Research On Fault Tolerance Diagnosis Of Data Center Network

At present,the number of data centers in the world has reached 8 million.The traditional switch-centric data center network is mostly based on a tree-shaped three-layer structure.This structure is difficult to meet the requirements of large-scale data centers.A large number of new data center network structures have been proposed.Among them,the BCube network is a server that uses recursive definition and can support millions of orders.It has excellent properties such as small diameter,large capacity,and high fault tolerance.The data center network may contain hundreds or even thousands of servers(nodes),but some of these servers may often fail,which may cause some key nodes to lose connectivity,and even the entire network is paralyzed.This has prompted efforts to solve the challenging problem of fault-tolerant reliability of data center networks.The process of distinguishing between faulty nodes and non-faulty nodes in the network system is called fault diagnosis.To the extent that the network can tolerate a fault,if a fault node is found,replace it with a fault-free node as soon as possible to maintain the reliability of the network.Therefore,the diagnosability of the network directly determines the fault tolerance of the networkTo diagnose the BCube structure,we use the classic diagnostic system model PMC for a multiprocessor system(usually represented by an interconnected network).Since the data center network is developed from the interconnected network,the PMC model can still accurately represent the degree of diagnosis of the network.It can also provide a very powerful diagnostic basis for fault tolerance in data center networks to better assess the reliability of the network structure.Therefore,diagnosis based on PMC model is of great significance for the development of data center networksThe main work of this project is to find out how many server failures a network structure can tolerate at most,and use it as a standard to measure the fault-tolerant diagnostic capability of the network structure.Firstly,according to the properties of the BCube network,the properties of the neighbor nodes of the BCube network are summarized and proven.It is proved that|NBk,n({u,v,x,y})|?4k(n-1)-4 for k>0,n?2.Based on this conclusion,t-diagnostic research is performed on the BCube network.As the most basic method of fault-tolerant diagnostic measurement of the network,t-diagnosability is(n-1)(k+1)-1(for n?2,k>0).Then based on t-diagnosis,we tried to use a more efficient fault-tolerant diagnostic method,t/t-diagnosis,to study and prove the fault-tolerant diagnostics of the BCube network.We found that this diagnostic method sacrificed some accuracy.Under the conditions,a higher diagnostic efficiency is obtained.After research,the t/t-diagnosis degree of the BCube network is((2k+1)(n-1)-1),(k?3,n?2).Finally,rules are added to the network itself,and the conditional diagnosability of the BCube network is studied.The conditional diagnosability of the BCube network is greater than[((n-1)(5k+1)-4)/2](n?2,k>0).In addition,this thesis proposes relevant diagnostic algorithms and conducts simulation experiments to further confirm the accuracy of the study.This thesis will use the system-level fault diagnosis model to study the fault tolerance of the BCube network.Through the study of a variety of diagnosable properties in the BCube data center network,it proves that it has a good diagnostic ability.Make sure that the BCube network has good fault tolerance diagnostics.