Error-driven QoS management in imprecise real-time databases

In applications such as Web-applications, e-commerce (electronic commerce), and engine control, the demand for real-time data services has increased. In these applications, requests have to be processed within their deadlines using fresh data. Since the workload of these systems cannot be precisely predicted, they can become overloaded and as a result, deadline and freshness violations may occur. To address this problem we propose a QoS-sensitive (quality of service) approach based on imprecise computation, applied on transactions and data objects. We propose two algorithms FCS-HEF (feedback control scheduling) (highest error first) and FCS-HEDF (feedback control scheduling) (highest error density first) that give a robust and controlled behavior of RTDB (real-time database) in terms of transaction and data preciseness, even for transient overloads and with inaccurate run-time estimates of the transactions. Further, performance experiments show that the propose algorithms out perform a set of baseline algorithms including FCS-EDF (feedback control scheduling) (earliest deadline first), which schedules the transactions using EDF (earliest deadline first).