Parallel Computing

Using parallelization protocols/standards for acceleration algorithm's performance

Table of contents

• General Info
• Features
• Prerequisites
• Setup
• How to run
• Acknowledgements

General Information

OpenMP Application Protocol Interface (API) and subset of functions of POSIX threadsstandard (pthreads) to speed up the Smith-Waterman algorithm for local alignment of sequences. A simplified form of omega statistic, to detect positive selection in DNA sequences. Exports performance statistics. Applied for N random data.

Features

• Serial program on SW algorithm ¹²
• Parallel Standards (OpenMP²³, Pthreads⁴⁵⁶)

Benchmarked on Intel(R) Core(TM) i7-1065G7 @ 1.30GHz 1.50 GHz with 8GB DDR3 memory.

Prerequisites

• Input .txt files for test. D_SIZE number of pairs of sequences of characters, with each sequence being on a separate line or extending to more lines for ease of reading. Read input file and reserve D,Q variables with variables from command line. For more info, see How to run.

dataset.txt

2

Q:      abc
D:      xxxabxcxxxaabbcc

Q:      aaabcd
D:      abababcabababcd

• Understanding of the SW algorithm

How to run

1. GCC installation

$ gcc --version
$ sudo apt install gcc

Reference

1. Compile .c file

gcc -o newserial newserial.c

2. Run in command-line flags and arguments on linux terminal

./newserial -name ID -input PATH -match INT1 -mismatch INT2 -gap INT3

where

• ID => string for .out file
• PATH => .txt path
• INT => int variable

OpenMP

1. OpenMP config

$ echo |cpp -fopenmp -dM |grep -i open
$ sudo apt install libomp-dev

3. Setting the number of threads⁷

$ export OMP_NUM_THREADS=8 

4. Run in command-line flags and arguments on linux terminal

gcc -fopenmp -o OMPX <OMPa.c>
./OMPX -name ID -input PATH -match INT1 -mismatch INT2 -gap INT3 -threads INT4

where
THREADS => num of threads , and X is the preference on OMP implementation (3 implementation of OMP based on task granularity ⁸ the different computation-to-communication ratio)

• OMPa : Fine grained
• OMPb : Fine grained
• OMPd : Course grained

Pthreads

1. Run in command-line flags and arguments on linux terminal

gcc -pthread  POSIXX.c -o POSIXX.
./POSIXX -name ID -input PATH -match INT1 -mismatch INT2 -gap INT3 -threads INT4

where
X is the preference on POSIX implementation (2 implementation of POSIX based on task granularity ⁹ the different computation-to-communication ratio)

• POSIXa : Fine
• POSIXc : Course

See scripting.sh ⁸ for more..

Setup

Script variables initialized as:

• N = 10000000.
• Threads = [2 4]
• Processors = [2 4].

Acknowledgements

• This project was implemented for the requirements of the lesson Architecture of Parallel and Distributed Computers

Footnotes

1. https://en.wikipedia.org/wiki/Smith-Waterman_algorithm#Linear ↩

2. https://computing.llnl.gov/tutorials/openMP/ ↩ ↩²

3. http://www.openmp.org ↩

4. https://computing.llnl.gov/tutorials/pthreads/ ↩

5. http://www.cs.cmu.edu/afs/cs/academic/class/15492-f07/www/pthreads.html ↩

6. https://www.ibm.com/developerworks/library/l-posix1/ ↩

7. Default number of threads (ignoring this command) is defined from the specification of CPU ↩

8. Run scripting.sh to compile and run all files. Must be executable. ↩ ↩²

9. https://en.wikipedia.org/wiki/Granularity_(parallel_computing)#:~:text=In%20parallel%20computing%2C%20granularity%20%28or%20grain%20size%29%20of,communication%20overhead%20between%20multiple%20processors%20or%20processing%20elements. ↩