Effects of deadline propagation on scheduling nested transactions in distributed real-time database systems

摘要

In this paper, we address issues concerning efficiently scheduling real-time transactions in a distributed database system using a nested transaction model. In this model, a transaction may consist of several subtransactions, each of which acts as a unit of work. Therefore, a subtransaction can have its own priority to compete for resources. In real-time applications, a transaction is usually submitted with a deadline. A policy that assigns priority to transactions based on their deadlines has been widely used. In order to determine priorities of subtransactions within a nested transaction, deadlines of subtransactions must be known. This means that an efficient deadline propagation scheme must be used to propagate deadline of a transaction to its subtransactions. In this paper, we describe several deadline propagation schemes to derive CPU and data conflict resolution priorities which are based on either top-level transactions' deadline or individual subtransactions' work amounts within a transaction. We then conduct simulation experiments to measure the overall system performance when these schemes are being incorporated into a priority-abort conflict resolution protocol, where earliest deadline first is used for priority assignment. The simulation results indicate the following: 1) a deadline propagation scheme which assigns all members of a transaction with the same deadline works best to derive deadlines for data conflict resolution priority assignment as well as for CPU priority assignment if there are high data conflicts in the system, and 2) when data conflicts are low, individual subtransactions' deadlines based on subtransactions' own execution times should be used for CPU priority assignment.