Maximal strips data structure to represent free space on partially reconfigurable FPGAs

Abstract

Partially reconfigurable devices allow the execution of multiple tasks simultaneously on the same chip. To schedule a set of tasks in a small amount of time, the scheduling algorithm will need to represent the free space efficiently. A data structure to represent free space should allow the scheduler to identify free space in which to place a new task and admit efficient updates after placing or removing a task. In this paper, we review some existing data structures and analyse their time complexity. We propose a new structure using maximal horizontal and vertical strips (MHVS) to represent the free space. These strips start and stop at task boundaries. Simulation and time analysis showed that this method has better time complexity than many other free space data structures and at the same time has a very reasonable rejection ratio on real-time tasks compared to other methods.

Keywords:

• online scheduling
• real time
• reconfigurable architecture
• task placement
• free space data structure

Acknowledgements

This work was supported in part by the National Science Foundation under grant CCR-0310916.