Research On Consistency Of Multi-Tenant Database

The important thinking of SaaS (Software as a Service) and cloud computing is to take a variety of resources as services. Single-instance multi-tenant technology is an important technology by integrating resources for hundreds of different tenants. SaaS applications need to support multiple tenants and each tenant may have a very large number of users, so it is a great challenge for the performance, availability, scalability of the supported software infrastructure platform. SaaS and cloud computing themselves demand high availability and scalability of applications. Replication technology is the most important technology to improve availability, however, CAP theory limits that in a distributed system environment it can only fulfill two of consistency, availability and partition tolerance.There are three most commonly used methods to implement multi-tenant technology in data management, independent database, shared database independent schema and shared database shared schema. Based on the method of highest sharing degree, shared database shared schema model way, the paper builds up Multi-tenant Quorum NRW system which can provide high availability in full replications. It aims to combine each tenant’s application transaction characteristics and needs to assign each tenant suitable R and W values, and in the condition that can ensure the strong consistency, R+W>N, it tries to provide individual tenants and the whole system higher performance. The characteristics and requirements of the application transaction of the respective tenants may change, which will need the corresponding appropriate RW value changes. The paper summarizes as the two left change and right change mode according to the differences in the changes and proposes transition left change method and transition right change method. The change process remains online, but the system will maintain strong consistency.Meanwhile, the paper gives study on concurrency and consistency of single replica node. According to the characteristics of shared database shared schema, more suitable tenant oriented granularity lock is proposed, and from the average granularity size for various tenants speaking, tenant oriented granularity lock provides different tenants different lock granularity which is between the upper boundary granularity and the lower boundary granularity by definition. Aimed at the feature of tenant oriented granularity lock, we build a multi-tenant transaction model to analyze the performance of tenant oriented granularity lock on several aspects, including transaction conflict rate, deadlock rates, system throughput, the number of locks and basic table plus extension table storage. Granularity of each tenant may not be consistent with the transactional characteristics of each tenant’s application, and it may change. So in view of that granularity size of tenant oriented granularity lock is determined by the multi-tenant data distribution, we propose adjustment method for tenant oriented granularity lock, cluster and diffusion methods, and combine transaction conflict rate, deadlock rate, system throughput, the number of locks and basic table plus extension table storage to solve the approximate suitable granularity which can be used in granularity adjustment by tenants.Combining with thesis analysis, tenant oriented Quorum system and tenant oriented granularity lock are evaluated by experiments and from multiple perspectives of the experimental data, tenant oriented dynamic Quorum system and tenant oriented dynamic granularity lock can obviously improve the performance of the system.