Scheduling real-time transactions with dynamic values: a performance evaluation

Abstract

A number of researches on real-time databases have focused on systems where the semantics of transaction time constraints include only deadlines or values that are constant during a transaction's life time. For some real-time applications, the values of transactions may change dynamically with time. Under such circumstances, existing real-time scheduling algorithms may not be applicable. A new real-time scheduling algorithm that can efficiently schedule transactions with time-variant values is proposed. By exploiting the estimated execution time of transactions to estimate the expected values of transactions at completion time, the proposed algorithm can generate a better priority ordering of transactions than the existing algorithms. Moreover, a concurrency control protocol based on conditional-restart is also proposed to reduce the costly restarts. Through performance evaluation, the proposed algorithm is superior to the existing algorithms under various system environments. Some suggestions for scheduling real-time transactions with time-variant values are also presented.