. Cardellini et al. Propose an extension to Store in order to execute a distributed QoS-aware scheduler. Kapsalis et al. Propose an architecture for resource management and load balancing between the Cloud and Fog.
22 Communication Paper Contribution Shi et al.
Study the inter-stratum communication between IoT/end-users stratum and Fog stratum. Slabicki et al. Study the intra-stratum communication between devices in IoT/end-users stratum. Krishnan et al. Propose an architecture composed of the Cloud and fog and a method to move computation from the Cloud to the Fog. Aazam et al. Propose an architecture for fog computing co-located within a gateway. Moreno-Vozmediano et al. Propose an architecture to interconnect geographically distributed Cloud and Fog domains.
23 Resource Sharing Paper Contribution Abedin et al.
Propose an algorithm to enable resource sharing among Fog nodes. Oueis et al. Propose an algorithm to cluster small cells to enable resource sharing among them. Nishio et al. Present a strategy to optimize the sharing of resources with the objective of maximizing the corresponding utility.
24 Task Scheduling Paper Contribution Oueis et al.
Introduce an algorithm to manage the execution of tasks in small-cell Fog stratum. Intharawijitr et al. Present three policies to select Fog nodes to execute tasks. Aazam et al. Propose a loyalty-based task scheduling algorithm. Zeng et al. Introduce a task scheduling and image placement algorithm that aims at minimizing the overall completion time. Agarwal et al. Introduce an algorithm to distribute workload to reduce response time and cost. Deng et al. Introduce an algorithm to distribute workload in Cloud/Fog systems at lowest power cost.
25 Response Time Reduction Power Consumption Reduction
A Fog server manager receives requests and is responsible for matching resources with demands. Depending on availability, it can: Execute all tasks Execute some tasks and delay others Transfer demand to cloud nodes Lowest existing response time in survey! Agarwal et al. In contrast, these authors account for: Computational capabilities Communication bandwidth limitations Delay constraints on user’s side By sacrificing modest computational resources to save communication bandwidth and reduce transmission latency, the use of Fog can significantly improve the performance of Cloud computing. Deng et al.
26 Adaptive Solutions The authors propose a proactive resource allocation algorithm incorporating historical data and attributes. Users loyal to the requested service/service provider receive higher QoS! Further, they complement this allocation strategy with a discriminative pricing scheme based on user loyalty. Aazam et al.
27 Offloading and Load Redistribution
Paper Contribution Hassan et al. Propose a strategy to offload applications on mobile devices. Ye et al. Propose a strategy to offload cloudlets and mobile devices to Fog-enabled buses. Fricker et al. Evaluate the performance of a data center offloading strategy in the Fog layer. Ningning et al. Introduce a dynamic load balancing algorithm in the Fog layer that allows coping with the dynamic arrival and exit of Fog nodes. Li et al. Propose a coding framework to handle redundancy in tasks computation in Fog computing. Ottenwalder et al. Propose a strategy that allows migrating operators at minimal migration cost.
28 Dynamic Load Balancing
The authors use dynamic graph balancing to repartition the load of their Fog system model. The strategy presented has additional overhead but outperforms previously presented hybrid load balancing strategies. Ningning et al.
29 Healthcare. Elderly care (Stantchev et al.)
Support for COPD (Fratu et al., Masip-Bruin et al.) Parkinson’s disease (Monteiro et al.) Speech disorders (Dubey et al.) ECG and EEG feature extraction (Gia et al., Zao et al.)
30 . and More Use of cellular networks and roadside units (RSUs)
Vehicular ad hoc networks (VANETs) Vehicular fog computing (VFC) Vehicular applications Smart living Smart grids Smart levee monitoring Smart city infrastructure Emergency alert management
31 Pricing, Pay-as-you-go, and Blockchain
Business models that determine levels of responsibility for users and agencies and tiers of compensation for services have yet to be fully discussed. A BC-based smart contract system lends itself to the geo-distributed mobile nature of Fog computing.
32 Thank you! Questions?
33 References F. Bonomi, R. Milito, P. Natarajan, and J. Zhu, “Fog computing: A platform for Internet of Things and analytics,” in Big Data and Internet of Things: A Roadmap for Smart Environments. Cham, Switzerland: Springer, Mar. 2014, pp. 169–186. C. Mouradian et al. “A Comprehensive Survey on Fog Computing: State-of-the-art and Research Challenges,” in IEEE Surveys and Tutorials