Service differentiation in real-time main memory databases

The demand for real-time database services has been increasing. Examples include sensor data fusion, stock trading, decision support, Web information services, and data-intensive smart spaces. In these systems, it is essential to execute transactions in time using fresh (temporally consistent) data. Due to the high service demand, many transactions may miss their deadlines regardless of their importance. To address the problem, we present a service differentiation architecture for real-time databases. Transactions are classified into several service classes based on their importance. Under conditions of overload, different degrees of deadline miss ratio guarantees are provided among the service classes according to their importance. A certain data freshness guarantee is also provided for the data accessed by timely transactions which finish within their deadlines. Feedback control is applied to support the miss ratio and freshness guarantees. In a simulation study, our service differentiation approach shows a significant performance improvement compared to the baseline approaches. The specified miss ratio and freshness are supported even in the presence of unpredictable workloads and data access patterns.