Source-linked AI summary
Game Theory for Multi-Access Edge Computing: Survey, Use Cases, and Future Trends
Jose Moura, David Hutchison
TL;DR
Wireless MEC networks must coordinate scarce resources and conflicting player objectives while handling latency, energy, information, and dynamic-behavior constraints. This paper surveys theoretical games applied to wireless networking and MEC, organizing the literature by game type and operational design aspects. It concludes that coalitional and evolutionary models are especially suitable for future MEC directions because of scalability, energy savings, adaptation, and learning.
Problem
Wireless MEC services require resource management under latency, energy, information, and dynamic-network constraints across diverse use cases.
Method
The paper surveys GT contributions for wireless communications and MEC, comparing classical and evolutionary games and examining cooperation, learning, information, and network scenarios.
Results
The survey concludes that coalitional and evolutionary game models are most suitable for future MEC research directions.
Takeaways & Limitations
Coalitional models offer scalable clustering and energy savings, while evolutionary models adapt to unexpected network behavior through learned strategies.
Takeaways & Limitations
The reviewed approaches face boundaries including complete-information requirements, hierarchical decision-making risks, high convergence time, and limitations in surveyed network settings.
Abstract
from arXiv · showhide
Game Theory (GT) has been used with significant success to formulate, and either design or optimize, the operation of many representative communications and networking scenarios. The games in these scenarios involve, as usual, diverse players with conflicting goals. This paper primarily surveys the literature that has applied theoretical games to wireless networks, emphasizing use cases of upcoming Multi-Access Edge Computing (MEC). MEC is relatively new and offers cloud services at the network periphery, aiming to reduce service latency backhaul load, and enhance relevant operational aspects such as Quality of Experience or security. Our presentation of GT is focused on the major challenges imposed by MEC services over the wireless resources. The survey is divided into classical and evolutionary games. Then, our discussion proceeds to more specific aspects which have a considerable impact on the game usefulness, namely: rational vs. evolving strategies, cooperation among players, available game information, the way the game is played (single turn, repeated), the game model evaluation, and how the model results can be applied for both optimizing resource-constrained resources and balancing diverse trade-offs in real edge networking scenarios. Finally, we reflect on lessons learned, highlighting future trends and research directions for applying theoretical model games in upcoming MEC services, considering both network design issues and usage scenarios.
1 Introduction
MEC brings data and services closer to users to reduce latency and backhaul overload, while introducing resource-management challenges for wireless networks. The paper surveys how Game Theory can address these challenges across diverse MEC scenarios, game models, and research directions.
- Survey scope: Game Theory models interactions among self-interested players competing for common scarce resources to seek stable resource allocations satisfying service requirements.The paper frames GT as a tool for analyzing interactions among independent players in wireless systems.
- MEC motivation: MEC moves services from remote clouds toward network-edge devices, reducing user-service latency and backhaul-link overload.Caching at the network periphery satisfies many requests locally.
- MEC motivation: Edge networking must manage metadata, context, environment information, and reactive or proactive data dissemination under changing conditions.Relevant information includes QoS/QoE, battery autonomy, popularity, location, mobility, and social relationships.
- Research question: The paper asks how Game Theory can help manage wireless resources for MEC requirements including energy efficiency, virtualization, storage, and processing.The discussion also considers mobile cloud computing, IoT integration, 5G access, WAVE, and SDN.
- Survey scope: The survey reviews classical and evolutionary games, cooperation, learning, incomplete information, and applications across heterogeneous access, small-cell, D2D, vehicular, and IoT networks.Its stated contributions include a GT taxonomy, comparisons of game types, and discussion of MEC architectures and future directions.
2 Background on Game Theory and Multi-Access Edge Computing
This section provides the background on Game Theory and Multi-Access Edge Computing needed for the paper's later discussion.
- Background: The section supplies foundational background on Game Theory and Multi-Access Edge Computing for non-specialist readers.The paper uses this background to support its subsequent review and analysis.
2.1 Game Theory Overview and some Relevant Enhancements for Multi-Access Edge Computing
The section introduces game-theoretic models for analyzing strategic resource allocation, then distinguishes classical and evolutionary games by rationality, information, interaction structure, cooperation, and learning. It connects these model choices to MEC coordination, efficiency, and practical constraints such as signaling overhead, scalability, and energy use.
- Game definition: A classical game models players, available actions, and utility functions in normal form, with payoffs constrained by all players’ decisions.The tuple is (N, A, u), where N is the player set, A the action profiles, and u the payoff functions.
- Equilibrium and efficiency: Nash equilibrium identifies strategy profiles from which no player benefits by unilateral deviation, whereas Pareto optimality evaluates allocations that cannot improve one player without worsening another.The Price of Anarchy compares the costs of Nash and Pareto designs, highlighting inefficiency associated with selfish behavior.
- Game timing: Stackelberg and repeated games represent sequential decision-making, with leaders or earlier actions influencing followers and later responses.Repeated games retain a history of decisions, allowing current actions to affect future behavior and potentially enforce cooperation.
- Classical and evolutionary games: Classical games assume rational choices among static strategies, while evolutionary games allow limited information, incomplete rationality, and strategies that evolve over time.The survey compares these perspectives as distinct approaches to modeling game solutions and player behavior.
- Game taxonomy: Non-cooperative games model individual players seeking their own utility, while cooperative games model groups sharing objectives and balancing coalition stability against network efficiency.Cooperation in non-cooperative games must be self-enforced, whereas cooperative games explicitly organize players into groups.
- Information and MEC constraints: Bayesian games address environments where players lack complete information, while distributed coordination may require signaling that overloads resource-constrained networks.The survey also notes that evolutionary models can support decentralized cooperation in MEC scenarios, although larger iterated games are harder to analyze for all players.
- Learning and practical constraints: Evolutionary learning can support decentralized cooperation and mitigate selfish routing behavior, but genetic algorithms may be unsuitable for real-time wireless applications because convergence and energy costs degrade performance.The survey reports reinforcement learning as a better-performing alternative than genetic algorithms for some wireless sensor-network functions.
2.2 Multi-Access Edge Computing Foundational Aspects
MEC places storage, computation, and dissemination capabilities at the network edge to support low-latency services and reduce backhaul load. Its use cases span consumer, operator, third-party, and network-performance services, each imposing distinct resource, mobility, and latency constraints.
- Foundational aspects: MEC extends cloud services to the network periphery, providing distributed storage, computational capabilities, and data dissemination.The paper describes MEC as supporting ubiquitous and efficient access to a small set of edge resources.
- Service categories: MEC use cases are organized into consumer-oriented, operator and third-party, and network performance and QoE improvement categories.The paper associates these categories with ETSI service classifications and corresponding design constraints.
- Consumer-oriented services: Consumer-oriented applications include edge gaming, augmented reality, assisted reality, virtual reality, and cognitive assistance.These applications require nearby computation and storage, often with short response times and support for user mobility.
- Consumer-oriented services: Personalized MEC sessions must remain continuous as users move between cloud, WiFi, and MEC environments.The described service session follows the mobile user from indoor cloud access to the MEC server serving the outdoor location without disruption.
- Operator and third-party services: Connected-vehicle MEC caches data and services near moving vehicles to reduce access RTT and disseminate safety warnings with very low delay.Roadside infrastructure can analyze vehicle data and propagate latency-sensitive hazard messages to nearby cars.
- Network performance and QoE improvements: Network-performance use cases apply edge resources to applications such as video management and analytics, while imposing explicit system design constraints.Video analytics processes camera streams at the MEC server and forwards low-bandwidth metadata for centralized searches.
3 Review of Theoretical Model Games and Multi-Access Edge Computing
The survey reviews classical and evolutionary game models for wireless networking scenarios aligned with MEC, emphasizing cooperation, learning, incomplete information, and resource-management trade-offs. It compares applications, benefits, and limitations across non-cooperative, Stackelberg, coalition, and evolutionary approaches.
- Survey scope: The survey broadens prior GT–MEC coverage beyond pricing models to review diverse research directions, use cases, learning, cooperation, and social connections.It considers heterogeneous access, small cells, D2D communications, vehicular networks, IoT, and energy consumption.
- Non-cooperative games: Non-cooperative games support robust distributed control with reduced signalling, but selfish behavior and distributed decisions can hinder global optimization.The survey discusses cooperation incentives and learning as mitigation strategies, including low-complexity control for energy–performance trade-offs.
- MEC applications: The reviewed models target MEC challenges involving resource constraints, incomplete information, mobility, cooperation, and trade-offs between energy efficiency and network performance.The survey positions game models as tools for configuring and optimizing network algorithms across diverse edge scenarios.
- Stackelberg games: Stackelberg games optimize virtualized computation, storage, and networking resources under Quality of Experience in hierarchical edge topologies.Their limitations include leader–follower synchronization, possible Stackelberg Equilibrium underperformance relative to Nash Equilibrium, and the need for complete information.
- Cooperative games: Coalition games divide large systems into smaller coalitions to simplify operation, supporting cooperation in vehicular, D2D, femtocell, and multi-hop networking scenarios.Vehicular coalition algorithms may face exponentially increasing convergence time, while dynamic mobility can complicate coalition formation.
- Evolutionary games: Evolutionary games let limited-rationality players learn from the environment and adapt strategies in distributed fashion across resource management, small-cell interference, and VANET scenarios.A VANET study reports that topologies with more clusters increase vehicle connectivity and cooperation, while evolutionary approaches can require high convergence time.
4 Research Directions for Theoretical Model Games and Multi-Access Edge Computing
The paper identifies research directions for applying theoretical games to dynamic MEC environments, including congestion mitigation, virtualization, social networking, and cognitive edge functions. It emphasizes multi-game and agent-based designs for heterogeneous systems with constrained resources and competing trade-offs.
- Future research directions: Future MEC game-theoretic directions include augmented reality, cognitive assistance, fair virtual resource allocation, seamless session transfer, and congestion control.These directions require location and environmental information, elastic storage and computing, communications capacity, traffic steering, and intelligent agents.
- Future research directions: Traffic offloading, D2D communication, relaying, pricing under incomplete information, and full-duplex communications are highlighted as congestion-mitigation technologies for MEC.Full-duplex communication is associated with improving upcoming 5G wireless access performance.
- Architectural challenges: A single game is difficult to use for heterogeneous MEC systems because they combine dynamic behavior, constrained resources, uncertain information, and simultaneous trade-offs.The cited trade-offs include energy versus relaying traffic, computation offloading versus channel interference, and data consistency versus latency.
- Architectural challenges: The proposed alternative is to run several virtual machines at MEC servers, supporting a hierarchical design for network domains such as D2D, vehicular, and sensor networks.The design also includes communities of people, devices, or systems connected by strong relationships, forming a basis for mobile social networking.
- Cognitive networking: Community devices can run agents in virtualized containers associated with edge providers, allowing the network edge to transform information into network knowledge.The paper connects this cognitive network with theoretical games, evolutionary algorithms, and artificial intelligence to infer human behavioral patterns.
- Cognitive networking: The paper anticipates gradual changes from IPv4 and client-server communication toward IPv6, Named Data Networking, publish-subscribe, peer-to-peer, and multicast communication.Forwarding would shift from interface IP addresses toward names associated with data chunks.
5 Summary and Future Trends
The paper concludes that coalitional and evolutionary games are especially suitable for future MEC research, while Bayesian games fit settings with limited or variable information. It identifies broad opportunities for applying GT to low-latency access, offloading, caching, IoT, and localization, while noting that the practical impact of theoretical games remains to be clarified.
- Summary and future trends: Coalitional and evolutionary games are identified as the most suitable models for future MEC research.Coalitional models offer scalable hierarchical clustering and potential energy savings, while evolutionary models adapt using previously successful strategies.
- Summary and future trends: Bayesian games are considered promising for scenarios with limited or highly variable context information, whereas non-cooperative models are judged inefficient for future MEC services.The paper attributes the non-cooperative limitation to the absence of learning and player selfishness, while noting cooperation incentives as an augmentation.
- Future opportunities: MEC offers research opportunities in low-latency access, distributed offloading, proactive edge caching, low-power IoT networking, and fingerprinting localization.The proposed caching factors include data popularity, social user connections, and available node battery energy.
- Future opportunities: Learning is presented as a promising route for improving game-theoretic predictive power while keeping models precise, limited in complexity, reproducible, and testable.These criteria are described as facets that make GT more useful for research investigations.
- Future opportunities: The real impact of theoretical model games on MEC services coordinated between edge networks and remote clouds remains to be clarified.