A Model for Programming Data-Intensive Applications on FPGAs: A Genomics Case Study

Genomics computing is indispensable in basic medical research as well as in practical applications such as disease prevention, pharmaceutical development, and criminal forensics. DNA sequencing, assembly and analysis are key components of genomics computing. Coupled with the increased use of computation for both synthesis and analysis of data in genomics is the astounding increase in the rate at which next-generation sequencing platforms are producing genomic data. Keeping up with the combination of increasing levels of algorithmic demands and an exponential increase in data represents a huge computational challenge that requires a corresponding revolution in how we process the data. Field Programmable Gate Arrays (FPGAs) are particularly well suited to the type of highly parallel, bit-level computations found in genomics algorithms. Unfortunately, the use of FPGA platforms among genomics researchers has been limited by the specialized hardware design expertise currently required to use these platforms. Another limiting factor has been a proliferation of FPGA platform architectures, each generally requiring a re-implementation of the algorithm. This paper describes a new programming model called Elan and an associated compiler that we are developing for FPGA-based genomic applications. The Elan model and compiler allow a programmer to use familiar concepts from parallel and distributed computing to develop an application at a relatively high level of abstraction, which can then be compiled automatically to large-scale FPGA platforms. One of the goals of Elan is to allow an application to be run seamlessly across both the CPU and FPGA portions of a platform, and to be parallelized easily across a system comprising many FPGAs and CPUs. We use the short read alignment application as the motivating example.