Design And Implementation Of Multipath Routing Mechanism In Data Center Net-Working

Data center is usually the center of data transmission,computation and storage.With the emerging of virtualization and cloud computing,Data Center Networking(DCN)is undergoing profound reforms from many aspects,such as structure,function,and scale,etc.It has become a hot research issue in both academia and industry.Tra-ditional DCN inherits the hierarchical structure,which results in that root nodes may become the bottlenecks of the whole network.To adapt the new requirements of DCN,researchers have proposed a variety of new network structures viewed from different angles.These new structures have multiple paths in nature,and most topologies of them are fixed and well-regulated.Given this,routing mechanisms can leverage mul-tipath to provide higher network bandwidth and better reliability guarantee.However,most existing multipath routing mechanisms cannot be optimized for general use,be-cause they are designed for specific network structures.Therefore,multipath routing mechanism in DCN offers the big room and value of research for this thesis.In this thesis,we study the multipath routing mechanism in DCN from two as-pects,i.e.,load-balancing and fault-tolerance.For load-balancing,two optimization objectives are designed.One is to minimize the average value of bandwidth utilization;the other is to minimize the variance of bandwidth utilization.This optimization problem belongs to NP problem.Thus,this thesis introduces the improved ant colony optimization algorithm(IACO),which is one kind of the intelligence optimization algorithms.IACO simulates the process of ants foraging and pathfinding.Bandwidth utilization and available bandwidth are used to calculate the selection probability when an ant chooses the next hop node.Then,this thesis conducts local update opera-tion of pheromone and global update operation of pheromone.After multiple itera-tions,three optimal paths are generated.Finally,data flows are segmented and solu-tions for packet reordering are proposed.In addition,traffic is assigned to the three paths according to certain rules in an even better fashion.For fault-tolerance,this the-sis first introduces the concept of path value which is used to represent the quality of each path.Path value considers both load-balancing and fault-tolerance and utilizes bandwidth information and other information to calculate path value.Then,this thesis designs fault-tolerant effect and cost function,which is used to calculate optimal path numbers.Fault-tolerant cost includes link congestion cost and elephant flow replica-tion cost.Finally,heuristic thoughts are adopted to design the algorithm of generating feasible path set.According to the heuristic algorithm,the final path set can be got.Correspondingly,data flows can be replicated and further transmitted to the destina-tion paths.The above mechanism improves fault-tolerance while considering load-balancing across the whole network.This thesis simulates the proposed multipath routing mechanism and compares its performance with equal cost multipath routing,stochastic load-balancing multipath routing,centralized multipath fault-tolerant routing and no fault-tolerant routing.Re-sults show that the proposed mechanism is superior in load-balancing and fault-tolerance to algorithms compared.Therefore,the mechanism proposed in this thesis is feasible and effective.