Source-linked AI summary
How to Place Your Apps in the Fog -- State of the Art and Open Challenges
Antonio Brogi, Stefano Forti, Carlos Guerrero, Isaac Lera
TL;DR
Fog application placement must distribute multi-service components across Fog and Cloud nodes while meeting functional, QoS, cost, and other requirements. This survey reviews existing algorithms and models, examines their evaluation practices, and identifies open challenges including security, mobility, and changing infrastructure conditions.
Problem
Application placement in the Fog is an NP-hard mapping problem requiring trade-offs among multiple objectives and constraints for distributed application components.
Method
The survey searched major scholarly libraries, screened 110 articles, and selected 38 studies to classify algorithms, models, prototypes, use cases, constraints, and optimisation metrics.
Results
The reviewed literature spans search-based and mathematical-programming approaches, with reported improvements in latency, network usage, response time, and resource allocation across varied use-case sizes.
Takeaways & Limitations
Fog placement research must account for where computation is best-placed across distributed infrastructures while balancing application and stakeholder requirements.
Takeaways & Limitations
Security is absent from the surveyed placement models, while mobility and infrastructure variations are considered in very few studies.
Abstract
from arXiv · showhide
Fog computing aims at extending the Cloud towards the IoT so to achieve improved QoS and to empower latency-sensitive and bandwidth-hungry applications. The Fog calls for novel models and algorithms to distribute multi-service applications in such a way that data processing occurs wherever it is best-placed, based on both functional and non-functional requirements. This survey reviews the existing methodologies to solve the application placement problem in the Fog, while pursuing three main objectives. First, it offers a comprehensive overview on the currently employed algorithms, on the availability of open-source prototypes, and on the size of test use cases. Second, it classifies the literature based on the application and Fog infrastructure characteristics that are captured by available models, with a focus on the considered constraints and the optimised metrics. Finally, it identifies some open challenges in application placement in the Fog.
1 INTRODUCTION
Fog computing extends Cloud capabilities toward distributed, heterogeneous devices to place application components closer to data sources and support latency-sensitive, bandwidth-hungry IoT applications. This survey systematically reviews Fog application placement algorithms, models, constraints, metrics, prototypes, experiments, and open challenges.
- Motivation: Bandwidth limitations and latency-sensitive workloads motivate processing data closer to where it is generated.The introduction also identifies reduced connectivity requirements and support for computation or storage near data sources as important needs.
- Fog computing context: Fog computing pushes application components and services toward geographically distributed devices spanning the IoT–Cloud continuum.The architecture includes personal devices, gateways, micro-data centres, and embedded servers.
- Placement challenges: Fog application placement must handle limited and heterogeneous resources, geographic distribution, variable network QoS, multiple operators, and failures.These characteristics distinguish Fog infrastructures from Cloud data centres and complicate application distribution.
- Survey objectives: The survey offers complementary algorithmic and modelling perspectives on Fog application placement.It reviews algorithms, prototypes, and experiments while classifying functional and non-functional constraints and optimisation metrics.
- Survey scope: The paper analyses state-of-the-art contributions and then identifies open problems and future research challenges.The organisation covers survey methodology, algorithmic analysis, modelling analysis, and open challenges.
2 SETTING THE STAGE
The survey formalises Fog application placement as constrained optimisation over mappings between application components and distributed infrastructure nodes. It systematically searches and classifies prior work by algorithms, constraints, objective functions, prototypes, and experimental characteristics.
- Problem formulation: FAPP maps application components to Fog or Cloud nodes while satisfying requirements and optimising objective metrics.Mappings may be many-to-many, allowing components on multiple nodes and nodes to host multiple components.
- Survey methodology: The survey searched scholarly databases and screened 110 collected articles, selecting 38 for detailed analysis after excluding adjacent topics.Excluded areas include offloading, dispatching, scheduling, and VNF embedding.
- Problem formulation: FAPP is NP-hard, with n^m candidate mappings for m application modules and n infrastructure nodes.Heuristics can find sub-optimal solutions, which is especially relevant for large-scale and dynamic Fog deployments.
- Algorithms: The reviewed literature uses search-based and mathematical-programming approaches most often, while bio-inspired, game-theoretical, deep-learning, dynamic-programming, and complex-network methods are less common.The classification organises the algorithmic analysis around the approaches used to solve FAPP.
- Constraints: Constraint taxonomies cover network topology, latency, bandwidth, link reliability, node hardware and software, energy, and application requirements.Node constraints include resources such as processors, memory, and storage, plus software dependencies such as operating systems and libraries.
- Objective functions: Common optimisation objectives include network usage, hardware allocation, energy consumption, and system performance or QoS.The survey also notes that placement can be solved or adjusted at runtime as infrastructure conditions change.
3 ANALYSIS OF THE STATE OF THE ART
The survey finds that Fog application placement research uses diverse search-based and mathematical approaches, while modelling heterogeneous infrastructure, network, energy, hardware, software, and application constraints. Evaluations report improvements across latency, network usage, energy, response time, and service quality, but many approaches remain limited in scale or lack released implementations.
- Algorithms: Distributed search strategies improve placement outcomes by reducing service distance, network usage, or latency under changing Fog conditions.One evaluation reported substantial improvements for frequently requested services, while MigCEP targeted migration and end-to-end latency in a dynamic connected-car infrastructure.
- Algorithms: 40% improvement in response time was achieved by a heuristic over first-fit, scaling to 20000 nodes while exhaustive search handled at most 150 nodes.The prototype implementations for these strategies were not released.
- Considered constraints: Network latency is the most common network constraint, while link reliability is seldom considered; application dependencies are common, but multiple application constraints are rarely combined.Energy matters because Fog infrastructures include power-intensive, mobile, and battery-powered devices, while topology may involve colonies or controller-managed device groups.
- Considered constraints: Most studies model hardware resource capacity, whereas software capabilities are included in only a small number of Fog placement studies.Resource models range from specific CPU, RAM, storage, and bandwidth vectors to general capacities or resource slots; software examples include operating systems, platforms, and frameworks.
4 CONCLUSIONS: OPEN PROBLEMS AND RESEARCH CHALLENGES
The survey identifies open challenges for Fog application placement, including newer and decentralised algorithms, common benchmarks, security-aware and adaptive models, Cloud–IoT integration, and larger-scale reproducible evaluation.
- Algorithms: Future work should assess evolutionary, swarm, deep-learning, network-science, and game-theoretic methods, alongside decentralised online algorithms for dynamic Fog infrastructures.The survey notes that search and mathematical programming dominate existing work, while distributed solutions may improve scalability and resilience.
- Evaluation: Standardised benchmark examples with optimal reference solutions would enable systematic comparison and quantification of performance improvements or degradation.Suggested benchmarks should model attributes such as hardware, software, IoT, QoS, and energy.
- Modelling: Security remains unmodelled in surveyed placement studies, despite the need to verify deployment-node security requirements with well-founded and explainable analyses.The concern is especially relevant to mission-critical applications such as e-health and disaster recovery.
- Modelling: Few studies account for node mobility or changing topologies, traffic, and workloads, motivating adaptive placement models for dynamic Fog conditions.Autonomous vehicles and flying drones exemplify Fog settings with continuously changing connectivity.
- Infrastructure: Research should jointly manage IoT, Fog, and Cloud resources, including QoS-based reachability and integration with production-ready Fog management tools.The survey frames the Cloud-to-Things continuum as a single system requiring simultaneous management.
- Evaluation: Most experiments use small or medium simulated environments and often omit code, creating a need for documented large-scale simulators and real Fog testbeds.Such infrastructure would support evaluation on larger, more lifelike examples and improve reproducibility.