Research On Multicast Routing Optimization And Reliability In Data Center Networks

As a key part of supporting for cloud computing, the performance of data centerdirectly determines the quality of cloud computing services. Data center network isthe communication bridge between the data centers. The load of data center networkis becoming heavier, which will affect the performance of data center. Multicast inthe data center attracts more and more attention. Multicast can effectively reduce theload of data center network and improve the performance of data center network.This paper mainly focuses on the optimization of multicast routing i.e., how toselect the recovery nodes and pathes for reliable multicast in data center networks.There are two types of data center network architecture, one is switch-orientednetwork structure,and the other is server-centric network structure. The servers inthe server-centric architecture also participates in forwarding packets. If the load onthe server is very large, it can affect the acceptance and forwarding of data packets,such as increasing the delay and packet loss rate. If the server need to forwardpackets when its load is large, it will undoubtedly further increase the burden onitself. When building a multicast tree considering the load on the server, it can makethe multicast tree more robust and balance the load between servers. Therefore, inthis paper, we design a new multicast tree building algorithm which is based on theload of server and distance between source and destination. The algorithm combinesthe ant colony algorithm and genetic algorithm. It exploits two parameters, theserver load and distance, to find the best solution. The algorithm tries to avoid theedges of the multicast tree passing the excessive load servers, which can effectivelyreduce the multicast latency and packets loss ratio.The traditional packets recovery method for reliable multicast is that the sourcenode sends packets to the destination via the original multicast tree route. As thedata center network has redundant links, the unicast approach is a feasible solotionfor packet recovery. Unicast can effectively reduce the burden of the original pathand increase the successful ratio of packet recovery. In this paper, we propose arecovery node selection algorithm which is based on the location of node that lossespackets and the priority of recovery node. We can estimate the packet lost ratiothrough the node position. The larger packet lost ratio the node has, the higherpriority the node has.Then the node and the recover node are established one-to-onerecovery relation. The algorithm can increase the recovery ability for the nodewhich is easy to loss packet and make each node has a average recovery pressure.Also, it needs less link resources. Finally, we verify the correctness and validity of the two algorithms throughthe simulation experiments.