Source-linked AI summary
Application Management in Fog Computing Environments: A Taxonomy, Review and Future Directions
Redowan Mahmud, Kotagiri Ramamohanarao, Rajkumar Buyya
TL;DR
IoT applications need timely processing, but Cloud distance and Fog heterogeneity complicate application execution and management. The paper reviews application management across architecture, placement, and maintenance, develops taxonomies and a perspective model, and identifies research gaps and future directions. Its review addresses shortcomings in earlier surveys by treating these aspects collectively and discussing distributed application management.
Problem
IoT-enabled CPSs require real-time processing, while Cloud distance and Fog nodes’ distribution, heterogeneity, and resource constraints complicate effective application management.
Method
The paper reviews Fog application management across architecture, placement, and maintenance, proposes separate taxonomies, and discusses a logically distributed management framework.
Results
The review identifies research gaps in Fog-based application management and presents a perspective model and future directions for improving application management.
Takeaways & Limitations
Application management in Fog requires coordinated consideration of application architecture, placement, maintenance, and the differing requirements of IoT-enabled CPS applications.
Takeaways & Limitations
The paper identifies unresolved limitations including uncertain renewable-energy availability, increased decision-making time from distributed placement, and missing standards for integrating Fog with other paradigms.
Abstract
from arXiv · showhide
The Internet of Things (IoT) paradigm is being rapidly adopted for the creation of smart environments in various domains. The IoT-enabled Cyber-Physical Systems (CPSs) associated with smart city, healthcare, Industry 4.0 and Agtech handle a huge volume of data and require data processing services from different types of applications in real-time. The Cloud-centric execution of IoT applications barely meets such requirements as the Cloud datacentres reside at a multi-hop distance from the IoT devices. \textit{Fog computing}, an extension of Cloud at the edge network, can execute these applications closer to data sources. Thus, Fog computing can improve application service delivery time and resist network congestion. However, the Fog nodes are highly distributed, heterogeneous and most of them are constrained in resources and spatial sharing. Therefore, efficient management of applications is necessary to fully exploit the capabilities of Fog nodes. In this work, we investigate the existing application management strategies in Fog computing and review them in terms of architecture, placement and maintenance. Additionally, we propose a comprehensive taxonomy and highlight the research gaps in Fog-based application management. We also discuss a perspective model and provide future research directions for further improvement of application management in Fog computing.
1 INTRODUCTION
IoT-enabled cyber-physical systems require timely, low-congestion processing, but Cloud distance and Fog heterogeneity make application management challenging. The paper reviews application management through architecture, placement, and maintenance, proposing taxonomies, a perspective framework, and future research directions.
- Motivation: IoT devices generate continuous data, while resource-constrained devices and time-sensitive CPS applications require responsive processing and uncongested networks.These requirements make heavyweight application execution on many IoT devices unsuitable.
- Fog Computing Context: Cloud datacentres can degrade service delivery because their multi-hop distance from IoT devices increases data-transfer time and network load.Fog computing addresses this by executing applications near data sources, reducing propagation delay and conserving bandwidth.
- Fog Computing Context: Fog computing uses distributed edge components such as gateways, personal computers, Raspberry PIs, nano-servers, and micro-datacentres to execute IoT applications near data sources.This architecture can improve application latency, reduce congestion, lower provider costs, and support location-aware communication.
- Application Management: Application management must address composition, placement, and execution because CPS applications differ substantially in service deadlines, workloads, and data volumes.The paper frames these decisions as aligning application characteristics with Fog infrastructure and platform services.
- Challenges: Fog application management remains constrained by heterogeneous, resource-limited nodes, uncertain failures, missing integration standards, interoperability problems, and absent business models.These issues complicate platform-independent execution, recovery, cross-paradigm services, application customization, and adoption.
- Existing Strategies: Existing strategies modularize applications, use web-service communication, optimize deadlines, cost, and energy, synchronize distributed nodes, and apply proactive or reactive fault tolerance.These strategies target resource constraints and reliable execution across distributed Fog nodes.
- Paper Contributions: The paper reviews application management by architecture, placement, and maintenance, proposes taxonomies and a logically distributed framework, and identifies research gaps and future directions.It positions the review as a comprehensive treatment of these aspects rather than a survey of only one management dimension.
2 BACKGROUND STUDY
Fog, Edge, and Mist computing all move processing toward data sources, but differ in execution location, capabilities, and management scope. Existing surveys address related aspects, yet comprehensive application-management coverage remains limited.
- Mist processes data within IoT devices, whereas Edge uses gateways and Fog uses edge-network components near data sources.
- Mist nodes have limited capacity for large-scale applications, while Edge management is user-centric and fairness across users is difficult.
- Definitions of Mist, Edge, and Fog remain inconsistent, with Edge described variously as a Fog subset, a superset, or an interchangeable term.
- Existing surveys cover resource management, application services, communication, security, data, actuation, offloading, and domain-specific Fog applications.
- The reviewed literature does not comprehensively examine application architecture, placement, and maintenance together, motivating this taxonomy and framework.
3 APPLICATION ARCHITECTURE
Application architecture in Fog is organized around functional layout, program model, service type, workload, and interaction methods. These elements guide distribution, resource provisioning, placement, and maintenance across constrained Fog environments.
- Application architecture should match Fog-environment specifications because its elements influence placement strategy, resource consumption, and service exposure.
- Functional Layout: Functional layouts are monolithic or distributed; distributed applications separate operations into programs that can run on one or multiple Fog nodes.
- Functional Layout: Distributed applications use tightly coupled modules or loosely coupled micro-services that can be attached to other applications.
- Program Model: Thread, map-reduce, and dataflow program models support concurrency, chunk-level parallelism, or DAG-linked operations with differing input characteristics.
- Service and Interaction: Application services include invocation, display, actuation, and storage, while interactions use shared databases or lightweight publish-subscribe messaging.
4 APPLICATION PLACEMENT
Application placement maps workloads to Fog resources while accounting for future commitments, resource requirements, offloading direction, resource organization, and controller location. These choices shape task distribution and application-management responsiveness.
- Considering future Fog-node processing commitments during placement can substantially address task distribution and support proactive application maintenance.
- Resource Estimation: Resource estimation classifies applications as compute-, I/O-, or disk-intensive to align placement with Fog-node capacity constraints.
- Resource Estimation: Profiling monitors execution parameters on each Fog node, whereas prediction uses past execution patterns and scales better but may be less precise.
- Resource Estimation: On-demand estimation derives resources from user expectations and instantaneous demand rather than application characteristics.
- Offloading: Offloading may be bottom-up from devices, top-down from Cloud, or hybrid among Fog nodes, depending on workload and network overhead.
- Resource Orientation and Controllers: Resource orientations include hierarchical, clustered, client-server, and master-slave arrangements, while controllers may be centralized, end-device-based, or broker-based.
4.5 Mapping Technique
Fog placement policies map applications to Fog nodes or virtualized instances using prioritization, optimization, or multi-objective trade-offs. These techniques differ in objectives, solution quality, and runtime.
- Placement policies map applications to Fog nodes and virtualized instances according to selected parameters.
- Prioritization uses heuristics such as best fit and first fit to select preferred Fog nodes or virtualized instances.
- Optimization maximizes or minimizes one objective and can provide a mathematically best solution, but requires more operating time than prioritization.
- Multi-objective Trade-off: Multi-objective trade-offs jointly optimize metrics such as time-energy, time-cost, and cost-QoE using meta-heuristic approaches.
4.6 Placement Strategy
Fog application placement is categorized by how often placement decisions are made during application execution. The taxonomy distinguishes static, dynamic, and event-driven strategies.
- Static placement: Static placement executes the placement algorithm once and keeps the application running on the selected host.
- Dynamic placement: Dynamic placement reruns the placement algorithm for every input arrival, supporting replicas or applications that terminate after one input.
- Event-driven placement: Event-driven placement performs additional scheduling after initial placement when mobility, preemption, node consolidation, or service migration occurs.
4.7 Resource Type
Fog application placement uses three resource types: bare metal, virtual machines, and containers. They differ in virtualization, isolation, and portability.
- Bare metal: Bare-metal placement gives applications direct access to Fog-node hardware through the host operating system without explicit execution-unit isolation.
- Virtual machines: Virtual-machine placement uses hardware-level virtualization and hypervisors to let multiple operating systems run independently on one Fog node.
- Containers: Container placement provides lightweight operating-level virtualization by isolating application processes and required packages while enabling portability across Fog nodes.
4.8 Placement Metric
Placement metrics define application-placement objectives according to user, service-provider, and physical-environment requirements. The reviewed metrics span timing, economics, context, energy, resources, mobility, and user experience.
- Time and deadline: Time and deadline metrics minimize service-delivery time while meeting specified deadlines, considering computation, propagation, and node-deployment time.
- Economic objectives: Profit and cost metrics respectively target provider revenue and minimization of infrastructure, operational, and instance-rental expenses.
- User experience: User-experience metrics account for changing service requirements and affordability because unmet expectations can degrade experience and discourage future Fog use.
- Context: Context metrics incorporate external factors such as relinquish rate, user activity, node reliability, service popularity, data size, and sensing frequency.
- Energy and resources: Energy and resource-status metrics address Fog-node consumption and heterogeneity in processing power, interfaces, storage, and operating platforms.
- Mobility: Mobility is considered because movement of IoT devices and Fog nodes can affect service delivery and trigger application migration.
4.9 Research Gaps in Application Placement
The review identifies unresolved application-placement gaps despite extensive prior research. In particular, it highlights uncertainty in renewable-energy availability and overlooked decision-time costs from distributed management.
- Overview: Table 4 summarizes existing application-placement techniques, while the review notes remaining research gaps in this area.
- Renewable energy: Few studies address the uncertainty and environmental dependence of renewable-energy availability when powering Fog nodes in remote areas.
- Distributed placement: Existing distributed placement approaches can reduce management overhead but have not considered the resulting increase in decision-making time.
5 APPLICATION MAINTENANCE
Application maintenance in Fog spans security, performance monitoring, monetary support, and resiliency strategies. The reviewed research also leaves unresolved trade-offs involving security overhead and resource reservation.
- Security: Security maintenance addresses threats to Fog infrastructure through data integrity, encryption, and authentication.These measures target information impairment, identity disclosure, replay, denial-of-service, unauthorized access, and data alteration.
- Performance Monitoring: Performance monitoring supports application maintenance through predictive implication-based techniques and threshold-based reactive techniques.Predictive monitoring uses application and resource states to anticipate trends, whereas reactive monitoring compares indicators such as processor usage and memory consumption against thresholds.
- Monetary Support: Monetary support includes compensation, incentives, and reservation to address service violations, resource participation, and subscription-based provisioning.Reservation provides a fixed number of applications at fixed charges regardless of current Fog load.
- Resiliency Strategy: Resiliency strategies continue application execution after failures through backup-restore, replication, and operator migration.Backup-restore stores intermediate results, replication runs application instances across nodes, and migration shifts execution dynamically between nodes.
- Research Gaps: Existing maintenance research must balance heavyweight security techniques against legitimate access time and reserved streaming resources against waiting time for less-interactive applications.The review identifies both concerns as unresolved issues in efficient Fog application maintenance.
6 A PERSPECTIVE APPLICATION MANAGEMENT FRAMEWORK FOR FOG
The perspective framework distributes application management across IoT application brokers, Fog and Cloud resource managers, and infrastructure levels. It offers an abstract basis for orchestration, provisioning, migration, and standardization policies.
- Fog Application Management Framework: IoT Application Brokers receive application specifications from CPSs, including workload, data frequency, service form, and QoS requirements.The requirements include service deadline, budget, and user expectations.
- Fog Application Management Framework: At the Fog gateway, an IAB combines a CPS Manager, Application Placement Engine, and Workload Scheduler.The CPS Manager stores application variants and obtains resource-state information through interaction with the placement engine.
- Fog Application Management Framework: Fog and Cloud resource managers store executables, allocate execution resources, monitor performance, and perform backup, replication, and failure-related maintenance.They also address resource outages and security attacks to support reliable application execution.
- Framework Scope: The framework is an abstract view of distributing application management across infrastructure levels rather than a complete operational specification.Its stated uses include developing policies for runtime orchestration, multilevel provisioning, execution migration, and Fog standardization.
7 FUTURE RESEARCH DIRECTIONS
The paper identifies future directions for adapting Fog application management to energy, intelligence, economic, trust, scalability, application-specific, and reuse requirements. These directions respond to resource constraints and heterogeneous IoT workloads.
- Trade-off between energy and accuracy: Future policies could dynamically tune application accuracy and IoT sensing frequency to meet Fog energy constraints.The proposed direction concerns the relation among application accuracy, sensing rate, and Fog-node energy consumption, including renewable-power settings.
- Artificial intelligence-based application management: Artificial intelligence-based management could predict resource requirements, context variation, and nodal failures more precisely.The paper notes that Fog can readily accumulate the training data required for intelligent systems.
- Pricing and detailed estimation of Fog resources: Fog pricing and resource estimation should account for localized demand, distributed deployment, CPS device counts, and future service requirements.Cloud subscription pricing cannot be directly transferred to Fog, where these conditions can also produce resource over-provisioning.
- Trusted service orchestration in Fog: Trusted orchestration is needed to support collaboration and reliability across public and private Fog infrastructures with different trust concerns.Public infrastructure is exposed to security threats, whereas private infrastructure may lack transparency.
- Fog node consolidation and scaling: Dynamic Fog-node consolidation and scaling could match computational demand while limiting the costs and interference associated with adding nodes.Adding nodes can increase deployment cost, communication interference, and edge energy consumption.
- Application-specific management: Application-specific policies could provide specialized support for complex IoT applications across healthcare, cities, agriculture, and industry.The paper links this direction to differing requirements among application domains.
- Task sharing and re-usability: Task sharing and executable reuse could optimize Fog computational load through shared caching techniques and policies.Applications may share tasks, and recently terminated application executables may be reused by other applications.
8 SUMMARY AND CONCLUSIONS
The paper reviews Fog application management through architecture, placement, and maintenance, proposes taxonomies and a perspective model, and identifies research directions for further improvement.
- Summary and Conclusions: The review organizes Fog application management around application architecture, placement, and maintenance.It also discusses associated research gaps and proposes separate taxonomies for these aspects.
- Summary and Conclusions: The paper presents a perspective model for managing applications in Fog environments and identifies directions for further research.The stated goal is to support continued improvement of Fog application management.