Utilization of execution histories in scheduling real-time database transactions

摘要

Real-time database systems support data processing needs of real-time systems where transactions have time constraints. Here we consider repetitively executed transactions, and assume that execution histories are logged. A well-known priority assignment technique called earliest-deadline-first is biased towards short transactions in which short transactions have better chances of completing their executions within their deadlines. We introduce the notion of “fair scheduling” in which the goal is to have “similar” completion ratios for all transaction classes (short to long in sizes). We propose priority assignment techniques that overcome the biased scheduling and show that they work via extensive simulation experiments.