A Supervised Approach-based Job Scheduling Technique for Distributed Real-Time Systems

Distributed real time systems have end-to-end jobs which are scheduled on multiple processors. These jobs are composed of several sub-jobs which do not have individual end-to-end constraints. To efficiently schedule these sub-jobs, their local deadline requirements are needed to be known. The local deadline assignment problem has been recognized as a crucial problem in distributed real-time system research. In this paper, we present a supervised machine learning based job scheduling technique for a distributed Real-Time System (RTS). We use linear regression, support vector machine, and artificial neural network machine learning techniques for predicting the local deadline of upcoming workload with a given release time and deadline of executed sub-jobs. We also develop a technique for labeled dataset creation in a distributed RTS. We demonstrate that the supervised machine learning based job scheduling technique reduces the job dropping rate and thereby enhances the utility of the distributed RTS.