High Performance Computing on an IBM Cell Processor

Download Report

Transcript High Performance Computing on an IBM Cell Processor

High Performance Computing on an IBM Cell
Processor --- Bioinfomatics
May08-24
Kyle Byerly
Shannon McCormick
Matt Rohlf
Bryan Venteicher
1
Advisor: Dr. Zhao Zhang
Introduction
 Problem Statement
– Researchers need to tackle more complex and
computational demanding problems, but are faced with
limited budgets
 Proposed Solution
– Use the PlayStation 3 and Cell processor to achieve
improved performance of BioInfomatics applications at a
low cost
2
Senior Design May08-24
12/7/07
Bioinformatics
 What is it?
– Field involving the use of computers for the organization
and manipulation of data related to biological structures
such as DNA, RNA and proteins
– Takes advantage of mathematics and science to solve
computational problems involving molecular structures
3
Senior Design May08-24
12/7/07
Bioinformatics (cont)
 Example bioinformatics application: Clustal
– A multiple sequence alignment tool
– Takes multiple sequences of DNA, RNA, or proteins as
input
– Arranges them to illustrate similarities between all
sequences
– Can be used to identify structural, functional, or
evolutionary similarities
4
Senior Design May08-24
12/7/07
Literature Survey
– Previous work done by Sachdeva, Vipin; Kistler, Michael;
Speight, Evan; Tzeng, Tzy-Hwa Kathy. 2007. “Exploring
the Viability of the Cell Broadband Engine for
Bioinformatics Applications.” IBM.
– Positive results – up to100x speedup
5
Senior Design May08-24
12/7/07
Cell Processor
 Sony, Toshiba, and IBM (STI) Alliance
– Work began in 2000
 February 2005 – First technical disclosures
 May 2005 – First public demonstration
 “Super-computer on a chip”
 Multi-core processor
 Home entertainment to distributed computing
 Heterogeneous Processor
– Power Processor Element (PPE)
– Synergistic Processing Element (SPE)
– Element Interconnect Bus (EIB)
www.power.org/resources/devcorner/cellcorner/CellTraining_Track, L1T1H1-02 Cell Overview
6
Senior Design May08-24
12/7/07
Project Plan Summary
 Project is broken into several steps
– Basic understanding of bioinformatics
• Will study selected applications algorithm in depth
– Research, experimentation with Cell/B.E. programming
– Understanding of parallel algorithms
– Determine which applications to port
– Actual porting process
– Verify port correctness, document performance
improvements
7
Senior Design May08-24
12/7/07
Project Schedule
 Next Semester Plans
– Finalize application selection
– Port applications
– Testing
8
Senior Design May08-24
12/7/07
System Block Diagram
9
Senior Design May08-24
12/7/07
Operating Environment
 Linux operating system on the Sony PlayStation 3
 Dry
 Temperature controlled environment
10
Senior Design May08-24
12/7/07
Deliverables
 Source code to two ported BioPerf applications
 Publishable paper if results warrant and time
allows
11
Senior Design May08-24
12/7/07
Requirements
 Functional
• Ported applications shall run on Cell/B.E.
• Ported applications shall produce the same output as unported applications
• Ported applications shall return their execution time
 Non-functional
• Ported applications shall run faster with the SPEs than
without
• User interface will not be altered
12
Senior Design May08-24
12/7/07
Work Breakdown Structure
13
Senior Design May08-24
12/7/07
Resource Requirements
 PlayStation 3
 Algorithms Books
14
Senior Design May08-24
12/7/07
Design Method
 Three methods to port applications to Cell/B.E.
– Depend entirely on the compiler
– Keep functional logic intact, but split application apart so
parts of it can run in multiple threads on the SPE’s
– Break application into multiple threads for SPE execution,
alter functional logic, vectorized code where possible
15
Senior Design May08-24
12/7/07
Specifications
 Input
– Two applications from the BioPerf suite
• Must not be applications already ported
• Limited data set sizes
• Should have parallelizable blocks/functions of code
 Output
• Execution time will be compared
• The GNU Profiler, GProf will be used to analyze modified
code
16
Senior Design May08-24
12/7/07
Specifications (cont)
 User Interface
– No plans to alter existing interface
• Sample
$ sh use-bioperf.sh
You can do the following:
[R] Run BioPerf
[I] Install BioPerf on your architecture
(if your architecture is not PowerPc, x86)
[C] Clean outputs in /home/seniord/BioPerf.profile/Outputs
[D] Display all versions of the installed codes
17
Senior Design May08-24
12/7/07
Specifications (cont)
 Hardware
– PlayStation 3
– Powerful servers to do comparisons with
• Quad-core Intel Xeon 3.0GHz, 6GB RAM
18
Senior Design May08-24
12/7/07
Specifications (cont)
 Software
– Fedora 7 with Linux 2.6.23.1 kernel
– IBM SDK 3.0
– IBM XLC 9.0, gcc 4.1.1
– GProf 2.17.5
 Testing
– Benchmark ported applications with 0, 1, 3, and 6 SPE threads
– Verify the output is correct
19
Senior Design May08-24
12/7/07
Conclusions & Questions
 Conclusions
– Project is interesting
– Project is feasible
– Porting process will be difficult
20
Senior Design May08-24
12/7/07