Flow-oriented Path Selection And Scheduling In Data Center Networks

With the growing of business needs,data center has become the pillar of the new information industry as a application service providers.It has long been payed closely attention and researched by academia industry.With the increase of traffic volume,the characteristic of resource sharing also brings the problem of resource competition in the data center network(DCN)when the network resources can not guarantee the excessive supply.Therefore,the network resource management in the data center network is very important for optimizing the performance of DCN.In recent years the data center's structure,function,scale and others rapid changes,the traditional way of resource allocation can't be fully adapt to the current data center.However DCN must be highly reliable and delivers satisfactory performance thusing the size of services that data center support are increasing growing.1.These three problems: the multi-constrained multi-path routing algorithm in traditional network can not take advantage of the topological characteristics of SCDC,the multi-constrained routing algorithm in the current SCDC don't provide multi-path,the multi-path routing algorithm in SCDC only consider hopcount are resulting in can't guarantee the good performance of the path and can't use data center network resources efficiently.In this paper,this paper bring the multi constraints multi path problem and multi constrained optimal path problem into SCDC structure.This paper designs the neighbor matrix with weight vector and establishes a multi constrained routing algorithm in SCDC model to evaluate the quality of the path by use of multi constrained(hop,delay,bandwidth,link reliability parameters).On the basis of this design,this paper put forward a multi constrained routing path resource management algorithm.This paper convergence the algorithm by using memory simulated annealing algorithm,it's also can help to find the optimal path of research process.Finally,it is proved that the algorithm reduces the average path cost by 12% when compared to TS_MCOP and reduces the delay by 17% when compared to MPTCP.2.By studying the traffic distribution in the data center,this paper find the traditional routing algorithm can not solve the problem of the low resource utilization and load imbalance witch is brought by elephant collisions in the data center due to the unreasonable distribution of the elephant flow load distribution.In this paper,this paper build the FatTree's forgotten routing model and study single-stream routing optimization algorithm,then this paper propose elephant flow collision avoidance routing optimization based on the non-equal probability.According to the model,the traffic is easy to collide on the core link.Fattree is CLOS structrue,so this paper make the non-blocking path in FatTree as the primary path of the elephant stream and select the alternate path with a low probability to balance the traffic by using the DCN multipathing feature.The algorithm does not need the centralized collection of network data.It's more suitable for distributed structure.Finally,it is proved that the algorithm can reduce the probability of collision of elephant flow and reduce the maximum link load by about 10% with respect to ORM.