Improving computation efficiency of schedulability of transaction-based task model for distributed real-time embedded systems

Abstract

In the transaction-based task model, the system consists of multiple end-to-end transactions; each transaction consists of multiple tasks, with or without explicit precedence constraints; each task has a release offset (with possible release jitter) relative to the transaction start time. The transaction-based task model is especially useful for schedulability analysis in distributed real-time systems. For schedulability analysis of this task model, some authors have developed an exact analysis algorithm with exponential time complexity and a pessimistic tractable analysis algorithm with polynomial time complexity. In this paper, we aim to improve the computation efficiency of the exact analysis algorithm to make it applicable to larger tasksets. We define the concept of dominance relationship between any pair of tasks in the same transaction with higher-priority than the task under analysis. If task τix dominates task τiy in the same transaction, then τiy can be omitted from the WCRT analysis equations to improve computation efficiency. We present effective techniques for determining dominance relationship between tasks. Performance evaluation shows that our technique can generate significant improvements in computation efficiency.