Research On Key Technologies Of Distributed Storage In Cloud Computing

The wave of development in a new generation of IT technology comes.As a core part of this development,cloud computing and its technologies related have affected our lifestyles imperceptibly.Distributed storage,which is the key technology for building infrastructure in cloud computing,faces new challenges.How to build a high-performance,high-reliable distributed storage system in a ultra-scale,mass data,high-available,highscalable environment,is a hot research area in both industry and academia.Based on the knowledge of cloud computing and the technologies in distributed storage,this dissertation has made a thorough analysis and study on research works currently,and discussed related technologies in distributed storage of cloud computing.The contributions of this dissertation are as below:1.This dissertation proposes a novel chunk coalescing algorithm of data deduplication.Existing works focus on the granularity of chunking,the calculation of chunk's fingerprint,the optimization of indexing and updating.These works do not consider the extra overhead during the process of chunk coalescing.The proposed algorithm takes into account both the minimum and maximum number of sub-chunks during coalescing.It adjusts the number of sub-chunks dynamically.As a result,the algorithm reduces the cost during sub-chunks coalescing and the time complexity of algorithm.Besides,it optimizes the utilization efficiency of storage space.2.This dissertation designs a Diamond-based message spreading strategy.Existing works either focus on the efficiency of message spreading,or when some part of nodes are down,other alive nodes can still receive messages.The proposed message spreading strategy can obtain good tradeoff between the two factors,and guarantee that messages are able to be spread to all nodes in storage system quickly.In addition,it makes sure that the remaining nodes are able to receive messages when parts of nodes are down.3.This dissertation proposes an immune optimization strategy based replica placement algorithm.Existing works only focus on the utilization of network bandwidth,the distance between nodes and clients,the cost of replica access and so on.The proposed algorithm introduces the core idea of immune system.With the process of clonal selection and immune memory,it calculates the affinity of nodes in distributed storage system,and picks up proper nodes for placing data replicas.The simulation results show that the convergence of proposed algorithm is fast,and it can reduce the response time when clients access data replicas.4.This dissertation proposes an agent-based consistency algorithm-A*Raft.Raft is a consistency algorithm,but it still has some problems.For example,the leader node is in a risk of single-point of failure,and the efficiency of log consistency is low.A*Raft defines the costs of log consistency on nodes,and selects proper agent nodes for helping and balancing the load of leader node.It can avoid the situations of single-point of failure in part,and reduce time complexity of consistency algorithm.In conclusion,this dissertation has made a systematic research work on distributed storage technologies of cloud computing,and done beneficial explorations for building distributed storage system with high-performance and high-reliability.