中央研究院 資訊科學研究所

In computing systems and applications, people are always looking for better performance of various types: timeliness, reliability, energy efficiency, to name a few. Often, it seems that there is a trade-off between meeting these performance metrics. For example, a system engineered for fault tolerance may run slow, and a system that computes fast may consumes high energy. But is this always the case? If not, is there any design principle that we may follow, to build a system that meets seemingly conflicting performance demands? This talk will give a case from real-time fault-tolerant data communication, showing that one may meet both real-time performance and fault-tolerance performance, by having a unified problem analysis from the timing perspective.