Efficient runtime performance monitoring of FPGA-based applications

Embedded computing platforms have long incorporated non-traditional architectures (e.g., FPGAs, ASICs) to combat the diminishing returns of Moore's Law as applied to traditional processors. These specialized architectures can offer higher performance potential in a smaller space, higher power efficiency, and competitive costs. A price is paid, however, in development difficulty in determining functional correctness and understanding the performance of such a system. In this paper we focus on improving the task of performance debugging streaming applications deployed on FPGAs. We describe our runtime performance monitoring infrastructure, its capabilities and overheads on several different configurations of the monitor. We then employ the monitoring system to study the performance effects of provisioning resources for Mercury BLASTN, an implementation of the BLASTN sequence comparison application on an FPGA-accelerated system.