Abstract: Heterogeneous computing has emerged as a powerful tool for modern highly-connected systems to efficiently distribute computing resources. Yet, many systems with real-time constraints may require expensive over-provisioning to meet their requirements. To address this problem, a real-time system may be designed to be overloaded on its primary processor, but be made viable through the support of a powerful secondary processor. In such a system, workload is scheduled on the primary processor by default, but some jobs that would cause a deadline miss can be offloaded, through a transfer of the necessary data, to the secondary processor. We propose an earliest-deadline-first (EDF) based scheduling framework to demonstrate schedulability conditions of such a system, including the costs incurred by offloading. This represents a first stage towards a complete understanding of this complex class of systems and their dynamics.
0 Replies
Loading