PaperlayerView exact revision on arXiv

Source-linked AI summary

A Comprehensive Survey on Fog Computing: State-of-the-art and Research Challenges

Carla Mouradian, Diala Naboulsi, Sami Yangui, Roch H. Glitho, Monique J. Morrow, Paul A. Polakos

arXiv:1710.11001v3cs.DCcs.SE

TL;DR

Fog computing addresses cloud limitations for latency-sensitive applications and location-constrained SLAs by provisioning resources and services near end devices. This survey evaluates fog architectures and algorithms, derives lessons and research directions, and discusses prospects for technologies such as Tactile Internet.

  • Problem

    Cloud distance can hinder latency-sensitive applications, while SLAs may require processing where cloud providers lack data centers.

  • Method

    The paper comprehensively surveys fog computing using evaluation criteria to critically review architectures, algorithms, challenges, and research directions.

  • Results

    The review covers 68 papers from 2013–2017 and finds that fog systems can reduce latency, with reported reductions of 48% and 73%.

  • Takeaways & Limitations

    Fog computing complements cloud computing by enabling edge processing while retaining interaction with the cloud.

  • Takeaways & Limitations

    The survey excludes security, confidentiality, and data-protection efforts because these topics are covered by dedicated surveys.

Abstract

from arXiv · show

Cloud computing with its three key facets (i.e., IaaS, PaaS, and SaaS) and its inherent advantages (e.g., elasticity and scalability) still faces several challenges. The distance between the cloud and the end devices might be an issue for latency-sensitive applications such as disaster management and content delivery applications. Service Level Agreements (SLAs) may also impose processing at locations where the cloud provider does not have data centers. Fog computing is a novel paradigm to address such issues. It enables provisioning resources and services outside the cloud, at the edge of the network, closer to end devices or eventually, at locations stipulated by SLAs. Fog computing is not a substitute for cloud computing but a powerful complement. It enables processing at the edge while still offering the possibility to interact with the cloud. This article presents a comprehensive survey on fog computing. It critically reviews the state of the art in the light of a concise set of evaluation criteria. We cover both the architectures and the algorithms that make fog systems. Challenges and research directions are also introduced. In addition, the lessons learned are reviewed and the prospects are discussed in terms of the key role fog is likely to play in emerging technologies such as Tactile Internet.

I. INTRODUCTION

This survey addresses cloud-computing limitations for latency-sensitive and location-constrained applications by reviewing fog computing from architectural and algorithmic perspectives. It proposes a structured literature classification, evaluation criteria, research directions, lessons learned, and a reading map.

  • I. INTRODUCTION: Cloud connectivity can be unsuitable for latency-sensitive applications, while regulations may require processing where providers lack data centers.Examples include connected vehicles, fire detection, smart grids, and content delivery; inter-cloud communication can further worsen latency.
  • I. INTRODUCTION: Fog computing extends cloud architecture to the network edge, enabling selected processing near end devices while other components remain in the cloud.The survey describes fog nodes as providing compute, storage, networking, low-latency processing, and distributed data-gathering points.
  • C. Paper Organization and Reading Map: The survey provides an a la carte reading map for readers focused on architecture, algorithms, or a high-level overview.The recommended section paths differ according to each reader’s primary interest.
  • I. INTRODUCTION: Unlike earlier surveys, this work critically evaluates contributions using defined criteria and comprehensively addresses architectural, algorithmic, and research-direction issues.It also derives lessons learned and discusses fog computing’s prospects for emerging technologies such as Tactile Internet.
  • B. Literature Classification: The survey classifies fog literature into architectures and algorithms, reflecting the two perspectives most researchers use to address fog-system issues.A work addressing both perspectives is reviewed in both sections.
  • B. Literature Classification: The review covers 68 papers from 2013–2017, including 32 architecture-focused papers and roughly half devoted to algorithms.The surveyed literature grew from four papers in 2013 and three in 2014 to 33 in 2015 and 22 in 2016.

II. FOG COMPUTING AND RELATED CONCEPTS

Fog computing brings computation to the network edge and complements cloud computing, while differing from cloudlets and MEC in scope, architecture, and application coverage.

  • Edge provisioning brings resources closer to end devices, reducing latency and enabling applications such as mobile data offloading.
  • The survey excludes industry initiatives focused solely on implementations and compares cloudlets, MEC, and fog rather than superseded cyber foraging.
  • Cloudlets serve mobile offloading through nearby virtualized servers, whereas MEC targets edge applications in mobile and non-mobile networks.
  • Fog computing extends cloud architecture to the edge while retaining cloud interaction and supporting applications spanning cloud and edge.
  • Fog systems use a three-tier architecture comprising cloud, fog, and IoT/end-user strata, with fog domains formed from heterogeneous edge devices.

III. ILLUSTRATIVE USE CASES AND FOG SYSTEM EVALUATION CRITERIA

The paper introduces illustrative fog-computing use cases and uses them to derive evaluation criteria for fog systems.

  • The section introduces illustrative use cases that highlight fog computing’s benefits.
  • The use cases provide the basis for listing evaluation criteria for fog systems.
  • The section proceeds from use-case discussion to criteria for evaluating fog systems.

A. Illustrative Use Cases

The survey illustrates fog computing through CDN and fire-detection use cases, then derives criteria addressing system capabilities and deployment constraints.

  • Illustrative Use Cases: The survey examines CDN and fire-detection-and-fighting use cases among broader fog-computing application scenarios.
  • Fog System Evaluation Criteria: The discussed fog use cases remain hypothetical, although some have been prototyped and commercial fog devices exist.
  • Illustrative Use Cases: Cloud-based CDN surrogates may be too distant from users, degrading video-streaming quality of experience.
  • Illustrative Use Cases: Proactive video caching at access points shortens initial delay and can improve feedback freshness for subsequent users.
  • Illustrative Use Cases: Moving fire-detection or robot-dispatcher components to the fog stratum helps meet tight end-to-end latency constraints.
  • Illustrative Use Cases: Concrete measurements reported for the firefighting scenario show that the fog approach significantly reduces end-to-end delays.
  • Fog System Evaluation Criteria: The evaluation framework covers heterogeneity, QoS, elastic scalability, and federation across fog and cloud resources and providers.

IV. ARCHITECTURES FOR THE FOG SYSTEM

The surveyed fog architectures address heterogeneity, QoS, scalability, mobility, federation, and interoperability with uneven coverage across designs. Evaluation shows recurring strengths in heterogeneity and selective support for scalability or mobility, while several architectures leave federation, interoperability, or scalability insufficiently addressed.

  • Application-specific architectures: Mobile-fog reduces end-to-end latency compared with a pure cloud-based approach, while its latency may fluctuate without QoS management.It supports dynamic workloads, predicts future user locations, and can move application components for early processing.
  • Application-specific architectures: The programming model supports heterogeneous hosting nodes, latency-aware placement, fog-node scaling, and IoT-device mobility, but not fog-node mobility or dynamic orchestration.Developers manually wire component flows, and scalability for IoT devices or fog domains is not considered.
  • Application-agnostic architectures: Yangui et al.’s architecture meets QoS, scalability, federation, and interoperability criteria, but provides no details on fog-resource descriptions or mobile fog nodes.Mobile IoT devices are supported through gateways, and the architecture was validated with temperature sensors and firefighting robots.
  • Application-agnostic architectures: Across three reviewed architectures, all support lifecycle phases, signaling and control interfaces, and scalability, but only Yangui et al. provide cloud–fog federation.The comparison concerns architectures reviewed in this subsection.
  • Application-agnostic architectures: Cardellini et al.’s distributed QoS-aware Storm scheduler meets heterogeneity, QoS, scalability for applications and nodes, mobility, and federation, but not interoperability.Its scalability does not require global knowledge of the entire data-stream-processing system.
  • Communication architectures: Communication architectures use inter- and intra-stratum links, including CoAP-based fog preprocessing, but evaluations identify unmet heterogeneity and QoS-management concerns.In one comparison, direct device communication is fastest, fog transmission is twice as high, and cloud transmission is more than twice the fog case.

V. ALGORITHMS FOR THE FOG SYSTEM

The survey reviews fog-computing algorithms across resource sharing, task scheduling, offloading, and load redistribution, using criteria tailored to each algorithm’s scope. Its review finds strengths in heterogeneity and QoS support but persistent gaps in scalability, mobility, and federation.

  • Algorithms for the Fog System: The survey covers application-agnostic and application-specific algorithms for computing, storage or distribution, and energy consumption.Its algorithmic review includes resource sharing, task scheduling, offloading, and load redistribution.
  • Evaluation criteria: The review evaluates each paper by scope, approach, methodology, contribution, and the criteria it meets, while excluding irrelevant criteria when appropriate.For example, heterogeneity need not apply to a vehicle-routing algorithm running on a node selected by task scheduling.
  • Resource Sharing: Resource-sharing approaches target different objectives, including utility-based node pairing and low-power clustering of fog-enabled small cells.The clustering strategy forms groups of small cells to share resources for offloading mobile-device workloads.
  • Resource Sharing: Compute resource-sharing studies generally meet heterogeneity and QoS criteria, but none meet elastic scalability or mobility support, and federation remains unaddressed.The reviewed works also focus on sharing within fog strata or domains rather than federation across domains.
  • Task Scheduling: Task scheduling determines where tasks execute across IoT/end-user, fog, and cloud strata and which nodes receive particular tasks.Relevant decision metrics include task blocking probability, latency, capacity, and resource availability.
  • Task Scheduling: A lowest-latency policy achieved the lowest blocking probability among three simulated fog-node task-scheduling policies.The alternatives were random selection and assigning tasks to nodes with maximum remaining resources from a candidate list.

7 This work is also discussed in Section V.C

The survey finds that fog-computing algorithms address QoS, energy, latency, and workload distribution, but commonly lack scalability, mobility, and federation support. Across reviewed studies, fog placement can improve latency, energy use, and request satisfaction relative to cloud- or device-centered alternatives.

  • Task scheduling: Task-scheduling algorithms mainly target the fog stratum, while heterogeneity and QoS are commonly addressed; elastic scalability, mobility, and federation remain unmet.The survey concludes that none of the reviewed task-scheduling algorithms meet criteria C3, C4, or C5.
  • Offloading and load redistribution: Fog offloading outperforms cloud offloading or fully local execution in response time and energy consumption, although the evaluation uses a small-scale environment.The studied graph-based approach splits mobile applications between the device and fog nodes while accounting for resource limitations and QoS constraints.
  • Algorithmic strategies: Fog-oriented strategies reduce blocking, latency, or energy consumption, but evaluations often remain limited by small-scale settings, absent mobility, or unconsidered federation.Examples include shortest-path alternatives, coded computation, and fronthaul-aware content delivery strategies.
  • Content delivery and caching: F-RAN analyses identify trade-offs among fronthaul capacity, caching, delivery mode, latency, delivery rate, and power constraints.Low fronthaul capacity favors high-capacity caches for latency, whereas high fronthaul capacity requires both caching and cloud use; hybrid delivery outperforms hard- or soft-transfer modes alone.
  • Energy efficiency: Fog computing improves energy-related outcomes in several studies, including lower latency, power consumption, and CO2 emissions than cloud computing.With 25% of applications requesting real-time services, one simulation reports 30% lower service latency, 42.2% lower power consumption, and more than 50% lower CO2 emissions.
  • Energy efficiency: Across energy-focused contributions, QoS is consistently met, heterogeneity is met by some studies, while scalability and mobility are generally unmet and federation is often irrelevant.The survey concludes that significant energy savings can be obtained in fog systems, but several evaluation criteria remain insufficiently addressed.

VI. CHALLENGES AND RESEARCH DIRECTIONS

The survey identifies unresolved architectural and algorithmic challenges in fog computing and organizes research directions around their evaluation criteria. Key needs include interoperability, federation, mobility, multi-objective optimization, and fog-specific SLA management.

  • None of the reviewed architectures and algorithms meets all identified evaluation criteria, leaving architectural and algorithmic challenges for future research.
  • Architectural challenges and research directions: Common resource ontologies and semantic Web technologies could unify heterogeneous resource representations across fog environments.
  • Algorithmic challenges and research directions: Fog-specific SLA schemas and management techniques must support multiple providers, domains, and business models in dynamic environments.
  • Architectural challenges and research directions: Federated fog systems still lack feasible mechanisms for composing distributed application components across multiple strata and domains.
  • Architectural challenges and research directions: Centralized cloud orchestration can become unscalable and introduce overhead and delays when coordinating remote fog domains, motivating distributed composition engines.
  • Architectural challenges and research directions: Research must address how to dimension and place additional fog nodes, determine usable resources on existing devices, and preserve services as fog nodes move.
  • Algorithmic challenges and research directions: Algorithmic research should integrate bandwidth and resource costs alongside latency, completion time, power consumption, and other potentially conflicting objectives.
  • Architectural challenges and research directions: Fog federation requires mechanisms for coordinating providers with different business models, including pricing, insourcing, and outsourcing decisions.

VII. LESSONS LEARNED AND PROSPECTS

The survey finds that fog computing can reduce latency, cloud-bound traffic, and energy consumption, while these benefits depend on deployment choices and application requirements. It identifies unresolved architectural and algorithmic challenges and highlights fog’s potential role in Tactile Internet systems.

  • Lessons Learned: Fog systems reduce latency relative to traditional cloud systems, with reported reductions of 48% and 73%.The survey cautions that latency reduction depends on application-component placement.
  • Lessons Learned: Local fog processing reduces data transmitted toward the cloud and can save bandwidth by avoiding unnecessary cloud turnaround.The benefit is especially relevant where immediate processing is required, including Industrial IoT and healthcare applications.
  • Lessons Learned: Fog systems are reported to consume less overall energy than cloud systems, implying lower costs.
  • Lessons Learned: Fog computing lacks mature standardization and consortium support beyond OpenFog, motivating reference architectures, developer guides, samples, and SDKs.The survey also identifies needs for semantic Web technologies, monitoring, and reconfiguration mechanisms.
  • Research Directions: Algorithmic research needs deployment-design decisions, broader QoS management, realistic large-scale evaluations, mobility models, and federation algorithms.
  • Prospects: Fog computing is presented as a potential enabler of Tactile Internet through integrated end-user, edge, and cloud strata.The survey connects this prospect to ultra-responsive and ultra-reliable communication, task scheduling, and distributed intelligence requirements.

VIII. CONCLUSION

The paper surveys fog-computing literature by describing systems, distinguishing related concepts, and applying evaluation criteria to architectures and algorithms. It also uses IoT and CDN use cases to support its critical review.

  • The survey derives evaluation criteria from IoT and CDN use cases and applies them to fog architectures and algorithms.
Loading 1710.11001v3…