Reactivity-based Scheduling Approaches For Internet Services

Understanding the characteristics of Internet services workloads is a crucial step to improve the quality of service offered to Web users. One aspect that is usually neglected during a performance evaluation is the user reactivity, i.e., how the users react to variable server response time. This paper addresses the use of reactivity to improve the quality of service (QoS) of Internet services. We propose and evaluate two new scheduling approaches: the patient-first impatient-next (PFIN), and the impatient-first patient-next (IFPN). We design and implement the USAR-QoS simulator that allows the evaluation of QoS strategies considering the dynamic interaction between client and server sides. We simulate the new strategies using a TPC-W-based workload. The experiments show the benefits of the reactive policies which can result in better QoS for Internet services, improving the user satisfaction