Area Fragmentation in Reconfigurable Operating Systems.

Abstract

Area utilization is poor in reconfigurable operating systems due to fragmentation of placement resources caused by dynamic addition and deletion of tasks. We attempt to quantify the amount of fragmentation of the area resources. Main assertion of our fragmentation model is that each empty cell on the FPGA offers different fragmentation depending upon other empty cells in its vicinity. Our model can be used to identify regions of high fragmentation on the FPGA. We introduce a novel and run-time efficient algorithm to measure fragmentation of a given area on the FPGA. We use our fragmentation model to obtain high quality placements in a real-time online placement engine. We show by experiments that our model produces better quality results than both first-fit and best-fit placement strategies. Lesser fragmentation leads to better area utilization, lesser task rejection ratio and faster execution of the tasks.