PaperlayerView exact revision on arXiv

Source-linked AI summary

All One Needs to Know about Fog Computing and Related Edge Computing Paradigms: A Complete Survey

Ashkan Yousefpour, Caleb Fung, Tam Nguyen, Krishna Kadiyala, Fatemeh Jalali, Amirreza Niakanlahiji, Jian Kong, Jason P. Jue

arXiv:1808.05283v3cs.NI

TL;DR

The paper addresses how growing IoT deployments produce security-critical and time-sensitive data that motivates computing closer to connected devices. It provides a tutorial and comparison of fog and related edge paradigms, develops a research taxonomy, surveys the literature, and identifies challenges and future directions. The survey concludes that fog computing is broadly applicable and more general than several related paradigms, while some extreme settings may favor alternatives.

  • Problem

    Rapid IoT growth generates large volumes of security-critical and time-sensitive data that cloud-centered computing may not handle efficiently under stringent bandwidth, latency, and connectivity requirements.

  • Method

    The paper combines a tutorial comparing fog with related paradigms, an exhaustive taxonomy, a comprehensive categorized literature survey, and a review of challenges and future directions.

  • Results

    The survey concludes that fog computing suits many data-driven and low-latency IoT use cases and is more general than edge computing, MEC, and cloudlets, while some extreme settings may favor other paradigms.

  • Takeaways & Limitations

    Fog computing offers a versatile computing model across the thing-to-cloud continuum, but paradigm choice should depend on the use case and network setting.

  • Takeaways & Limitations

    Fog nodes are more resource-constrained than adequately secured cloud data centers, making them more prone to denial-of-service attacks.

Abstract

from arXiv · show

With the Internet of Things (IoT) becoming part of our daily life and our environment, we expect rapid growth in the number of connected devices. IoT is expected to connect billions of devices and humans to bring promising advantages for us. With this growth, fog computing, along with its related edge computing paradigms, such as multi-access edge computing (MEC) and cloudlet, are seen as promising solutions for handling the large volume of security-critical and time-sensitive data that is being produced by the IoT. In this paper, we first provide a tutorial on fog computing and its related computing paradigms, including their similarities and differences. Next, we provide a taxonomy of research topics in fog computing, and through a comprehensive survey, we summarize and categorize the efforts on fog computing and its related computing paradigms. Finally, we provide challenges and future directions for research in fog computing.

1 INTRODUCTION

The growth of IoT data and connected devices strains cloud-centered architectures, especially for bandwidth-intensive, geographically dispersed, ultra-low-latency, location-aware, and privacy-sensitive applications. The survey introduces fog computing as a computing paradigm closer to IoT devices and maps its related paradigms, research topics, and future challenges.

  • Motivation: IoT growth is producing data volumes that challenge current mobile-network and cloud-centered architectures.The passage cites more than 1 zettabyte of digital data in 2010, 2.5 exabytes generated daily since 2012, and an estimate of 50 billion connected devices by 2020.
  • Motivation: Moving increasing IoT data to distant clouds can be inefficient or infeasible because of bandwidth constraints and strict application requirements.Time-sensitive and location-aware applications such as patient monitoring, real-time manufacturing, self-driving cars, drones, and cognitive assistance require capabilities the distant cloud may not satisfy.
  • Fog Computing: Fog computing places computing, storage, networking, decision making, and data management on network nodes near IoT devices and along the path to the cloud.It is presented as a bridge between cloud infrastructure and IoT devices.
  • Fog Computing: The survey compares fog computing with edge, mist, cloud-of-things, cloudlet, and other related paradigms, arguing that fog is more general because of its broad scope and flexibility.The comparison addresses similarities and differences among computing paradigms proposed for the same application pressures.
  • Survey Scope: The paper provides a tutorial, a taxonomy of fog-computing research topics, a comprehensive categorized survey, and challenges and future research directions.Figure 1 is described as the survey’s structure and reading map.

2 RELATED SURVEYS

Existing surveys cover focused aspects of fog, edge, and MEC computing, including architectures, algorithms, connectivity, device configuration, management, and enabling technologies. This paper distinguishes itself through an integrated tutorial, exhaustive taxonomy and survey, and consolidated challenges and future directions.

  • Existing Surveys: Prior fog-computing surveys emphasize architectures, algorithms, emerging technologies, and prospects such as the Tactile Internet.
  • Existing Surveys: Other surveys examine edge-computing cooperation, software-defined networking, connectivity, device configuration, infrastructure for smart cities, architecture design, or system management.
  • Existing Surveys: MEC surveys cover enabling technologies, reference architecture, standardization activities, deployment scenarios, and recent MEC research.
  • This Paper’s Contributions: This paper contributes a detailed tutorial explaining fog computing and its relationships with cloud computing, cloudlets, edge computing, and MEC.
  • This Paper’s Contributions: It proposes an exhaustive taxonomy, presents a comprehensive fog-computing survey, and compiles challenges and future research directions.

3 A COMPARISON OF FOG COMPUTING AND RELATED COMPUTING PARADIGMS

This section compares fog computing with cloud computing and related paradigms, emphasizing how their placement, capabilities, and trade-offs suit different connected-device use cases.

  • Fog Computing: Fog computing extends cloud capabilities along the IoT-to-cloud path by placing computing, storage, networking, and data management near IoT devices.This placement can support local processing such as compressing GPS data before transmission to the cloud.
  • Cloud Computing: Cloud computing provides scalable, remotely accessed infrastructure, platforms, and software, but access latency can be too high for mission-critical or ultra-low-latency applications.Cloud provisioning dynamically allocates resources according to application demand, commonly under pay-as-you-go pricing.
  • Edge Computing: Edge computing addresses privacy, latency, and connectivity, but its latency advantage depends on having sufficient local computation power.Compared with MACC, edge computing uses small data centers and can form hybrid architectures with peer-to-peer and cloud models.
  • Fog and Edge Computing: Fog and edge computing both move resources toward end-nodes, but fog is hierarchical and spans cloud-to-things functions, whereas edge computing is generally limited to edge computation.The paper treats the paradigms as related but non-identical.
  • Use-Case Suitability: Fog suits many data-driven and low-latency use cases, while ad hoc computing or extreme edge clouds may better fit disaster zones and sparse network topologies.The paper notes that paradigm strengths and weaknesses make some approaches better suited to particular use cases.
  • Comparative Scope: No globally unanimous distinction exists among fog computing, edge computing, mist computing, and cloudlets, motivating the survey’s comparative clarification.The section summarizes comparisons across fog computing and related paradigms using their attributes and features.

4 TAXONOMY OF FOG COMPUTING THROUGH A COMPLETE SURVEY

The survey organizes fog-computing research into a taxonomy spanning foundations, architectures, resource management, operations, software, experiments, and security or privacy. It uses this taxonomy to categorize fog and relevant related-paradigm research.

  • Taxonomy scope: The taxonomy categorizes fog-computing research primarily from the networking perspective and includes relevant, generalizable work from related paradigms such as edge computing.The survey’s taxonomy is shown in Figure 8, with referenced papers categorized in Table 4.
  • Research categories: Foundation papers survey or define the fog-computing field, while frameworks and programming-model papers introduce architectures, frameworks, or fog-based concepts.The framework category includes examples such as vehicular fog computing.
  • Research categories: Resource-management and provisioning papers study service provisioning, VM placement, control, and monitoring, whereas operation papers address task scheduling, load balancing, and resource discovery.These categories distinguish resource preparation from ongoing system operation.
  • Research categories: Software and tools, testbeds and experiments, and security and privacy form additional categories for implementation resources, empirical studies, and protection concerns.Testbed papers focus on developing testbeds or conducting extensive experiments.
  • Foundations: Fog-computing definitions and fog-node meanings remain unsettled, despite early definitions and standardization efforts such as OpenFog and IEEE 1934.OpenFog emphasizes security, scalability, openness, and agility as pillars of an open fog architecture.
  • Foundations: The survey also covers communication technologies, fog-service QoS classes, distributed deep learning, content islands, node reliability, and fogonomics.These topics extend the taxonomy beyond architectures and infrastructure management.

4.2 Foundations: Surveys

The survey situates fog computing among earlier and related paradigms, including edge, cloudlet, MEC, and hybrid architectures. Prior work addresses architectures, applications, resource management, privacy, and infrastructure planning, while the survey compares these paradigms systematically.

  • Related surveys: Earlier surveys cover fog architectures and algorithms, edge-computing development and use cases, privacy and security, and key technologies or deployment characteristics.The reviewed application areas include gaming, real-time image processing, smart grids, and smart transportation.
  • Architectures and frameworks: Fog and edge architectures are described through decentralized, hierarchical, and cloud-connected designs, reflecting different ways to place computation across the network.One cited three-layer design separates IoT, fog, and cloud domains.
  • Architectures and frameworks: Research proposes combining MEC and fog computing for 5G, alongside architectures integrating cloud, MEC, IoT, and heterogeneous cloud-fog layers.The surveyed paper also presents its own three-layer fog architecture.
  • Resource providers: Fog-resource provisioning remains an open question involving service providers, network providers, and end users that may share consumer devices or edge resources.Indie Fog is cited as using consumer-premises equipment to provide fog services.
  • Architectures and frameworks: Related architectures extend fog toward 5G, information-centric networking, hierarchical cloudlets, unified fog-cloud management, and telecom-edge deployment.Field, shallow, and deep cloudlets are positioned at progressively higher network locations to absorb peak loads.

4.4 Frameworks and Programming Models: Concepts and Frameworks using Fog

Fog-based concepts and frameworks adapt computation, storage, and service delivery to specialized settings such as vehicles, UAVs, V2G systems, IoT devices, and mobile users. These proposals emphasize distributed resources, local processing, coverage, and service flexibility.

  • Vehicular and aerial fog: Vehicular fog computing uses vehicles as communication and computation infrastructure, supporting geodistribution, local decision making, and real-time load balancing.Its proposed architecture has application-and-services, policy-management, and abstraction layers.
  • Vehicular and aerial fog: UAVs can act as fog nodes to provide computing capabilities and enhanced coverage for IoT nodes, while vehicular micro clouds aggregate and preprocess data.Vehicular micro clouds are virtual edge servers formed by clusters of cars.
  • Hybrid fog concepts: The Foud model combines cloud and fog resources for vehicle-to-grid services, with cloud resources providing virtualization and fog integrating stationary and mobile edge resources.The fog component temporarily expands computing capacity at the V2G-network edge.
  • IoT-oriented concepts: Fog of Things defines IoT services at the network edge and distributes them through message- and service-oriented middleware in a self-organized architecture.The paradigm includes FoT devices, gateways, and servers.
  • IoT-oriented concepts: Human-driven edge computing uses devices carried by people to ease provisioning and extend the coverage of fixed MEC solutions.This approach is presented as a new model for extending traditional MEC coverage.
  • Additional concepts: Other proposals include mobile IoT federation, transparent computing for scalable cross-platform IoT services, volunteer edge resources, path computing, and IoT Hubs for heterogeneous networks.IoT Hubs bridge different physical networks and merge them through an all-IP network.

4.5 Frameworks and Programming Models: Programming Models and Data Modeling

Fog programming and data-modeling research addresses distributed data sharing, workflow placement, deployment annotation, QoS-aware provisioning, and application development across IoT-fog-cloud environments. Related network and capacity-planning studies connect these programming concerns to deployment constraints.

  • Programming models and data modeling: Firework enables distributed data sharing by creating virtually shared data views across geographically distributed stakeholders while addressing resource, security, and privacy concerns.The model is designed for distributed big-data analytics in IoT settings.
  • Programming models and data modeling: WM-FOG lets developers customize workflow synchronization and choose how much data is sent to the back-end cloud versus fog nodes.The model directly exposes workload-distribution choices to application developers.
  • Application deployment: Process-aware deployment models identify application fragments for different edge nodes and annotate them with deployment locations.Users can define a distributed IoT application in one place before assigning components to locations.
  • Application deployment: QoS-aware programming supports application deployment in IoT-fog-cloud infrastructure, while Foglets provides spatio-temporal abstractions and container-based distributed programming across fog nodes.Foglets processes are associated with geospatial regions and manage components on fog nodes.
  • Platforms and tools: Fog programming research also targets smartphone-based fog stacks, dynamic-scaling PaaS models, and hybrid fog-cloud provisioning architectures.The hybrid PaaS architecture supports development, deployment, and management phases.
  • Infrastructure planning: Network-design studies address cloudlet planning, computing-versus-communication tradeoffs, SDN and virtualization, and edge-data-center placement and capacity allocation.Capacity planning considers cost-effectiveness together with application bandwidth and performance requirements.

4.7 Design and Planning: Resource Analysis and Estimation

This section surveys resource analysis, pricing, energy estimation, elasticity, migration, orchestration, and lightweight deployment for fog and edge environments.

  • Resource pricing and estimation: Fog resource studies address service pricing, infrastructure utility, and estimation for providers serving customers near their locations.Service providers seek proximity to customers, while infrastructure providers rent edge resources to service and content providers.
  • Energy estimation: Fog and cloud energy efficiency depends on workload: fog is favored for little computation, whereas cloud is more efficient for high data processing under centralized-grid power.
  • Resource elasticity: Fog resources must scale dynamically because heterogeneous IoT networks vary spatially and temporally; analytical models scale allocations with incoming workloads.
  • Migration and replication: Service migration and replication studies use network state, server response time, and proactive neighboring replicas to accommodate mobility and service performance.
  • Orchestration and deployment: Fog orchestration research targets reliability, scalability, security, automated deployment, QoS-aware scheduling, and lightweight virtualization on constrained IoT devices.Frameworks include Foggy, service-oriented middleware, Docker-based provisioning, FADES unikernels, and PiCasso orchestration.

4.9 Resource Management and Provisioning: Placement (VM and Service)

This section surveys placement methods for services, virtual machines, network functions, containers, caches, and cloudlets under QoS, cost, availability, storage, and traffic constraints.

  • Service placement: Service placement is formulated using user association, task distribution, VM placement, QoS constraints, and multi-component application requirements.
  • Service placement: Optimization-based placement seeks to maximize fog utilization while satisfying QoS and potentially incorporating resource availability, reliability, and cost.
  • VNF placement: VNF placement determines resources and locations across two-tier edge-cloud infrastructures while considering service-level agreement requirements.
  • VM placement: VM placement in fog radio access networks minimizes back-haul traffic from VM replication and cloud data transmission.
  • Caching and cloudlets: Caching and cloudlet-placement studies target retrieval or download delay while respecting cache coherence, fog storage capacity, and deployment cost.

4.10 Resource Management and Provisioning: Control and Monitoring

This section covers control, monitoring, scheduling, dispatching, offloading, load balancing, network formation, and privacy-aware management in fog networks.

  • Control and monitoring: SDN-based fog control and monitoring provides network-wide resource visibility and supports orchestration for large IoT data flows.
  • Control and monitoring: Fog SDN architectures combine centralized and distributed control, including hybrid control planes and wide-area coordination for edge-cloud management.
  • Virtualization: Lightweight container-based virtualization addresses the heavy footprint of conventional platforms and supports VNF deployment on edge devices.
  • Scheduling and offloading: Scheduling and offloading methods distribute tasks among local devices, neighboring fog nodes, and the cloud to reduce delay or balance load.
  • Scheduling and offloading: Offloading is not universally beneficial because wireless last-mile latency and fog-node bottlenecks can increase processing delay; collaborative edge resources provide an alternative.
  • Scheduling and offloading: Offloading frameworks also address dynamic fog formation, computation-time estimation, energy and communication overhead, privacy, and quality-of-results trade-offs.

4.12 Operation: Resource Discovery

This section surveys resource discovery and fog-enabled applications, including distributed lookup, surrogate brokering, stream analytics, transportation monitoring, healthcare, video, and image recognition.

  • Resource discovery: Resource discovery aims to provide location-independent access to IoT and fog resources through gateways, peer-to-peer overlays, and distributed hash tables.
  • Resource discovery: Context-aware and geographically distributed systems discover, virtualize, and pool IoT resources into cloud-like networks near the edge.
  • Resource discovery: A surrogate broker enables computation offloading by matching client requests with advertised micro-clouds, fog nodes, or cloudlets using network and hardware attributes.
  • Applications: Fog and edge applications process geo-distributed, heterogeneous, mobile, and high-volume streams, including real-time traffic video and federated weather-data queries.
  • Applications: Edge systems reduce data movement through secure deduplication, incremental document synchronization, and vehicle-based monitoring and mobility analytics.
  • Healthcare: Healthcare applications use fog nodes for low-bandwidth, low-latency ECG processing, hierarchical patient monitoring, and real-time brain-state classification.
  • Video and imaging: Edge platforms support low-latency video analytics, lightweight fog-based gaming, and image-recognition caching or fog-assisted resolution to reduce cloud-bound processing and traffic.

4.14 Software and Tools

This subsection surveys software, simulation, emulation, and platform implementations developed for fog and related edge computing environments.

  • Simulation and emulation: iFogSim extends CloudSim to simulate fog networks with cloud-only and edge-ward placement strategies.It supports modeling resource management, latency, throughput, and scalability of fog-based applications.
  • Simulation and emulation: EdgeCloudSim models both network and computational resources for modular, multi-tier edge-computing simulations.Its architecture supports a variety of critical functionality and builds on CloudSim.
  • Simulation and emulation: EmuFog provides an extensible framework for repeatable, controllable fog experiments using Docker applications on emulated network topologies.Users can design topologies and embed fog nodes connected through the emulated network.
  • Edge platforms: ParaDrop enables Docker-container services to run on WiFi access points at the extreme edge of the network.The platform is open source and includes documentation and tutorials.
  • Edge platforms: Other surveyed implementations include backend-based edge onloading, OpenStack-based Stack4Things, vendor-neutral EdgeX Foundry, OCI, EAaaS, DDNN, and cloudlet-based trusted identity systems.These efforts address service placement, IoT provisioning, edge analytics, distributed inference, and identity establishment in disconnected environments.

4.15 Testbeds and Experiments

The surveyed testbeds and experiments evaluate edge architectures, deployment tools, micro-clouds, containers, and latency-sensitive applications under practical resource and network conditions.

  • Network and infrastructure experiments: Measurement of LTE deployments provides realistic delay values for edge computing and examines where latency arises in the access network.The study specifically analyzes the first hop between user equipment and the base station.
  • Network and infrastructure experiments: Raspberry Pi clusters are used to test edge-based PaaS architectures designed for cost-efficiency and low power consumption.The experiments target resource-limited edge devices.
  • Network and infrastructure experiments: Micro-cloud experiments compare traditional cloud architectures with portable, low-cost infrastructures suitable for rougher deployment environments.Micro-clouds can be built with devices as small as a Raspberry Pi and are described as easier to deploy than mini data centers.
  • Orchestration and containers: Docker Swarm, Kubernetes, and Apache Marathon are evaluated against requirements for seamless fog-node changes and application scheduling.The surveyed evaluation identifies Docker Swarm as meeting all stated requirements.
  • Orchestration and containers: Docker is evaluated for edge hosting across deployment, resource management, fault tolerance, and caching using a data-center-plus-edge-sites testbed.The testbed simulates one data center and three edge sites.
  • Application experiments: Latency-sensitive cognitive assistance, serverless MEC, and mobile augmented-reality applications are examined through empirical and architectural studies.The serverless architecture offloads computation among edge nodes to target high throughput while keeping latency low.

4.16 Hardware and Protocol Stack

This subsection reviews specialized fog hardware and protocol stacks for integrating heterogeneous IoT devices, optical networks, resource management, and edge services.

  • Hardware: Intel’s Fog Reference Unit is a self-contained chassis that serves as a generic fog node for testing and demonstrating fog use cases.The reference design is intended for experimentation and demonstration.
  • Hardware: A proposed fiber-wireless architecture combines MEC with optical fiber networking and connects centralized cloud servers through a backbone network.The architecture is designed for IoT networking.
  • Hardware: A multi-Microcontroller industrial IoT gateway addresses heterogeneous communication, management, and big-data services.Its FPGA-based parallel bridge controller targets serial communication bottlenecks in traditional MCU architectures.
  • Network architectures: Cloud-Fog Radio Access Networks place cloud services on fog nodes and use NFV to process baseband signals from remote radio heads.Related F-RAN work considers interference mitigation, resource optimization, and mobility management.
  • Protocol stacks: The IoT hub protocol stack supports multiple device protocols and provides discovery, resource directories, origin services, and protocol proxies with caching.Supported device connectivity includes IEEE 802.15.4, IEEE 802.11, and Bluetooth Low-Energy.
  • Protocol stacks: FogOS organizes fog management into service abstraction, application management, resource management, and device abstraction layers.The abstraction layers expose service APIs and device data models for the IoT ecosystem.

4.17 Security and Privacy

The survey covers fog-based security and privacy mechanisms while identifying unresolved challenges in trust, offloading integrity, platforms, standards, hardware, resource models, and heterogeneous authentication.

  • Security and privacy mechanisms: Chaff services can reduce location-tracking accuracy to close to zero when moved carefully alongside users’ real services across edge clouds.The strategy adds fake services to confuse eavesdroppers observing service migrations.
  • Security and privacy mechanisms: Homomorphic encryption supports secure private-proximity detection while addressing communication and computation costs associated with existing approaches.The proposed protocol protects data transmission through fast encryption and decryption.
  • Security and privacy mechanisms: Differential privacy prevents colluded fog nodes from inferring IoT data by adding artificial noise before outsourcing.The surveyed scheme targets privacy of data shared by IoT devices.
  • Security and privacy mechanisms: Privacy-preserving aggregation lets control centers compute averages and variances without learning sensor readings and remains fault tolerant when devices malfunction.Fog nodes and control centers cannot learn the original collected data.
  • Open challenges: Fog offloading requires secure and private load balancing together with mechanisms that let receivers verify offloaded-task correctness and integrity.The challenge is explicitly linked to security and privacy risks incurred by offloading tasks to fog nodes.
  • Open challenges: The survey identifies the absence of a transparent PaaS spanning fog, IoT, and cloud, with protocol and API support for diverse applications.It proposes plugins for different fog-computing applications.
  • Open challenges: Fog computing lacks a universally agreed definition, motivating standardization and clarification of distinctions among related paradigms.The survey notes independent definitions and no globally unanimous distinction across researchers and industry.
  • Open challenges: Future directions include new hardware and communication technologies, peer-to-peer resource frameworks, and authentication mechanisms for heterogeneous fog nodes and IoT devices.The proposed P2P framework is intended to address heterogeneity, mobility, and operation without cloud connectivity.

5 CHALLENGES AND FUTURE RESEARCH DIRECTIONS

The survey identifies gaps spanning fog-system scalability, mobility, monitoring, energy, protocols, resilience, resource sharing, standardization, and hardware integration, and maps these gaps to future research directions.

  • Research coverage is organized by taxonomy categories and objectives, with article counts used as indicators of potential future research directions.Figures 10 and 11 summarize the distribution of surveyed research articles across taxonomy categories and supported objectives or features.
  • Scalable Design of Fog Schemes: Existing fog schemes often neglect scalability, so researchers should verify that algorithms can handle IoT-network magnitude, including through implementation.An example is online offloading that does not require information about individual IoT nodes for decisions.
  • Future work should address mobile fog nodes, multi-operator resource monitoring, and schemes that jointly consider objectives such as QoS, bandwidth, energy, and cost.Mobility complicates resource availability, discovery, offloading, and provisioning, while multi-operator access motivates monitoring techniques.
  • Green Fog Computing: Energy research remains limited across fog systems, motivating work on energy harvesters, battery storage, fog-node placement, and proximity to renewable energy sources.The survey distinguishes energy consumption at IoT devices, the interconnecting network, and fog nodes, and proposes routing toward renewable-powered fog nodes.
  • Fog communication and control need adaptation for multi-domain systems and high-speed users, including standardized SDN support and quick or stateless authentication handshakes.The survey specifically notes gaps in SDN interfaces and protocols for users in cars, trains, and vehicular-computing environments.
  • Resilient Fog System Design: Resilience requires coordinated fog-cloud provisioning, protection and restoration across layers, failure handling, and defenses against denial-of-service attacks.Limited fog resources make replica allocation nontrivial, while resource-constrained fog nodes are more exposed to DoS attacks than adequately secured cloud data centers.
  • Promising directions include peer-to-peer fog resource sharing, standardized fog definitions, and use of newer storage, networking, and FPGA technologies.The survey describes current fog models as primarily client-server and notes that most studies do not use the listed emerging hardware or communication technologies.

6 CONCLUSION

The paper situates fog computing within the growth of IoT and presents it as one approach for addressing the resulting computing demands.

  • IoT is expected to connect billions of devices and humans, and fog computing is presented as one response to this growth.
Loading 1808.05283v3…