A Worst-Case Behavior Analysis for Hard Real-Time Transactions

Most hard real-time systems that incorporate database components assume that the data is resident in main memory. For these systems, priority ceiling protocols can be used to limit priority inversion. In this paper we consider disk resident data, additionally we assume optimistic concurrency control with abort and restart. Hence, when a high priority transaction requires access to the disk it will cause any lower level transactions that are accessing the disk to be aborted. Analysis is presented that allows the worst case response times of transactions to be calculated. A feature of this analysis is that the I/O channel and the CPU are considered separately. Thus while a high priority transaction is using the channel, a lower priority transaction may gain access to the CPU.