Dynamic Scheduling, Allocation, and Compaction Scheme for Real-Time Tasks on FPGAs

Abstract

Run-time reconfiguration (RTR) is a method of computing on reconfigurable logic, typically FPGAs, changing hardware configurations from phase to phase of a computation at run-time. Recent research has expanded from a focus on a single application at a time to encompass a view of the reconfigurable logic as a resource shared among multiple applications or users. In real-time system design, task deadlines play an important role. Real-time multi-tasking systems not only need to support sharing of the resources in space, but also need to guarantee execution of the tasks. At the operating system level, sharing logic gates, wires, and I/O pins among multiple tasks needs to be managed. From the high level standpoint, access to the resources needs to be scheduled according to task deadlines. This thesis describes a task allocator for scheduling, placing, and compacting tasks on a shared FPGA under real-time constraints. Our consideration of task deadlines is novel in the setting of handling multiple simultaneous tasks in RTR. Software simulations have been conducted to evaluate the performance of the proposed scheme. The results indicate significant improvement by decreasing the number of tasks rejected.