Modeling and control of server systems: Application to database systems

Servers technology is a mean to support different Internet services and applications ranging from web servers, to email systems, streaming media services, enterprise servers, and database servers. However, the ad-hoc configuration of servers poses significant challenges to the performance and economical costs of applications. This paper precisely addresses this issue. Firstly, we present the design of a server model as a non-linear continuous-time model. A fluid flow approach allows us to partly free ourselves from accurate stochastic considerations. Secondly, a utility function for characterizing the impact of server configuration on performance and cost is presented. Thirdly, a utility-aware capacity planning algorithm is developed to efficiently control the configuration of servers. Model and control algorithm were implemented and applied to the standard PostgreSQL database server running the TPC-C warehouse application. The experiments show that the proposed method provides significant benefits for database servers management.