On- and off-line model identification for power management of Web service systems

In the context of Web services, hosting centers need to comply with the service level agreements (SLAs) stipulated with their customers while minimizing the operating costs, mainly related with the energy expenses due to servers operations. The problem can be effectively formalized using system identification and control theory: the SLAs are translated into set-points for the response time of the servers and tracking performance are traded-off with energy saving objectives on the basis of suitable models for server dynamics. Two approaches are analysed for modeling such systems: recursive and LPV state space system identification methods. The suitability of both identification methods is investigated and their performance assessed on experimental data measured on a custom implementation of a workload generator and a microbench-marking Web service application.