ScalableSystolicArray

ScalableSystolicArray - Reconfigurable Supercomputing with...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon
1 Reconfigurable Supercomputing with Scalable Systolic Arrays and In-Stream Control for Wavefront Genomics Processing C. Pascoe, A. Lawande, H. Lam, A. George NSF Center for High-Performance Reconfigurable Computing (CHREC) University of Florida Y. Sun, W. Farmerie Interdisciplinary Center for Biotechnology Research (ICBR) University of Florida M. Herbordt Department of Electrical and Computer Engineering Boston University I. INTRODUCTION Computational challenges in genomics data mining and analysis are an impending roadblock in modern health-sciences research, where algorithms for DNA sequence processing grow alarmingly in computational needs as datasets from new instruments continue to dramatically expand. Reconfigurable computers featuring customizable processing devices (e.g. FPGAs) hold the key in addressing these challenges with high performance, productivity, and sustainability, where the architecture adapts to match the unique needs of each application instead of vice-versa. In this paper, we present the novel use of a method for incorporating control information into the data stream, limiting wasted cycles and increasing hardware utilization. This method is described in Section II, and then featured in Section III as device-level reconfigurable architectures for in-stream control with scalable systolic arrays to accelerate two leading genomics applications based on wavefront algorithms, Needleman-Wunsch (NW) and Smith-Waterman (SW) [1], and a third application, Needle-Distance (ND) [2], an augmentation of NW. These architectures are experimentally evaluated on Novo-G, the reconfigurable supercomputer in the NSF CHREC Center at Florida, where results from these genomics applications achieve unprecedented levels of sustained performance. Case study results are reported in Section IV, followed by Section V with summary and conclusions. II. SCALABLE SYSTOLIC ARRAYS WITH IN-STREAM CONTROL Control for conventional systolic-array datapaths usually consists of a separate centralized controller, many small distributed controllers, or combinations of the two. Depending upon the complexity of the underlying algorithm, these conventional control methods can add significant overhead in terms of both chip area (complex state machines, additional control lines, etc.) and execution time (non-computational control states, pipeline stalls, etc.). In the case of DNA sequence alignment, the simplest design of such conventional controllers (i.e. with minimum area overhead) can achieve hardware execution time equal to N × (setup time + pipeline latency + PE configuration time + time to process streamed sequence + time to record results), assuming N successive alignments need to be calculated. Given that everything except “time to process streamed sequence” is overhead, such a controller is immensely inefficient. By contrast, the best possible performance from a systolic-array architecture is one that overlaps all of the overhead with useful work, achieving hardware
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 2
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 03/27/2012 for the course EEL 4930 taught by Professor Staff during the Spring '08 term at University of Florida.

Page1 / 4

ScalableSystolicArray - Reconfigurable Supercomputing with...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online