Reliability comparison of schedulability test in ubiquitous computing

The development of ubiquitous intelligent has increased the real-time requirements for computing system. If one real-time computation does not complete before its deadline, it is as worse as that the computation is never executed at all. Ineffective computation not only wastes computational resources, but also might bring system overload and collapse. Hence, a schedulability test is necessary to ensure the stability of ubiquitous system. The schedulability test is concerned with determining whether a set of tasks is schedulable on a cluster. Although a number of schedulability tests have been developed, they can not be compared due to distinct test principles. In this paper, we propose a reliability indicator, through which the probability that a random task set succeeds in schedulability test can be evaluated. The larger the probability is, the better the test is. The reliability of two sufficient deadline monotonic tests are compared, and the comparison result is further validated by detailed experiments. Both analysis and experimental results show that the performance discrepancy of schedulability test is determined by a prerequisite pattern. Since this pattern can be deduce by reliability indicator, it may help system designers choose a good schedulability test in advance.