一种面向边缘计算环境的去中心化服务请求分发方法

Decentralized Service Request Dispatching for Edge Computing Systems

边缘计算模式的出现使得软件服务可以部署在距离用户较近的边缘服务器上,从而减小服务调用过程中产生的数据传输开销,并提升服务质量.然而,由于计算和存储等资源的限制,边缘服务器通常只能部署有限数量的服务实例,无法满足复杂多样的服务请求,加之边缘计算环境下服务请求分布不均导致了边缘服务器之间负载的不均衡性.因此,边缘计算环境下需要对服务请求进行再分配来满足不同的服务请求并优化服务供应.如何合理地将服务请求分发到合适的边缘服务器以优化系统的负载均衡并提升服务质量成为亟待解决的关键问题.现有的研究方法通常采用集中式的方法来解决该问题,存在单点故障等严重缺陷.因此,本文基于博弈理论设计了一种面向边缘计算环境的去中心化服务请求分发方法.该方法将边缘服务器间的服务请求分发问题建模为分布式的非合作博弈模型,通过多轮次的博弈和竞争达到系统的Nash均衡状态,以此获取服务请求的分发策略.实验结果表明,本文所提出的方法可以有效优化边缘服务器之间的负载均衡,降低服务的响应时间,并且随着系统规模的扩大表现出良好的可扩展性.

Edge computing paradigm enables software services to be deployed on edge servers that are close to users,thereby reducing the data transmission costs produced during service invocation processes and improving the quality of services.However,due to the limited computing and storage resources,individual edge servers can typically be deployed with only a limited number of service instances and thus cannot satisfy diverse service requests.Moreover,service requests are distributed unevenly in edge computing systems,which leads to an imbalance in the load of edge servers.To solve these problems,a collaboration mechanism has been proposed to bring neighboring edge servers together as a cohort to provide services for their users collaboratively.In this mechanism,on receiving service requests,edge servers re-dispatch them to appropriate servers with the objective to satisfy these diverse service requests and to optimize their quality.Thus,how to reasonably dispatch service requests to appropriate edge servers to optimize the load balancing among edge computing systems as well as to improve the quality of services has become a key problem to be solved.Existing research methods usually adopt centralized methods to solve this problem,which assume that there is a central node in each edge computing system that can get the global information of the system such as the available resources of all edge servers,service execution status,arriving service requests and network conditions between edge servers,and use this information to calculate the optimal request dispatching strategy.However,centralized methods rely heavily on central nodes and thus have serious defects including single point of failure,etc.Therefore,this paper utilizes decentralized mechanism for service request dispatching in edge computing systems.In decentralized frameworks,due to the lack of central nodes,it is hard to get the global real-time information of the system.Hence,centralized methods are not available for decentralized systems,and edge servers can only obtain the status and request information of each other through continuous communication.In this paper,we propose a game theory-based method to solve this problem.Service request dispatching among edge servers is modeled as a distributed non-cooperative game,where edge servers,as the game participants,iteratively communicate with each other and make their own decisions for their received requests according to the status and decisions of each other,with the aim to minimize the average response time of their received service requests.Through multiple rounds of mutual game and competition,these edge servers can finally achieve the Nash equilibrium state,thereby obtaining the optimal service request dispatching strategies.We present the game model and theoretically prove that it can achieve the Nash equilibrium.Moreover,we split service response time to data transmission time and service execution time and give the calculation method of them based on Shannon's Theorem and Queuing Theory,respectively.The experimental results show that compared with benchmark methods,the proposed method can greatly reduce the response time of services,effectively optimize the load balancing between edge servers and perform good scalability as the scale of the system increases.