Service Differentiation in Real-Time Main Memory DatabasesReport
The demand for real-time database services has been increasing recently. Examples include sensor data fusion, decision support applications, web information services, e-commerce, 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 and stringent timing/data temporal consistency constraints, real-time databases can be overloaded. As a result, users may suffer poor services. Many transaction deadlines can be missed or transactions may have to use stale data. To address these problems, we present a service differentiation architecture. Transactions are classified into several service classes based on their importance. Under 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. Our approach also achieves a relatively low miss ratio for the less privileged service classes, thereby reducing potential starvation.
All rights reserved (no additional license for public reuse)
Kang, K, Sang Son, and John Stankovic. "Service Differentiation in Real-Time Main Memory Databases." University of Virginia Dept. of Computer Science Tech Report (2002).
University of Virginia, Department of Computer Science