PaperlayerView exact revision on arXiv

Source-linked AI summary

Fog Computing: Survey of Trends, Architectures, Requirements, and Research Directions

Ranesh Kumar Naha, Saurabh Garg, Dimitrios Georgakopoulos, Prem Prakash Jayaraman, Longxiang Gao, Yong Xiang, Rajiv Ranjan

arXiv:1807.00976v1cs.DC

TL;DR

Fog computing research lacks a requirements-based taxonomy spanning infrastructure, platforms, and applications, alongside comprehensive coverage of key operational areas. This survey reviews trends and architectures, proposes such a taxonomy, and identifies open research issues while reporting that Fog computing searches increased 2.5 times from 2010 to 2017.

  • Problem

    Existing Fog computing studies lack a requirements-based taxonomy spanning infrastructure, platforms, and applications, with incomplete coverage of resource management, fault tolerance, and microservices.

  • Method

    The survey reviews Fog trends and architectures, maps applications to a requirements-based taxonomy, and examines resource allocation, fault tolerance, simulation tools, and microservices.

  • Results

    Fog computing scholarly searches increased 2.5 times from 2010 to 2017, indicating rapidly increasing research interest.

  • Takeaways & Limitations

    The survey organizes current Fog computing research and identifies application- and platform-related gaps requiring further exploration.

Abstract

from arXiv · show

Emerging technologies like the Internet of Things (IoT) require latency-aware computation for real-time application processing. In IoT environments, connected things generate a huge amount of data, which are generally referred to as big data. Data generated from IoT devices are generally processed in a cloud infrastructure because of the on-demand services and scalability features of the cloud computing paradigm. However, processing IoT application requests on the cloud exclusively is not an efficient solution for some IoT applications, especially time-sensitive ones. To address this issue, Fog computing, which resides in between cloud and IoT devices, was proposed. In general, in the Fog computing environment, IoT devices are connected to Fog devices. These Fog devices are located in close proximity to users and are responsible for intermediate computation and storage. Fog computing research is still in its infancy, and taxonomy-based investigation into the requirements of Fog infrastructure, platform, and applications mapped to current research is still required. This paper starts with an overview of Fog computing in which the definition of Fog computing, research trends, and the technical differences between Fog and cloud are reviewed. Then, we investigate numerous proposed Fog computing architecture and describe the components of these architectures in detail. From this, the role of each component will be defined, which will help in the deployment of Fog computing. Next, a taxonomy of Fog computing is proposed by considering the requirements of the Fog computing paradigm. We also discuss existing research works and gaps in resource allocation and scheduling, fault tolerance, simulation tools, and Fog-based microservices. Finally, by addressing the limitations of current research works, we present some open issues, which will determine the future research direction.

I. INTRODUCTION · II. OVERVIEW OF FOG COMPUTING

The paper surveys Fog computing as a decentralized layer between cloud servers and IoT devices, motivated by the latency and scalability challenges of processing rapidly growing sensor data in the cloud. It reviews Fog architectures, requirements, research gaps, and open issues across infrastructure, platforms, and applications.

  • I. INTRODUCTION: Fog computing addresses IoT’s massive sensor-data growth and cloud-processing limitations by placing computation and storage closer to users.Cloud computing offers accessibility, scalability, and pay-per-use services, but data from billions of sensors cannot be transferred and processed entirely in the cloud.
  • I. INTRODUCTION: Fog is a decentralized computing paradigm that uses nearby devices’ idle resources to reduce cloud latency while retaining cloud support for complex processing,.Unlike cloud computing’s reliance on a central component, Fog distributes computation across capable devices around users.
  • I. INTRODUCTION: The survey identifies heterogeneity as a central deployment challenge, requiring research on resource management and failure handling across diverse Fog devices.It also highlights gaps in resource allocation and scheduling, fault tolerance, simulation tools, and Fog-based microservices.
  • I. INTRODUCTION: The survey presents research trends, reviews Fog architectures in detail, and proposes a requirements-based taxonomy spanning infrastructure, platforms, and applications.Its stated contributions include investigating publication and Google Scholar trends, detailing architecture components, and classifying Fog research by infrastructure, platform, and application requirements.
  • II. OVERVIEW OF FOG COMPUTING: Fog research builds on edge computation and cloudlets: cloudlets target mobile networks, whereas Fog computing targets connected IoT things,,,.The term “Fog computing” was proposed in 2012 by Cisco researchers.
  • II. OVERVIEW OF FOG COMPUTING: Fog provides virtualized or non-virtualized networking, storage, and computation between cloud servers and IoT devices, primarily supporting low-latency applications,,.These services are distributed and not completely located at the network edge; Fog also supports non-latency-aware services.
  • II. OVERVIEW OF FOG COMPUTING: The paper surveys Fog definitions and research trends, compares Fog technically with cloud computing, and synthesizes prior architecture and taxonomy studies,,,,,.Prior work covers hierarchical architectures, enabling technologies, requirements, challenges, and evaluation criteria including heterogeneity, QoS, scalability, mobility, and federation.

A. DEFINITION OF FOG COMPUTING · B. FOG COMPUTING RESEARCH TRENDS

Fog computing is defined as a distributed platform that processes data on virtualized and non-virtualized edge devices while using the cloud for non-latency-aware processing and long-term storage. Research interest is rapidly increasing as edge processing addresses IoT latency needs and enables utilization of idle nearby resources.

  • A. DEFINITION OF FOG COMPUTING: Fog computing places compute, storage, and networking between IoT devices and cloud data centers, supporting latency-sensitive applications near data-producing devices.The paradigm integrates with cloud infrastructure while moving processing toward the network edge.
  • A. DEFINITION OF FOG COMPUTING: Edge processing reduces reliance on transferring all IoT data to the cloud, which would require massive bandwidth and storage and may require protocol translation.IoT devices may use industrial protocols rather than direct IP connections, creating an additional translation requirement.
  • A. DEFINITION OF FOG COMPUTING: Fog definitions vary, but they consistently describe decentralized or edge-proximate resources that provide processing and storage between users, IoT devices, and centralized cloud infrastructure.Some definitions include heterogeneous, autonomous devices cooperating on storage and processing tasks, while others emphasize moving resources from centralized clouds to network ends.
  • A. DEFINITION OF FOG COMPUTING: Fog computing performs most processing on virtualized and non-virtualized edge devices while associating with the cloud for non-latency-aware processing and long-term storage.The authors include all computing- and storage-capable devices as Fog devices and clarify the cloud’s role in the environment.
  • B. FOG COMPUTING RESEARCH TRENDS: Research attention has grown around processing data closer to users because handling IoT-generated data at the edge can improve overall processing time.The section connects Fog computing with latency-aware smart-home services, particularly emergency-response applications.
  • B. FOG COMPUTING RESEARCH TRENDS: Edge computing was the most searched related paradigm in Google Scholar, although its search trend decreased by more than three times over eight years.Mobile cloud computing and mobile edge computing followed edge computing among the most searched paradigms.
  • B. FOG COMPUTING RESEARCH TRENDS: Fog computing articles more than doubled between 2015 and 2016, and about 564 journal and conference articles were published across four major digital libraries.The first paper with “Fog computing” in its title appeared in 2012, supporting the observed rapid growth of the field.
  • B. FOG COMPUTING RESEARCH TRENDS: The authors conclude that Fog research interest is rapidly increasing and that idle resources in devices near users can support market adoption and technological development.Their trend analysis combines Gartner’s hype-cycle context, Google Scholar searches, and publication counts from digital libraries.

C. DIFFERENCE BETWEEN FOG AND CLOUD · III. RELATED PARADIGMS AND TECHNOLOGIES · A. MOBILE CLOUD COMPUTING (MCC)

Fog computing differs from cloud computing through its distributed, geographically orchestrated Fog clusters near users, enabling lower latency, energy use, and operating costs but incurring higher failure risks. Related paradigms include MCC, which uses near-user cloudlets and cloud collaboration to overcome mobile-device limitations and support computationally intensive applications.

  • C. DIFFERENCE BETWEEN FOG AND CLOUD: Fog uses geographically distributed Fog clusters near users, whereas cloud computing relies on centralized datacenters, with lower Fog operating costs and energy consumption.Fog devices cooperate for processing, while cloud infrastructure is more centralized and Fog infrastructure is more distributed.
  • C. DIFFERENCE BETWEEN FOG AND CLOUD: Fog enables real-time interaction by reducing latency, but wireless connectivity, decentralized management, and power failures increase its failure rate, –.Most Fog devices are expected to connect wirelessly, including smart gadgets and handheld devices.
  • C. DIFFERENCE BETWEEN FOG AND CLOUD: Fog cannot replace or be judged categorically better than cloud computing because the paradigms fulfill different perspectives and requirements.The paper presents Fog and cloud as complementary rather than interchangeable.
  • III. RELATED PARADIGMS AND TECHNOLOGIES: Fog computing places resources between traditional cloud infrastructure and edge devices to provide faster application processing and services.The paper situates Fog among related paradigms including MCC, MEC, Edge Computing, Dew Computing, and Fog-dew computing.
  • A. MOBILE CLOUD COMPUTING (MCC): MCC supports remote execution of offloaded mobile services through near-user cloudlets, overcoming smart devices’ computational, energy, and storage limitations,,.MCC provides functionality, mobility, and storage through heterogeneous network connectivity and follows a pay-per-use model.
  • A. MOBILE CLOUD COMPUTING (MCC): MCC provides mobile users Internet access to data, applications, and cloud resources, with expected applications in education, development, healthcare, and social networking.The paper also identifies computation-intensive applications such as augmented reality, computer vision, speech recognition, and machine learning.
  • A. MOBILE CLOUD COMPUTING (MCC): Computation-intensive mobile applications require edge processing and cloud collaboration, motivating low-latency middle tiers, seamless remote execution, and optimized cloud infrastructure.Simply designing more powerful mobile devices does not satisfy these application requirements.

B. MOBILE EDGE COMPUTING (MEC) … E. FOG-DEW COMPUTING

The paper distinguishes MEC, edge, Dew, and Fog-dew computing by their placement, connectivity, service models, and degree of edge processing. Across these paradigms, computation and storage move toward users and IoT devices, while Dew and Fog-dew computing additionally support decentralized or offline operation.

  • B. MOBILE EDGE COMPUTING (MEC): MEC co-locates computing and storage at cellular base stations, enabling two- or three-tier deployments with mobile end devices.A MEC server is deployed near base-station towers, with mobile users, network operators, Internet infrastructure providers, and application service providers participating in the ecosystem.
  • B. MOBILE EDGE COMPUTING (MEC): MEC evolved from mobile base stations through collaborative telecommunications and IT networking, supporting IoT, location, augmented-reality, caching, video-analytics, and local-content services.It provides real-time low-latency local access or cached content, but installing MEC servers is a principal limitation.
  • C. EDGE COMPUTING: Edge devices both produce and consume data, performing cloud offloading, storage, processing, caching, request distribution, and cloud-provided services for users.These devices must be designed to satisfy privacy requirements.
  • D. DEW COMPUTING (DC): Dew Computing operates at the ground level of the cloud–Fog hierarchy as a vertically distributed microservice sub-platform connecting sensors, tablets, and smartphones.Unlike Fog, which supports latency-sensitive IoT applications and dynamic network configurations, Dew is not dependent on a centralized device, server, or cloud.
  • D. DEW COMPUTING (DC): Dew Computing can support smart traffic control by placing data-collection and processing units between traffic signals to generate collective traffic conditions and warn vehicles about congestion-related risks [40].The example includes alerts for low-fuel cars and hybrid vehicles approaching congestion.
  • E. FOG-DEW COMPUTING: Fog-dew computing lets IoT devices connect to a community server without an active Internet connection, while that server interacts with the cloud to provide services.Unlike cloud computing, which generally requires Internet connectivity, Fog-dew computing facilitates offline services; Fog-related paradigms differ in connectivity and how much processing occurs at the edge.

IV. ARCHITECTURE OF FOG COMPUTING … 1) Physical layer

The paper reviews Fog computing architectures because no standard architecture yet exists, progressing from a three-layer high-level model to varied proposed architectures and detailed component descriptions. Its physical layer begins with sensor-generated data sources supporting sensing, monitoring, and actuation-oriented applications.

  • IV. ARCHITECTURE OF FOG COMPUTING: Because no standard Fog architecture exists, the paper reviews proposed designs and presents a detailed architecture with component-level descriptions for deployment.The review covers high-level architecture, existing proposals, and component functionality.
  • A. HIGH-LEVEL ARCHITECTURE OF FOG COMPUTING: The high-level model has three layers: IoT devices perform sensing and actuation, the Fog layer provides intermediate computation, and the cloud handles other processing.Container-based virtualization is described as more appropriate for the resource-constrained Fog layer than traditional virtualization.
  • A. HIGH-LEVEL ARCHITECTURE OF FOG COMPUTING: For time-sensitive applications, processing should occur exclusively on the Fog plane, which controls what data is sent to the cloud while users access both services.The Fog layer accumulates sensor-generated data and manages cloud communication.
  • B. VARIOUS PROPOSED ARCHITECTURES FOR FOG COMPUTING: Reviewed Fog architectures use three, four, five, or six layers, with layer counts varying between high-level representations and more implementation-oriented designs.The cited works justify their layer choices differently, and omitting the user plane yields a three-level high-level view.
  • B. VARIOUS PROPOSED ARCHITECTURES FOR FOG COMPUTING: Aazam and Huh organize six layers around specific tasks, whereas Dastjerdi et al. define five layers from a network perspective.Other reviewed designs include hierarchical Fog, OpenFog, Fog network, Internet of Energy, nervous-system-based, and IFCIoT architectures.
  • C. COMPONENTS OF FOG COMPUTING ARCHITECTURE: The detailed architecture groups components by layer and functionality to connect IoT devices with Fog devices, servers, gateways, and the cloud.Figure 8 presents the architecture’s layers and components.
  • 1) Physical layer: The physical layer receives data from sensors and other sources, including smart devices, environmental sensors, smart homes, surveillance systems, traffic systems, and vehicles.These sources provide the basic data input for Fog computing.
  • 1) Physical layer: In smart traffic management, roadside sensors, devices, cameras, and GPS sensors support current-condition monitoring, traffic-signal management, and future-demand prediction.Virtual sensors are also important because decisions such as road closures cannot rely on a single physical sensor.

2) Fog device, server, and gateway layer … 8) Application layer

The paper describes Fog’s layered architecture from device, server, and gateway management through monitoring, processing, storage, resource management, security, and applications. Together, these layers support distributed, latency-aware IoT services by coordinating resources, data, reliability, scalability, and privacy.

  • 2) Fog device, server, and gateway layer: Fog devices, servers, or gateways may be standalone or IoT-integrated, with servers requiring higher configurations to manage multiple Fog devices.Their roles depend on hardware, connectivity, and the number of managed devices; cross-cluster applications may require coordinated processing across multiple Fog devices and servers.
  • 3) Monitoring layer: The monitoring layer tracks performance, resources, services, and responses to select appropriate resources during operation.It addresses situations where computation or storage becomes unavailable on Fog devices or servers.
  • 4) Pre and post-processing layer: The pre- and post-processing layer analyzes, filters, trims, and reconstructs data, then routes it to local storage or the cloud for long-term storage.Real-time aggregation can retain application-relevant summaries instead of every sensor reading, reducing unnecessary storage.
  • 5) Storage layer: The storage layer uses storage virtualization to pool networked devices as one manageable resource, while backup improves data availability and mitigates loss.Virtualization can provide enterprise-class functionality using less-expensive or commodity hardware.
  • 6) Resource management: Resource management coordinates allocation, scheduling, energy saving, reliability, and scalability, with Fog targeting both horizontal and vertical scalability.Application scheduling handles different objectives, while energy efficiency can reduce operational costs and environmental impact.
  • 7) Security layer: The security layer maintains encrypted communications, secure data storage, and user privacy while supporting utility-style access to Fog services.Users connect to Fog infrastructure for services, while Fog middleware manages the underlying resources.
  • 8) Application layer: Fog supports latency-aware IoT, Wireless Sensor Network, CDN, utility, and Augmented Reality applications.For Augmented Reality, Fog can address real-time processing needs and support continuous service improvement.

V. TAXONOMY OF FOG COMPUTING … 3) Fog devices

The paper proposes a requirements-based Fog computing taxonomy spanning infrastructure, platform, and applications. Its infrastructure analysis covers physical resources, network constraints, and heterogeneous IoT, processing, and gateway devices that collaboratively support edge computation.

  • V. TAXONOMY OF FOG COMPUTING: The proposed taxonomy organizes Fog computing requirements across infrastructure, platform, and applications, providing a framework for research and enterprise deployment.It identifies infrastructure and network requirements, device types, platform allocation and scheduling, security and privacy, services, management, multitenancy, and application and user requirements.
  • A. INFRASTRUCTURE: Fog infrastructure comprises the physical devices, network devices, and gateways that participate in computation, with requirements determined by their capabilities and connectivity.The infrastructure must provide computation, networking, storage, and memory resources for efficient Fog services.
  • 1) Infrastructure requirements: The proliferation of connected devices, sensors, and actuators requires infrastructure capable of supporting a vast Fog computing environment.The paper cites an estimate of 50 billion handheld devices by 2020, alongside expanding sensor and actuator deployments.
  • 2) Network requirements: Billions of edge devices make the network a key Fog bottleneck, requiring existing LAN, MAN, WAN, and PAN technologies to be further investigated and amended.Fog networks must support many simultaneously connected devices, often with low individual bandwidth requirements.
  • 3) Fog devices: Fog computing addresses sensor-rich IoT settings where sending all data to the cloud in real time is impractical, so Fog layers provide intermediate processing, storage, management, and gateway services,.The Fog plane relies on collaboration among sensor-management, processing, storage, and gateway devices.
  • 3) Fog devices: IoT devices provide sensing and actuation, generating application data from heterogeneous sensors such as speed, camera, temperature, vehicle-monitoring, and GPS sensors.A typical sensor generates 10 data samples every second, and sensors convert environmental variables into electrical signals.
  • 3) Fog devices: Devices such as smartphones and Raspberry Pi boards can function simultaneously as IoT and Fog devices when equipped with sensing or actuation capabilities.This illustrates the taxonomy’s support for overlapping device roles across the IoT and Fog layers.
  • 3) Fog devices: Fog processing devices require computing, storage, and network connectivity, while gateway boards provide heterogeneous hardware and protocol support for edge data handling.Possible processing devices include controllers, switches, routers, servers, and surveillance cameras; gateways can condition, analyze, filter, and transfer data to the cloud.

B. PLATFORM · 2) Service requirements

The Fog platform manages applications and infrastructure through resource allocation, scheduling, fault tolerance, multi-tenancy, security, and privacy. Its service requirements emphasize continuously updated outcomes, fault-tolerant operation, and QoS based on reliability, delay, throughput, and energy consumption.

  • B. PLATFORM: The Fog platform manages applications and infrastructure while coordinating resource allocation, scheduling, fault tolerance, multi-tenancy, security, and privacy.These requirements are defined through a Fog computing platform taxonomy.
  • B. PLATFORM: Heterogeneous Fog devices make resource allocation and scheduling difficult, while inefficient task placement can undermine latency-aware IoT application processing.Efficiently scheduling tasks on idle devices is necessary to use their computation power without creating complex processing issues.
  • 2) Service requirements: Fog services may comprise single or multiple user requests whose outcomes continuously change while users remain subscribed.Fog devices and servers perform intermediate processing between user requests and service outputs, with Fog servers potentially communicating with the cloud.
  • 2) Service requirements: Fault tolerance keeps Fog systems operating after software, hardware, or network failures, potentially with reduced capability rather than complete shutdown.Although fault tolerance has been investigated mainly in cloud computing –, Fog computing also requires dedicated investigation.
  • 2) Service requirements: Fog QoS depends on reliability, network delay, throughput, and energy consumption, alongside resource management, power modeling, scheduling, and power-failure handling.These factors are needed to ensure service quality in Fog environments.
  • 2) Service requirements: Because Fog targets latency-sensitive systems, it must maintain high reliability even when sensor failures could reduce outcome or action accuracy.Sensor failures can affect the accuracy of resulting outcomes or actions, making reliability a central QoS concern.

3) Security and privacy … 1) Application requirements

Fog computing requires coordinated security, privacy, tenancy, management, and application capabilities because processing is distributed across heterogeneous, mobile, resource-constrained environments. Applications particularly require scalability, heterogeneity handling, timely interaction, and mobility support, while users and providers need protections for networks, data, and privacy.

  • 3) Security and privacy: Fog security must protect network connections, data, and user privacy because sensitive user data may be processed on intermediate devices outside users’ or applications’ full control.Application providers must also prevent unwanted data theft.
  • 4) Multi-tenancy: Multi-tenancy lets one Fog-device instance serve multiple isolated tenants, conserving limited Fog resources through container-based or virtualization-based isolation.Container-based virtualization is described as more lightweight and powerful than usual virtualization-based approaches.
  • 5) Management: Fog management may be centralized or decentralized, but distributed management is often necessary because devices belong to different domains and applications may execute across multiple Fog clusters.Decentralized deployments require similar management processes across different Fog devices.
  • C. APPLICATION: Fog applications must satisfy explicit execution requirements defined for the application layer.The application requirements include scalability, heterogeneity, interaction timeliness, and mobility.
  • 1) Application requirements: Scalability requires autonomous handling of rapidly increasing IoT devices and services, including changing device states, availability, and device querying and selection.Cloud-dependent processing and maintaining frequently changing device states and availability become difficult as the number of devices grows.
  • 1) Application requirements: Heterogeneity requires abstracting device complexity across diverse services, protocols, and horizontal and vertical levels of the Fog architecture.Device heterogeneity is presented as a fundamental characteristic of IoT systems.
  • 1) Application requirements: Interaction timeliness requires Fog and cloud infrastructure to support the perception-action cycle and its application communication logic through multiple interaction models.Giang et al. identified four interaction models for the perception-action cycle in Fog environments.
  • 1) Application requirements: Mobility support must coordinate mobile edge, computing, and storage devices across two planes, using task distribution, duplication, and migration to preserve resource availability and task completion.Managing and synchronizing mobile devices across these planes is challenging.

2) User requirements … D. HEALTHCARE

The paper organizes Fog-computing applications around user constraints, application models, and service domains, showing how Fog addresses latency-sensitive and real-time requirements across transportation, vehicles, AR/VR, and healthcare. Its taxonomy maps related work to application dimensions and exposes areas requiring further investigation.

  • 2) User requirements: User requirements may prioritize deadlines, budget limits, response time, or tolerable accuracy depending on the application constraint.Some users value fast results over accuracy, while others impose explicit time or budget requirements.
  • 3) Application modeling: Fog applications can model periodic IoT data as real-time streams or process previously stored datasets through microservice-based applications.Microservices can bind application functionality and required libraries together.
  • VI. DIMENSION OF FOG COMPUTING-BASED APPLICATIONS: Fog-based application research spans smart transportation, augmented and virtual reality, healthcare, video streaming, smart homes, and smart cities, with taxonomy mapping used to identify research gaps.The paper evaluates works according to platform and application requirements, noting that service dimensions are interrelated although studies often focus on only some aspects.
  • A. SMART TRANSPORTATION SYSTEM: Fog-based smart transportation research addresses VANET requirements for rapid decisions and communication among vehicles, infrastructure, and base stations.Truong et al. combined SDN programmability, flexibility, global knowledge, and scalability with Fog location awareness and time sensitivity in an FSDN architecture; Giang et al. examined Fog-based VANET applications for urban driving decisions.
  • B. VEHICLES AS FOG INFRASTRUCTURE: Vehicular Fog Computing uses moving vehicles as computation and communication infrastructure, aggregating their resources to serve nearby edge devices and enhance service quality.Hou et al. used quantitative analysis across different scenarios to study relationships involving connectivity and mobility.
  • C. AUGMENTED AND VIRTUAL REALITY: Because AR and VR applications are extremely time sensitive, Fog can support real-time processing for interaction and game experiences.Zao et al. used Fog and cloud infrastructure for real-time signal processing in an augmented brain-computer interaction game.
  • D. HEALTHCARE: Fog enables real-time sensor-based healthcare by processing medical data near the source and generating emergency notifications.Rahmani et al. proposed a Fog-assisted architecture centered on a smart e-health gateway that processes sensor data and generates an Early Warning Score for medical emergencies.

E. SMART CITY … 1) Resource allocation and scheduling

The reviewed Fog-computing research spans smart-city applications and four research areas, with resource allocation and scheduling receiving substantial attention. Existing work addresses collaborative Fog–cloud deployment, workload distribution, preprocessing, mapping, and dynamic estimation, while application- and platform-level aspects remain underexplored.

  • E. SMART CITY: Fog computing supports real-time smart-city applications through distributed architectures, including noise mapping, urban drainage, and smart streets.Giordano et al.’s Rainbow framework uses a distributed agent-based intermediate layer between physical infrastructure and applications.
  • E. SMART CITY: Across Fog-based applications, research has focused mainly on infrastructure and application modeling, leaving application- and platform-related aspects as gaps.Table 4 summarizes the reviewed applications according to the proposed taxonomy.
  • VII. STATE-OF-THE-ART FOG COMPUTING: The state-of-the-art review covers resource allocation and scheduling, failure handling, simulation tools, and Fog-based microservices.These areas organize the paper’s discussion of existing Fog-computing research.
  • A. RESOURCE ALLOCATION AND SCHEDULING IN FOG COMPUTING: Most reviewed studies address resource allocation and scheduling in cloud–Fog environments, reflecting Fog computing’s rapid evolution and heterogeneous edge-level computation.The review compares these works to identify research gaps in the field.
  • 1) Resource allocation and scheduling: Fog–cloud allocation methods manage requests and workload using VM capacity, completion time, and service size, while optimization frameworks address power-consumption–delay trade-offs.Alsaar et al. use linearized decision-tree rules; Deng et al. apply Hungarian and Generalized Benders Decomposition algorithms.
  • 1) Resource allocation and scheduling: Deployment-planning research verifies hardware, software, and QoS eligibility, maps IoT modules to suitable Fog–cloud nodes, and emphasizes latency over resource consumption and communication links.FogTorchQ combines Monte Carlo simulation with preprocessing and backtracking, while the Module Mapping Algorithm uses lower-bound searches.
  • 1) Resource allocation and scheduling: Collaborative computation dynamically assigns preprocessing and workloads across Fog, cloud, and SDN resources, while customer history informs dynamic resource estimation.Yin et al. formulate social-welfare maximization with H-ADMM; Aazam et al. use relinquish probability, set to 0.3 for new customers and starting at 0.1 for existing-history-based estimation.

2) Resource allocation and scheduling for a Fog environment · 3) Summary of resource allocation and scheduling in Fog · B. FAULT TOLERANCE IN FOG COMPUTING

The reviewed Fog resource-allocation studies address autonomous selection, penalty-aware placement, and incentive-based resource participation, but research remains concentrated on allocation and cloud-based simulation. Fault tolerance is especially important in heterogeneous Fog environments, where hybrid failure-handling methods appear more appropriate than single techniques.

  • 2) Resource allocation and scheduling for a Fog environment: Ni et al. proposed priced timed Petri nets that let users autonomously select resources from pre-allocated groups while considering time, price, and credibility.Users with higher credit limits can allocate highly reliable resources.
  • 2) Resource allocation and scheduling for a Fog environment: Pooranian et al. modeled allocation as penalty-aware bin packing, rewarding or penalizing servers using idle energy, maximum frequency, and maximum energy.VM placement is constrained by server capacity, frequency, and time limitations.
  • 2) Resource allocation and scheduling for a Fog environment: Sun and Zhang proposed crowd funding with incentives, rewards, and punishments to integrate idle local-network resources and encourage owners to perform computations.The approach is described as similar to Pooranian et al..
  • 3) Summary of resource allocation and scheduling in Fog: The survey finds that most studies focus on Fog resource allocation, leaving resource sharing and workload allocation comparatively underexplored.Table 5 summarizes the surveyed resource-allocation and scheduling research.
  • 3) Summary of resource allocation and scheduling in Fog: Cloud-based simulations are considered unconvincing for Fog because Fog environments are more heterogeneous and dynamic, motivating Fog-specific workload generation and simulation.Further investigation is explicitly requested for both workload generation and simulations in Fog.
  • B. FAULT TOLERANCE IN FOG COMPUTING: Fog’s highly distributed and heterogeneous nature makes device failures more probable than in cloud, yet the survey reports no prior Fog-specific fault-tolerance study.Fault tolerance has instead been studied mainly in cloud computing.
  • B. FAULT TOLERANCE IN FOG COMPUTING: Cloud fault tolerance commonly uses proactive or reactive techniques at workflow or task level, including migration, checkpointing, replication, rollback, recovery, and resubmission.Fault handling in Fog must cover processing as well as transmission and reception.
  • B. FAULT TOLERANCE IN FOG COMPUTING: Hybrid failure handling is considered more appropriate for Fog because of its unstable failure behavior and heterogeneous characteristics.Reviewed approaches include standby replacement, checkpointing, migration, and proactive monitoring based on hardware-health indicators.

C. SIMULATION TOOLS FOR FOG COMPUTING … 2) Microservices and IoT applications

Fog-computing simulation research remains limited and omits important network and virtualization factors, while microservices offer a modular approach for resource-constrained, time-sensitive IoT applications. Existing microservice studies address heterogeneous IoT and Fog architectures, but Fog-based microservices remain an open area with unresolved distributed-computing challenges.

  • C. SIMULATION TOOLS FOR FOG COMPUTING: Fog simulation research remains in its infancy, with existing studies targeting specific aspects such as resource prediction, pricing, and smart-city incident-detection queries.Aazam and Huh modeled resource requirements from future user behavior, while Dastjerdi et al. studied query DAGs; both used CloudSim.
  • C. SIMULATION TOOLS FOR FOG COMPUTING: Fog deployment requires simulators that support containers, Software-defined Networking, and Network Function Virtualization for emerging 5G-integrated environments.The paper identifies integration with 5G Technologies, NFV, and SDN as a deployment challenge.
  • C. SIMULATION TOOLS FOR FOG COMPUTING: The two commonly used Fog simulators omit bandwidth distribution, round-trip delay, and container-based virtualization, limiting latency-focused and deployment-realistic evaluations.These omissions matter because network parameters heavily affect results when latency minimization is the goal.
  • D. FOG-BASED MICRO SERVICES: Microservices partition distributed applications into independently implemented functionality-specific processes that communicate through message passing and unambiguous APIs.This structure improves scalability and maintainability while allowing mainstream programming models.
  • D. FOG-BASED MICRO SERVICES: Microservices suit Fog-based IoT development because their agility and modularity reduce complexity under Fog resource limitations.Independent services divide large systems into smaller components, making deployment more convenient for IoT applications.
  • 1) Current research aspects of microservice: Fog-based microservices remain insufficiently investigated, although studies address heterogeneous devices, communication protocols, service orchestration, and programmable self-managing IoT platforms, –.Reported architectures show benefits over monolithic designs in scalability, flexibility, and platform independence, but face network faults, delay, serialization, and transaction-processing challenges.

E. FOG BASED MOBILE COMPUTING … 3) Interoperability and federation of Fog

Mobile Fog computing addresses growing, geographically distributed content demand by offloading computation and content to nearby Fog devices, while requiring distributed management, interoperability, and federation. The paper identifies deployment-tier, standardization, and scalability challenges that shape Fog’s future research and market adoption.

  • E. FOG BASED MOBILE COMPUTING: Mobile Fog computing offloads content and computation to nearby Fog devices, but demand-aware distribution, monitoring, and cache management remain challenging for resource-limited nodes.These challenges arise as densely populated mobile users generate high-volume concurrent content requests.
  • E. FOG BASED MOBILE COMPUTING: Mobile Fog research explores programming models for heterogeneous geospatial applications, peer-to-peer resource sharing with CoAP microservices, content offloading, and hierarchical WSN–cloud communication.Hong et al. target simplified development and dynamic resource scaling, Shi et al. introduce mobile-device resource sharing, Khan et al. investigate content offloading, and Wang et al. bridge WSNs and the cloud through routing and Fog layers.
  • VIII. OPEN ISSUES AND FUTURE RESEARCH DIRECTIONS: The paper identifies Fog computing research issues and future directions as an open-issues area requiring continued investigation.Figure 10 summarizes the research issues discussed in the paper.
  • A. INFRASTRUCTURE-RELATED ISSUES: Fog must evolve toward market adoption to support time-sensitive applications, with OpenFog established by academia and industry as a key standardization initiative.The consortium was founded by ARM, Cisco, Dell, Intel, Microsoft, and Princeton University in November 2015.
  • 1) Deployment issues: OpenFog’s N-tier deployment can introduce latency as tiers proliferate, so the number of levels should be selected according to workload, sensor scale, device capability, latency, and reliability requirements.Deployment decisions depend on the tasks assigned to each tier and the capabilities and requirements of the target use case.
  • 2) Standard architecture for Fog computing: Fog computing lacks a defined standard architecture, although OpenFog released architecture versions in 2016 and 2017 covering performance, manageability, security, data analytics, and control.The second draft expanded the initial architectural overview with greater detail.
  • 3) Interoperability and federation of Fog: When Fog devices cannot handle simultaneous latency-aware requests, sending them to the cloud violates latency requirements, motivating interoperability and federation among Fog clusters and servers.The issue arises because nearby Fog processing minimizes latency, but overloaded Fog devices need coordinated alternatives.

B. PLATFORM-RELATED ISSUES … IX. CONCLUSION

The paper identifies platform and application challenges in Fog computing, including dynamic resources, failures, connectivity, participation, security, service management, and interoperability. It concludes that Fog computing remains immature and needs further investigation across its taxonomy and supporting technologies.

  • B. PLATFORM-RELATED ISSUES: Fog resource management is difficult because devices are highly dynamic and heterogeneous, with varying resources and local application workloads.
  • 2) Failure management: Fog devices have high failure risk because distributed management, hardware and software faults, user activity, unreliable wireless links, mobility, and power constraints can disrupt operation.
  • 3) Communication between different layers: Continuous cross-layer connectivity among IoT devices, Fog, and cloud is essential for time-sensitive applications, even when cloud connectivity fails.Connectivity failures could seriously harm autonomous-car, drone, or emergency-surveillance applications.
  • 4) User participation management: Fog service management must address uncertain user participation through feasible resource deployment and incentive- or reward-based policies.
  • C. APPLICATION-RELATED ISSUES: Distributed ownership creates security and privacy risks for participant devices, user data, and cross-layer communication, requiring distributed security management independent of a central component.
  • 1) Application service management: Fog must manage diverse service availability and quality across billions of IoT devices, motivating microservice-based services and further Fog-based research.
  • 1) Application service management: Fog application modeling is challenging because IoT devices use different protocols and code sets, creating a need for standard communication protocols and easily deployable generic applications.
  • IX. CONCLUSION: The survey concludes that Fog computing is still in its infancy and requires broader investigation spanning its taxonomy, architecture, state of the art, resource allocation, scheduling, fault tolerance, and simulation tools.
Loading 1807.00976v1…