Using real-time scheduling principles in web service clusters to achieve predictability of service execution

Real-time scheduling algorithms enable applications to achieve predictability in request execution. This paper proposes several request dispatching algorithms based on real-time scheduling principles that enable clusters hosting web services to achieve predictability in service execution. Dispatching decisions are based on request properties (such as deadline, task size and laxity) and they are scheduled to achieve designated deadlines. All algorithms follow three important steps to achieve a high level of predictability. Firstly, requests are scheduled based on their hard deadlines. Secondly, requests are selected for execution based on their laxity. Thirdly, the underlying software infrastructure provides means of achieving predictability with high precision operations. The algorithms use various techniques to increase the number of deadlines met. One decreases the variance of task sizes at each executor while another increases the variance of laxity at an executor. The algorithms are implemented in a real-life cluster using real-time enabled Apache Synapse as the dispatcher and services hosted in real-time aware Apache Axis2 instances. The algorithms are compared with common algorithms used in clusters such as Round-Robin and Class-based dispatching. The empirical results show the proposed algorithms outperform the others by meeting at least 95% of the deadlines compared to less than 10% by the others.