Research Of Fast Failure Recovery Algorithm In Data Center Network

The data center expands rapidly to cope with the rapid growth of Internet business and user data,and accordingly the scale of data center network(DCN)increases fast,which,nevertheless,brings frequent network failures and seriously damages network availability.Fast failure recovery algorithm provides high availability transmission guarantee for the DCN and therefore,it has become a research hotspot in academia and industry.The DCN consists of two part,inter data center network(Inter-DCN)and intra data center network(Intra-DCN).Inter-DCN builds the network interconnection between the data centers.The long transmission distance and the large number of intermediate routers are the network characteristics of InterDCN,for which the main challenges of fast failure recovery in Inter-DCN are low efficiency of routing deployment,high cost of flow retransmission,and the path redundancy problem.Intra-DCN builds the connection between the servers in DCN and has short transmission distance while the flows are burst and high concurrent.Consequently the rapid change of network state and fierce bandwidth competition are the main challenges of Intra-DCN fast recovery.This thesis analyzes the network characteristics of Inter-DCN and IntraDCN and the main challenges of fast failure recovery,and proposes targeted fast failure recovery algorithms,which focuses on the following main work:1.This thesis proposes a fast and low redundancy algorithm for Inter-DCN failure recovery(FLAIR).Aiming at the challenges of low efficiency of routing deployment and heavy retransmission,FLAIR reduces the updating of routing in failure recovery through an incremental rerouting algorithm and introduces Value at Risk theory to establish a traffic loss risk model for selecting flow schedule strategy of low re-transmission expectation.Aiming at reducing the redundancy,during network idle period,FLAIR efficiently computes and updates low redundant paths through a heuristic routing optimization algorithm based on a fast path state estimation method.The experiment results show that,compared with TEAVaR and other recovery algorithms,FLAIR averagely improves 70.2%recovery efficiency and simultaneously outperforms 1.14%on network availability.Meanwhile,FLAIR improves 36.0%transmission efficiency at most and shows good stability and scalability.2.For Intra-DCN,a fast awareness and competition avoidance failure recovery algorithm with stochastic remaining bandwidth balancing(FaCa),is proposed.Aiming at the challenges of rapid change of network state and achieving fast awareness of network state,FaCa introduces network probes technology and a bandwidth predictive algorithm,and accelerates the speed of reaction through the fast failure rerouting and the logically structural design.Aiming at alleviating the bandwidth competition,the random residual balancing algorithm is introduced to allocate flows,which ensures high transmission bandwidth.The experimental results show that FaCa achieves the efficiency of the state-of-the-art failure recovery algorithm PLB,and reduces 14.28%bandwidth competition at average,which strikes a strong balance between fast recovery and high transmission bandwidth.