Collaborative Resource Allocation Algorithms Over Hybrid Networks Based On Primal-dual Methods

With the network technology development,the service request processed by cloud center and edge server is becoming more and more popular.This framework incorpo-rates the precise processing power of cloud center with the quick response ability of edge server and is applied in many fields such as video service,surveillance network and Internet of vehicles.From the view of mathematics,edge servers constitute a de-centralized network,and every edge server links with cloud.so on the whole,the cloud is the center of the network.This network combines decentralized network with central-ized network and is called hybrid network.How to design efficient resource allocation algorithms for the hybrid network is the research emphasis of this paper.This paper focuses on resource allocation problem of the hybrid network.This problem is an extension of classic resource allocation problem of the centralized or de-centralized network.We first abstract two-layer hybrid network of edge server and cloud center,three-layer hybrid network of end users,edge servers and cloud center from the practical network.Then,we establish two network utility maximization problems for the two hybrid networks respectively.At last,we solve them by a distributed parallel manner.Specifically,at the beginning,every end users send their request service to the corresponding edge server,then every edge server will consider the computing and storage resource of the neighboring edge servers and cloud center,also the latency of links between them to redistribute the initial amount of service.And this redistributed manner is executed by every node in hybrid network that can take the situation of it-self and its neighbour nodes into account.This is a high self-government,distributed manner.And in this way,the hybrid network can achieve maximal utility with high efficiency.Because the resource allocation problem over a network is a constraint problem.In order to allocate resource by a distributed manner,we have to use primal-dual meth-ods.So,this paper uses two classical primal-dual methods to design two distributed parallel resource allocation algorithms for the two hybrid networks respectively.We first use Alternating Direction Method of Multipliers(ADMM)to design a resource al-location algorithm for the two-layer hybrid network.Then,on the premise of reducing processing time of the algorithm as well as keeping convergence accuracy,we use Pre-dictor Corrector Proximal Multiplier Method(PCPM)to design a resource allocation algorithm for the three-layer hybrid network.At last,we verify the convergence rate and convergence accuracy of the two algorithms by simulation.As an extension,we use linearization to reduce computing amount of the algo-rithms.In the case that the amount of service is discrete,we come up with a approximate projection algorithm.And we show the favourable dynamic property of the algorithms by simulation.At last,we show the algorithms have a nice dynamic property in the situation where the initial amount of service changes over time.