Scalability Studies of the BLASTn Scan and Ungapped Extension Functions

Abstract

BLASTn is a ubiquitous and important tool used for large scale DNA analysis. As such, it is a good candidate for acceleration with FPGAs. The aim of this paper is two-fold. First, building upon our prior BLAST work we describe a design composed of multiple cores that can be scaled in two dimensions. The ungapped extension and a second dimension are new in this work. Second, we use this non-trivial example to explore spatially scalable designs. To provide the ability to move the design to a future generation chip, a mathematical model of performance that incorporates all of the system design parameters and the user’s preference (high throughput vs low latency) is developed. We demonstrate here that the model correctly predicts the optimal ratio between the two dimensions on a Xilinx Virtex-4 and measures four to five times faster performance figures as compared to a state of the art general purpose processor.