Abstract: Multi-tenancy is one of the key characteristics of cloud application. In shared data storage model, how to achieve dynamically scalable data storage of multiple nodes according to different tenants' performance requirements is the key of cloud data management. It faces many challenges, such as how to forecast performance requirements of different tenants under shared data storage, how to place and control data layout in response to bursty in workload, i.e. data items automatic partitioned, coalesced or copied among the nodes of the system according to variable tenants' performance requirements. In this paper, we propose a scalable and self-adjust multi-tenant data storage strategy to address these challenges. It mainly consists of piecewise multiple performance boundary model，which is used to predict whether the data node can meet the performance requirements of different tenants, and data storage layout adjustment strategy based on greedy, which is used to make decisions about data movement on overloaded node and data coalescence on underloaded node. Through the analysis of experiment system, this method can accurately predict whether the system is overloaded or not, reduce the influence on system performance by less data movement, and make the shared cloud data node satisfy performance requirements of different tenants.

Key words: multi-tenant, scalable, self-adjust, performance prediction, data storage