04SAAHPCpresentationScalableSystolicArray

04saahpcpresentationscalablesystolicarray

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Reconfigurable Supercomputing with Scalable Systolic Arrays and InStream Control for Wavefront Genomics Processing SAAHPC'10 C. Pascoe (speaker), 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 University of Florida M. Herbordt Tuesday, July 13, 2010 Department of Electrical and Computer Engineering, Boston University Motivation n Impending roadblock in health-sciences research q q Current-generation sequencing technology capable of producing millions of sequences per day Next-generation sequencing instruments poised to increase production more than 10-fold n Conventional sequence-analysis processing has not kept pace with raw sequence production Objective: Develop sustainable solution for increasing performance of sequence-analysis applications Approach: Develop in-stream control approach for systolic array to maximize hardware acceleration Capitalize on reconfigurable computing at scale to achieve sustainable supercomputing performance 2 Presentation Outline q Background n n n n Sequential Score Calculation Backtrack to Produce String Representation Sequence-Alignment Acceleration Common Hardware Acceleration Techniques q q Scalable Systolic Arrays with In-Stream Control Novo-G System Architecture n GiDEL PROCStar-III Boards q q q NW/SW/ND Performance Evaluation on Novo-G Conclusions Q&A 3 Background: SequenceAlignment Example Sequence S1: ACGGCTAACTG Sequence S2: GGCCCAACTGCGGGT n n Want to quantify similarity between S1 and S2 Use NW or SW or other algorithms to align sequences S (i - 1, j - 1) + Sub (i , S (i - 1, j ) + eog j ) = max H (i - 1, j ) + e S (i , j - 1) + eog V (i , j - 1) + e for 1 i imax , 1 j j max j ) S (i , S (i -1, j ) + og H ( i , j ) = max H (i -1, j ) + eg S (i , j -1) + og V ( i , j ) = max V (i , j -1) + eg S (0, i ) = S (i,0) = og +i eg H (0, i ) =V (i,0) = og Sub(i, j ) = penalty matrix Needleman-Wunsch Equations 4 Sequential Score Calculation S1 S2 G G C C C A A C T G C G G G T 0 -10 -10.5 -11 -11.5 -12 -12.5 -13 -13.5 -14 -14.5 -15 -15.5 -16 -...
View Full Document

Ask a homework question - tutors are online