Source-linked AI summary
Fog Computing: Principles, Architectures, and Applications
Amir Vahid Dastjerdi, Harshit Gupta, Rodrigo N. Calheiros, Soumya K. Ghosh, Rajkumar Buyya
TL;DR
The chapter addresses the limits of centralized cloud processing for IoE and IoT applications requiring low latency and handling high data volumes. It presents fog computing as an edge extension of cloud services, develops a reference architecture, and surveys applications and developments. The chapter reports reduced network usage and end-to-end latency when operators are placed across fog devices with sufficient capacity, while noting continuing security, resource, and energy issues.
Problem
Centralized cloud processing does not scale to IoE requirements and can violate latency and privacy constraints for applications such as health monitoring, emergency response, and high-rate IoT flows.
Method
The chapter defines fog computing, presents a reference architecture, and discusses its characteristics, applications, and related developments.
Results
Fog processing can reduce network usage and end-to-end latency relative to in-cloud processing when operators are placed across fog devices with enough capacity.
Takeaways & Limitations
Fog computing extends cloud services toward edge devices to support low-latency processing and reduce data sent through the core network.
Takeaways & Limitations
Fog computing still faces issues involving security, resource usage, and energy usage.
Abstract
from arXiv · showhide
The Internet of Everything (IoE) solutions gradually bring every object online, and processing data in centralized cloud does not scale to requirements of such environment. This is because, there are applications such as health monitoring and emergency response that require low latency and delay caused by transferring data to the cloud and then back to the application can seriously impact the performance. To this end, Fog computing has emerged, where cloud computing is extended to the edge of the network to decrease the latency and network congestion. Fog computing is a paradigm for managing a highly distributed and possibly virtualized environment that provides compute and network services between sensors and cloud data centers. This chapter provides background and motivations on emergence of Fog computing and defines its key characteristics. In addition, a reference architecture for Fog computing is presented and recent related development and applications are discussed.
4.1 Introduction
IoT environments generate distributed, continuously captured data that centralized cloud processing can scale but cannot always process with indispensable low latency. Real-time analytics and decision making therefore expose limits in-cloud processing must address.
- 4.1 Introduction: IoT environments include loosely connected devices and heterogeneous networks that send aggregated data uplink to cloud servers.The described arrangement centralizes storage and processing after collection from smart devices.
- 4.1 Introduction: IoT environments collect transient streams and persistent data that support real-time analytics and decision making.The passages distinguish continuously captured “Little Data or Big Stream” from persistent data and knowledge stored centrally.
- 4.1 Introduction: Cloud computing provides scalable, on-demand infrastructure for storing and processing large volumes of centralized data.Cloud platforms can be configured on demand and used in a pay-per-use mode for Big Data processing.
- 4.1 Introduction: When data sources are distributed across multiple locations and low latency is indispensable, in-cloud data processing fails to meet application requirements.The limitation concerns collecting and processing IoT data for pattern mining, predictive analysis, optimization, and timely decisions.
4.2 Motivation Scenario
The Endomondo scenario illustrates why centralized cloud processing is inadequate for high-rate, privacy-sensitive activity data. Fog computing addresses this motivation by bringing computation closer to geographically distributed sensors and using edge devices to form local data views.
- 4.2 Motivation Scenario: 25,000 tuples per second is the reported peak data flow generated by Endomondo’s 30 million users.A single workout generates 170 GPS tuples, and monthly GPS tuples can reach 6.3 million.
- 4.2 Motivation Scenario: Real-time city data sources can generate millions of tuples per second, extending the Endomondo-scale motivation to smart-city environments.Examples include GPS sensors in cars and air- and noise-pollution sensors.
- 4.2 Motivation Scenario: Centralized cloud servers cannot process flows with this velocity in real time, while users also express privacy concerns about cloud storage.The figure describes centralized processing as failing users’ latency and privacy constraints.
- 4.2 Motivation Scenario: Fog computing brings computation to capable devices geographically closer to sensors and connected to the Internet.These edge devices can build local views of data flows and aggregate data for later cloud analysis.
4.3 Definitions and Characteristics
Fog computing extends cloud services to the network edge, placing distributed compute, networking, and storage between sensors and cloud data centers. Its characteristics target reduced traffic, local IoT processing, low latency, and scalability.
- 4.3 Definitions and Characteristics: Fog computing is a distributed paradigm that extends cloud services to the network edge.Its components may run across cloud and edge devices, including smart gateways, routers, and dedicated fog devices.
- 4.3 Definitions and Characteristics: Fog provides compute, networking, and storage services between sensors and the cloud, while supporting mobility, heterogeneous interfaces, protocols, and cloud integration.It also supports distributed data analytics for applications requiring low latency across wide, dense geographical areas.
- 4.3 Definitions and Characteristics: Fog reduces cloud traffic by filtering and analyzing sensor data near its source and generating local data views before cloud transfer.The passages describe this as reducing the traffic sent to the cloud.
- 4.3 Definitions and Characteristics: Local fog processing suits IoT queries about nearby surroundings, such as locating people playing a similar sport nearby.The passages state that such requests can be served without global information from the cloud.
- 4.3 Definitions and Characteristics: Fog brings communication distance closer to physical distance by moving processing toward the network edge.This supports applications that need local processing and shorter communication paths.
- 4.3 Definitions and Characteristics: Fog supports real-time responses for control systems by processing required data close to robots.The motivation includes cloud robotics, fly-by-wire aircraft, and vehicle anti-lock brakes.
- 4.3 Definitions and Characteristics: Processing incoming data closer to its source reduces the cloud’s processing burden and addresses scalability issues from increasing endpoints.The scalability rationale concerns avoiding continued transmission of all raw end-device data to the cloud.
4.4 Reference Architecture
The reference architecture organizes fog computing from sensors and edge devices through network and cloud resources, with software-defined services managing applications and resource use.
- Layered architecture: The architecture’s bottom layer contains sensors, edge devices, and gateways, which communicate through the network with cloud services and resources.Cloud resources support processing of IoT tasks, while resource-management software manages the infrastructure and enables quality of service.
- Layered architecture: The application layer contains applications that leverage fog computing to deliver services to end users.
- Resource management: Software-defined resource management provides middleware-like services that optimize cloud and fog resources by pushing task execution to fog nodes.Its goal is to reduce cloud-use cost while maintaining acceptable application latency.
- Resource management: Flow and task placement tracks resource states and identifies candidates for incoming tasks, while monitoring can trigger new allocation rounds when demand is high.
- Supporting services: Knowledge Base, Performance Prediction, Raw Data Management, Monitoring, Profiling, Resource Provisioning, and Security provide decision support, data views, dynamic allocation, and protection.Resource provisioning uses profiles, performance predictions, monitoring, user latency requirements, and security credentials; low-latency tasks are pushed to the network edge when resources are available.
- Scope: The listed architecture elements are reference elements, so complete fog stacks may omit some or add services not shown in the figure.
4.5 Applications
Fog computing supports applications requiring responsive, distributed processing by placing latency-critical work near users or devices and using cloud resources for complementary tasks.
- Application range: Fog applications span healthcare, augmented reality, web performance, IoT-cloud integration, and distributed analytics.The chapter presents these application areas as examples benefiting from fog computing.
- Distributed analytics: Fog-assisted analytics performs latency-critical analysis at the edge and latency-tolerant computation in the cloud.
- Healthcare: Healthcare architectures use fog layers to provide low latency, mobility support, location awareness, and security for healthcare and elderly-care applications.One architecture models process flows with BPMN and maps them to devices through a service-oriented approach.
- Augmented reality: Combining fog and cloud servers enables continuous real-time brain-state classification from EEG streams, while models are tuned in the cloud.
- Wearable assistance: Wearable cognitive-assistance workloads require external computation with crisp real-time responses, but cloud offloading incurs considerable latency.Nearby devices handle latency-sensitive work, while the cloud can serve delay-tolerant jobs such as error reporting and logging.
- Web performance: Edge devices improve web-page response times through caching, page reorganization, object-size reduction, and adaptation to user behavior and network conditions.Under congestion, they may provide lower-resolution graphics to reach acceptable response times.
- IoT-cloud integration: Smart gateways preprocess IoT data before cloud transmission and use fog services for latency-sensitive, context-aware operations.The approach is presented as supporting richer user experiences for IoT applications.
4.6 Research Directions and Enablers
Research directions address how fog systems can realize low-latency, scalable services while managing security, reliability, resources, energy, and computation offloading.
- Challenges: Researchers identify computation offloading, energy use, security, reliability, resource management, and policy management as major fog-computing challenges.
- Offloading platforms: Cloudlets provide nearby, discoverable, VM-based infrastructure that supports computation offloading, data staging, disconnected operation, flexibility, mobility, scalability, and elasticity.Provisioning policies affect payload sent, application-ready time, and client energy spent.
- Development enablers: Mobile Fog provides an API for applications leveraging fog infrastructure’s geo-distribution and low-latency characteristics through components distributed across device hierarchy levels.
- Security: Security work focuses on authentication across fog-node levels, with PKI, trusted execution environments, and measurement-based rogue-device detection identified as potential solutions.
- Security: Policy-driven security management is proposed to support secure sharing and data reuse in heterogeneous fog environments, with a smart-transportation use case illustrating the approach’s efficiency.
- Reliability: Reliable fog systems must plan for failures of individual sensors, networks, service platforms, and applications; existing wireless-sensor-network reliability protocols address packet and event reliability.
- Resource management: Fog devices have less resource capacity than traditional servers and cloud systems, making judicious resource management essential.A service-oriented framework predicts customer usage and pre-allocates resources based on behavior and future-use probability to improve fairness and efficiency.
- Energy and delay: Fog computation can reduce communication latency while incurring slightly greater energy consumption, highlighting a power-delay trade-off.Related work models power and delay and allocates workloads between fog and cloud systems.
4.7 Commercial Products
Commercial fog-computing products place software and analytics capabilities across edge devices, gateways, and cloud systems. They emphasize interoperability, distributed processing, real-time decisions, and controlled data exchange.
- Cisco IOx: Cisco IOx combines IOS and Linux to support applications and device abstractions on fog-enabled devices.The platform includes an app store and app-management capabilities for IOx devices.
- Industry deployments: Commercial deployments include industrial asset management and plant-wide real-time event and data management using Cisco IOx.Rockwell’s FactoryTalk AssetCentre and OSIsoft’s PI system are cited as examples.
- Cisco Data In Motion: Cisco Data In Motion provides edge data management, rule-based RESTful APIs, and runtime rule updates without downtime.It analyzes incoming data by finding relevant information, summarizing data, and generating new results in distributed deployments.
- LocalGrid: LocalGrid standardizes and secures communication across heterogeneous devices, minimizing customization and services costs.Its platform supports reliable device-to-device communication without routing through the cloud, enabling real-time edge decisions with lower cloud-communication latency.
- ParStream: ParStream provides distributed, real-time IoT analytics by pushing query execution to the edge through a compact in-memory database.Its small footprint supports deployment on embedded and fog-enabled devices with limited disk space.
- VORTEX: VORTEX enables secure, interoperable data sharing across IoT devices, edges, gateways, and cloud systems using DDS 2.0 and common IoT protocols.Its fog-domain deployment models include Fog + Cloud, Fog + Cloud-link + Cloud, and Federated Fog, with Cloud-link controlling inter-domain information exchange.
4.8 Case Study
The smart traffic case study evaluates fog-based stream processing for incident detection and finds lower response delay and bandwidth consumption than cloud processing when operators are placed near data sources.
- Fog computing improves smart traffic application performance in terms of response time and bandwidth consumption.
- Query and data sources: The study simulates the DETECT_TRAFFIC_INCIDENT query on fog-based and traditional cloud-based stream-processing engines.
- Query and data sources: The incident-detection query computes average vehicle speed, lane congestion, and whether an incident has occurred.
- Performance evaluation: The fog engine dynamically places operators across fog devices when capacity is available to save bandwidth and minimize latency.
- Performance evaluation: Once operators are placed on fog devices, end-to-end tuple delay falls below in-cloud processing delay because data are processed closer to sources.
- Performance evaluation: Running the query on edge devices reduces tuples reaching the core network and lowers cloud workload and network usage considerably.
- Performance evaluation: These reductions require a placement algorithm to push operators downward when edge devices have enough capacity; otherwise, resource contention can increase delay.
4.9 Summary
Fog computing addresses IoT data-processing demands by using capable devices near the network edge, while the chapter presents a reference architecture and identifies unresolved challenges.
- Fog computing uses edge devices with more processing power than end devices and closer proximity than cloud resources to reduce application latency.
- The chapter introduces a reference architecture for IoT and discusses ongoing academic and industry efforts supporting the fog-computing vision.
- Security, resource usage, and energy-usage minimization remain challenges requiring solutions.
- Open protocols and architectures are identified as future-research topics that could make fog computing more attractive to end users.