Hardware support for dynamic scheduling in multiprocessor Operating System

In the present paper, we propose a mechanism to support dynamic task scheduling on multiprocessor systems, which assists the scheduler in efficiently and adaptively managing the platform resources. The main objective of this method is to maximize system resource utilization, by allocating available platform resources to a task based on the individual task characteristics and performance requirements. The proposed mechanism consists of a high speed database which stores the information about the task characteristics with respect to certain system topologies (called as metadata) and a supporting infrastructure which provides the real time availability of the different system resources. The scheduler uses this information to quickly and efficiently map the tasks on the available resources, such that the performance requirements of each task are met. The compactness of the data and the low latency of the database overcome the limitations posed by the existing dynamic methods. The paper deals with the different design alternatives for the proposed mechanism. An analysis of the schedule efficiency for a video decoding application is given, which helps us to understand the importance of the proposed mechanism.