Key Techniques Of Multi-Tenant Database Design

As the typical model of software as a service, multi-tenant database consolidates the data of di?erent tenants, reduces the service price for each tenant and obtains the large profit through the long-tail e?ect for the multi-tenant service provider. Therefore, a great deal of research work has been done based on the multi-tenant database system in both academic and industrial communities. However, as the scale of the tenants becomes larger, the service requirement improves and the appearance of the new storage, multitenant database system is faced with big challenges. First, current system design solutions cannot consider the performance, space requirement and system scalability at the same time. Second, current multi-tenant database system does not support full customization of di?erent tenants. Third, existing multi-tenant database does not provide enough support for the new storage, e.g., phase change memory. Based on these challenges, this paper proposes lots of optimization solutions for the multi-tenant database system. The research work and contribution of our paper can be summarized as follows:1. Adaptive schema design for the multi-tenant database with limited customization: Current multi-tenant database system cannot consider the performance, space requirement and scalability at the same time. Our paper proposed an adaptive schema design framework which considers the fact the multi-tenant service usually has high relationship and we build the table from the attribute level in order to ensure the scalability of the system. The method can adapt to the various configurations of di?erent tenants and the changes of the workloads dynamically so as to achieve high performance and low space under di?erent applications.2. Full Customized schema design for the multi-tenant database system : Current multi-tenant database provides limit support to the tenants, our paper devised a full customized schema design framework. In this framework we propose the recommendation techniques to help tenants better design customized schemas. In order to ensure the system scalability under the full customization, we design the two phase method to integrate customized schemas. The first stage is cluster-based schema integration and the second is the interactive-based schema integration. Besides, we propose the graph partition method to optimize the integrated schemas from both the space and query aspects at the same time.Therefore, our framework supports high scalability, good performance and low space with full customization property.3. Database index design based on the phase change memory: There is little work to support phase change memory especially for the database index field. Our paper proposed a Bp-Tree database index which is designed for PCM based on the predictive model. The Bp-Tree index tries to minimize the number of index node writes by preallocating the node space based on the predictive model because the writing operation in PCM has high latency and consumes lots of energy compared with the PCM reading operations. We relax the strict order constraint of the index leaf nodes. In addition, we design the index status parameters to instantly monitor the index condition so as to ensure the accuracy of the predictive model.