Research On A Tunable Consistency Strategy Of Distributed Database

As the internet grows,the data inside it become massive,in order to support heavier hits and data,using distributed databases has become an inevitable choice.But distributed databases will face the problems of higher latency of the strong consistency.As a result,a large number of databases use eventual consistency in systems.For example,the database used by Alibaba-DRDS,as a distributed database with eventual consistency shows a good performance when there are tens of millions of hits.At present,the Cassandra database allows users choose different consistency strategy at operation level.As one of the freshest technologies nowadays,the development of Blockchain is attracting more and more concerns.Although using Blockchain as the database still has a lot of problems,the decentralization of Blockchain can increase security and certainty of data.Hence,Blockchain is introduced in this paper,which can improve the reliability of data.This paper designs a strategy of distributed databases with tunable consistency through analyses and researches on distributed databases and consistency strategies.Firstly,adaptive replication factors algorithm is applied into the calculation of consistency level of each request in this paper.The network latency and accesses arrival distributions are computed.Thus,steal read rate of each read request is obtained.An estimation of the stale read rate in the system is computed and then compared to the application stale read that can be tolerated in order to provide the number of replications that the request should read in the cluster.And that number is the adequate consistency level for the running application.The request scheduling layer is introduced in the design of the tunable consistency strategy.It consists of the consistency level processing module,the request distribution module,the priority request processing queue,the request scheduling module and the request processing module.Cassandra cluster and BigchainDB are built as data access layer in this paper.YCSB is used as a test tool and test data sets.The test is completed under the Ubuntu 16.10 operation system.The experimental results show that the tunable consistency strategy proposed in this paper can control the order of processing requests and maintain consistency level of each request based on the tolerance of steal read rate of the application.In the context of massive data,the research of this paper is meaningful to consistency level control in operational level of a distributed database.Besides,this paper is also having reference to Blockchain technology used as a database.To summarize,the research work in this paper has some significance and practical value.