A Massively Parallel Architecture for Bioinformatics - Department of ...

clumpfrustratedΒιοτεχνολογία

2 Οκτ 2013 (πριν από 3 χρόνια και 6 μήνες)

74 εμφανίσεις

A Massively Parallel Architecture
for Bioinformatics

Presented by

Md

Jamiul

Jahid

Introduction


Bioinformatics algorithms are demanding in
scientific computing


In general most of the bioinformatics
algorithms are fairly simple


Dealing with huge amount of data


The size of DNA sequence database doubles
every year

Introduction


A typical DNA contains 3.4 billion base pairs


Maximum algorithms use only simple
operations with input data like


Arithmetic operation


String matching


String comparison

Introduction


Standard CPUs are designed for providing a
good instruction mix for almost all commonly
used algorithm


For a target class of algorithm they are not
effective


Results


High runtime


Energy


Money

Contribution


Present a massively parallel architecture


Using low cost FPGA(Field Programmable Gate
Array)


They called it COPACOBANA 5000


Meaning
C
ost
-
O
ptimized
P
arallel
C
ode
B
raker

AN
d

A
nalyzer

COPACOBANA 1000


This machine is for cryptanalysis: fast code
breaking


120 low cost FPGAs


20 subunits


Each has
Xilinx Spartan


-
3 XC3S1000 FPGAs

COPACOBANA 1000


Assumptions


Programs are
parallelizable


Demand of data
transfer is low


All node needed
very little local
memory which
can be served
from on
-
chip
RAM of FPGAs


COPACOBANA 5000


Bus Concepts


Point to point connection two neighboring FPGA
-
cards


Point to point connection contain 8 pairs of wire


Each 250MHz, total 2Gbit/s


COPACOBANA 5000


Controller


Root entity of control is running on a remote host
computer


Connected to COPACOBANA5000 by LAN


Two scenario


Data on remote host


Data on COPACOBANA5000

COPACOBANA 5000


FPGA
-
Card


Xilinx Spartan
-
3 5000 is used


Contains 8 FPGAs


All FPGAs are globally clocked



Performance Estimation


Between


PC


COPACOBANA1000


COPACOBANA5000

Performance Estimation

Conclusion


In this paper a new hardware for running
bioinformatics algorithm is proposed


The hardware are


Cheap


Low power consumption


Efficient

Questions

Thank You

Reference


Gerd

Pfeiffer, Stefan
Baumgart
, Jan
Schröder
, and Manfred
Schimmler
,
A
Massively Parallel Architecture for Bioinformatics
, 9th International
Conference on Computational Science (ICCS 2009).