Research On Resource Allocation And Cooperative Control In Edge Networks

For the surge of connected devices to network and the development of innovative applications,data traffic increases exponentially.It is also shown that services have different demands and performance requirements.Since cloud computing platform locates remotely from users,users need to transmit data over long path.The frequent communications between the centralized cloud and devices also increase the burden on backhaul traffic.Thus,it is difficult for the cloud computing platform to satisfy the requirements of real-time interaction,higher data rate and lower energy consumption of services.In terms of this,edge computing paradigm emerges,which brings cloud servers with computing capacity and caching resources at base stations in radio access networks(RAN).It is seen as a promising technology for computation services,as resource-constrained devices can offload computing tasks to edge computing servers for lower computing delay as well as lower energy consumptions.In addition,by proactively caching contents which may be requested in the future at edge computing servers,it can remarkably reduce the downloading delay for contents.Edge network integrates edge computing and RAN,and forms a resource pool with computing resources,caching resources,radio resources and decentralized devices.Based on the development of ultra-dense networks and device-to-device(D2D)communications,edge networks become distributed and heterogenous.Then,it is challenging to effectively schedule and control the hierarchical and distributed architecture of base stations and devices,as well as efficiently allocate diverse resources for services.According to these issues,this dissertation studies the joint scheme of resource allocation and cooperative control based on the diverse requirements of services and multiple choices of communications in edge networks.The research conclusions provide theoretical analysis and practical methods for future exploration on cooperative management of edge networks.The main contributions and innovations are shown as follows.1.Resource allocation and cooperative control scheme of edge computing for multiple servicesWhen computation service and content delivery service execute in edge networks,they compete for the resources in edge computing server.In this dissertation,in order to improve the delay-aware performance of multiple services,the resource allocation strategy of computing resources and caching resources is studied.Furthermore,in the multi-cells network,in terms of load balance across cells,the joint strategy of resource allocation and user association is studied to reduce the sum of delays.The main innovations are shown as follows.Firstly,in a single cell scenario,the problem of computing and caching resource allocation of edge computing server is formulated for computation service and content delivery service.To solve the problem efficiently,an alternative algorithm of multi-dimensional resource allocation is proposed to improve the delay-aware performance.Secondly,in a multi-cells scenario,according to the network status(e.g.,redundant resources,channel status)and unbalanced services' requests in different cells,the joint problem of resource allocation and user association assignment is addressed to minimize the weighted sum of delay.With convex optimization methods,a Coalition-Game-Based Algorithm(CGBA)is then proposed to obtain the scheme of resource allocation and user association.Simulation results show that the proposed algorithm reduces the weighted sum of delay by average 82.1%compared with the NEarest Assignment Scheme(NEAS).It also maintains the delay-aware fairness for users based on cooperative resource management across different cells.2.Resource allocation and cooperative control scheme of heterogeneous edge networksBoth uplink/downlink communications to base station and D2D communications exist in edge networks.In terms of the heterogeneity of communications,this dissertation studies the strategy of resource management based on users' collaboration for computation service and content delivery service,respectively.The main innovations are shown as follows.Firstly,under the consideration of sharing property of computation applications and users'collaboration,a joint optimization problem of task offloading and resource allocation is addressed to minimize the energy consumption with delay constraints for tasks.Then,an algorithm of Joint Resource Allocation and Offloading Assignment(JRAOA)is proposed,and numerical results shows it has benefits on energy saving compared with Non-Collaborative Resource Allocation and Offloading Assignment(NCRAOA)scheme.Secondly,for the scenario with both edge caching and device caching,the problem of proactive caching and content sharing for mobile user equipments is formulated to minimize the delivery costs of contents under the consideration of the social relationship and mobility patterns of users.Based on the long-term content requests of users,an incentive mechanism is designed to encourage local cooperation among users,and a Distributed Proactive Caching Scheme(DPCS)is obtained using game theory.Numerical results show the advantages of local cooperation on cost reduction of content delivery.3.Similarity-based caching and delivery scheme in arbitrary caching networksFor similarity-based caching,there exists a tradeoff between delivery delay reduction and the increase of dissimilarity cost.With the consideration of coordination across the caches in arbitrary caching networks,this dissertation studies the strategy of similarity-based caching and delivery decisions to minimize content delivery delay,while also satisfy the similarity requirement of delivered files.The main innovations are shown as follows.Firstly,the dissimilarity matrix between requested files and delivered files is modelled.For the scenario in which the requests rates are known,the offline optimization problem of cache allocation and delivery decision is formulated,such that the weighted sum of delay and dissimilarity cost is minimized.A variant of gradient descent ascent algorithm,i.e.,HiBSA,is used to solve the minimax primal-dual formulation of the relaxed optimization problem and obtain the offline scheme.Numerical results show that when dissimilarity sensitivity of files is low,the offline scheme obtains much lower average delay compared to exact caching scheme,and as dissimilarity sensitivity of files increases,the offline scheme converges to exact caching scheme.Secondly,for the scenario in which the arrival rates of requests are a priori unknown,an online stochastic version of the gradient descent ascent algorithm is proposed,such that the scheme of cache allocation and delivery policy is updated according to the requests observed over time.Numerical results show that the online collaborative algorithm gains performance improvement of delay by nodes collaboration,and achieves lower average delay compared with the up-to-date per-cache similarity-based strategy.