Journal of Computer Aided Parallel Programming
Volume 2 Issue 1
Optimization of Sorting Algorithm Using OpenMP
C.D. Vaidya1, Ankit Manwatkar2, Ankit Pande3, Shritej Pochalwar4
Professor1
Department of Computer Science & Engineering
Rajiv Gandhi College of Engineering & Research, Nagpur, Maharashtra, India\
Corresponding Author: [email protected]
Abstract
In today’s world, multi-core processors have become popular due to performance, and their
ability to performed multiple tasks simultaneously. We are trying to explore OpenMP for
parallelization of sequential program in multicore architecture to translate sequential program
to parallel using Parallel Processing. So, we are working on parallelizing bubble sort to reduce
execution time. OpenMP pragmas are found to be useful in parallelizing the sorting operation to
reduce sorting time to a great extent. As we observed that the traditional bubble sort algorithm
cannot be implemented in a parallel processing because of the iteration dependency. So, there is
need of some different technique to run bubble sort in a parallel processing. Thus, we are trying
to parallelize traditional bubble sort without Iteration dependency.
Keywords: Serial Processing, Parallel processing, OpenMP, Sorting, Bubble Sort, Time
Complexity, Multi-Threading
INTRODUCTION
In computes, Parallel Processing is the
Serial processing is the attending to and
processing of programs by dividing them
processing one item at a time, while parallel
into multiple segments with a unique thread
memory processing is the act of attending to
and this is done to reduce the execution time
and processing all items simultaneously.
of program.
1
Page 1-9 © MANTECH PUBLICATIONS 2017. All Rights Reserved
Journal of Computer Aided Parallel Programming
Volume 2 Issue 1
We are focusing on the bubble sort
II. LITERATURE REVIEW
algorithm. Bubble sort is one of the oldest,
The
the simplest
sorting
programming interface that creates an
algorithm in use having a complexity level
environment in which programs could be
of O(n2). Bubble sort works by comparing
implemented in parallel processing. The
each item with its next item in the list and
bubble sort algorithm takes (n-1) iterations
swapping them if they are in erroneous
to sort „n‟ number of unsorted elements.
order.
and the slowest
Repeats this
process
until
OpenMP[1]
is
an
application
no
swapping is required for items. Such a
A sorting algorithm is one that puts
situation means that all the items are in the
elements of a list in a certain order. It makes
correct order. By this way, larger values
easy searching and locating the information.
move to the end of the list whereas smaller
values remain towards the beginning of the
The most-used orders are numerical order
list. It is also used in order to sort the array
and lexicographical order. An efficient
such like the smaller values comes after the
sorting algorithm is that takes less time and
larger values. In other words, all items are
space complexity. In this paper, I make
in the correct order.
contrastive analysis of bubble sort[2] and
merge sort and tried to show why required
In the sorting algorithm, if you observe the
some new approach to get best sorting
move of the elements with higher orders,
results.
they are like bubbles in the water, floating
slowly from the bottom to the top.
Based on the analysis of the traditional
bubble sort algorithm, this paper proposes
Bubble sort is simple to program, but it is
two bidirectional bubble sort algorithm
worse than selection sort for a jumbled
design ideas. After giving C language
array. It will require many more module
description of the three algorithms, the
exchanges, and is just good for a well-
paper tests the new algorithms, analyzes
ordered array. More importantly bubble sort
comparatively the time complexity [3] and
is usually the easiest one to write correctly.
space complexity of the three algorithms,
and summarizes such a conclusion that the
first new algorithm has the same average
2
Page 1-9 © MANTECH PUBLICATIONS 2017. All Rights Reserved
Journal of Computer Aided Parallel Programming
Volume 2 Issue 1
time complexity and space complexity as
thread assigns a specific number of threads
the original algorithm, while the second is
to the slave threads and a task is divided
time efficient than the original one, and thus
between them. Open MP is basically
truly reaches the optimization purpose.
designed
for
shared
memory
multiprocessors, using the SPMD model
III. METHODOLOGY
(Single Program, Multiple Data Stream). All
Parallel processing makes a single large task
the processors are able to directly access all
to be executed on multiple threads, for this
the shared memory in the machine, through
reason the divide-and-conquer approach is
a logically directed connection.
usually preferred over other techniques. The
main idea of the proposed method is
Programs will be executed on one or more
distributing the elements of the input
processors that may share some or all of the
datasets into many partitions according to
available memory. The program is typically
an analysis made on dataset elements and
executed by multiple independent threads
each partition is sorted with a bubble sort
that share data, but may also have some
algorithm using unique threads.
additional private memory zones.
Bubble Sort in Openmp
Open MP provides a straight forward inter
Open MP is a widely adopted shared
face to write software that can be used for
memory parallel programming interface
multiple core computers. Using Open MP,
providing high level programming. The
the programmer can write code that will be
range of Open MP applicability was
able to use all cores of a multicore
significantly extended recently by the
computer, and will be run faster if the
addition of explicit tasking features.
number of cores is increased. In this section,
we will execute our sequential bubble sort
The idea behind Open MP is that the user
as parallel by using an Open MP, in the
specifies the parallelization strategy for a
early stage of the program writing we
program at a high-level by providing the
should adapt our environment to be able to
program
the
understand Open MP statements. It is not
implementation of multithreading [9], a
worthwhile just to write #include header in
code.
Open
MP
is
parallel execution scheme where the master
3
Page 1-9 © MANTECH PUBLICATIONS 2017. All Rights Reserved
Journal of Computer Aided Parallel Programming
Volume 2 Issue 1
our program but also required to include
less execution time that the sorting in serial
some configurations to C environment.
processing, for doing this job we used
parallel processing concept for optimization
IV. PROPOSED METHOD
in terms of time. In serial processing bubble
Sorting of elements of large dataset takes
sort algorithm takes countably large amount
large amount of time when executed in
of time. So, to reduce the execution time
serial processing and whole execution done
one of the option is to run this traditional
using the single thread and single core,
bubble sort in parallel processing, but
because of this proper utilization of
theoretically it is not possible to directly
multicore architecture was not achieved. In
parallelized
this project, we proposed a technique by
algorithm with multiple threads because of
which proper utilization of system is done
iteration dependency is shown in fig1.
traditional
bubble
and the sorting operation is performed with
Fig.1 Example of traditional Bubble sort
4
Page 1-9 © MANTECH PUBLICATIONS 2017. All Rights Reserved
sort
Journal of Computer Aided Parallel Programming
Volume 2 Issue 1
Iteration dependency is a scenario in which
dataset has been made once, on each
execution of one iteration is depend on the
partition separately apply bubble sort and
result of previous iteration, unless execution
each partition get executed with unique
of one iteration is get complete later
thread is shown in fig 2.
iterations cannot be executed. So, we
proposed a method by which we can
The result of each thread is get combined in
execute a bubble sort algorithm in parallel
a single dataset. At this combining stage no
processing without iteration dependency.
additional sorting is needed because initially
This can be achieved by analyzing the
the partitions of a dataset are created
dataset on which sorting is to be done and
according to specific range of an elements.
according to this analysis partitions of a
Fig.2 Example of proposed Bubble sort
5
Page 1-9 © MANTECH PUBLICATIONS 2017. All Rights Reserved
Journal of Computer Aided Parallel Programming
Volume 2 Issue 1
V. RESULT
The given dataset is applied with proposed
In result, we get the variation in execution
method to get the desired outcome. Based
time of proposed parallel bubble sort as
on comparison of serial bubble sort against
compare
sort.
parallel bubble sort (i.e. Time taken by
Evaluation is done on the basis of different
serial bubble sort for execution compare
datasets as shown in following table.
with time taken by parallel bubble sort
to
traditional
bubble
running on 4 & 16 threads)
Table 1. Comparison of serial and parallel sorting
Sr. No.
Dataset Size
Serial time (in sec)
1 threads
Execution time of proposed parallel
bubble sort (in sec)
4 threads
16 threads
1
25000
2.179
1.423
1.253
2
65000
18.717
9.454
8.641
3
100000
44.571
22.534
20.731
4
150000
99.995
52.32
47.253
5
175000
137.354
72.088
68.536
6
200000
178.921
92.454
83.002
6
Page 1-9 © MANTECH PUBLICATIONS 2017. All Rights Reserved
Journal of Computer Aided Parallel Programming
Volume 2 Issue 1
Fig3. Graphical analysis
CONCLUSION
sort. The better performance is obvious
We have implemented the bubble sort
because parallel sorting algorithm take the
algorithm using multithreading (Open MP).
advantage of parallelism to reduce the
The proposed work tested on datasets with
average running time. In future, same
different size taken. The effect of the
analysis can be performed with parallel
number of cores on the performance of
sorting algorithms and parallel sorting by
bubble sort has been experimentally studied.
regular sampling architectures and the
processors with more than two cores.
It is observed that parallel sorting algorithm
i.e. parallel bubble sort performs well in all
respects in comparison to sequential bubble
7
Page 1-9 © MANTECH PUBLICATIONS 2017. All Rights Reserved
Journal of Computer Aided Parallel Programming
Volume 2 Issue 1
5) Shiying
REFERENCES
1) Wang Min, “Analysis on Bubble Sort
Wang,
Yunxia
Ren,
“Diagnosability of bubble-sort graph
Algorithm Optimization”, Institute of
networks
Electrical and Electronics Engineers,
diagnosis model”, IEEE, 823 - 826, Dec.
Volume 1, July 2010. 0 50 100 150 200
2015.
under
the
comparison
Execution Time (In sec) 0 50000
100000
150000
200000
Datasets
6) Monika
Gurbhele,
Chandu
Vaidya,
Graphical Analysis Serial Processing 1
“Fair share resource allocation for load
threads Proposed BS 4 threads Proposed
sharing” International Conference on
BS 16 threads
Innovations in information, Embedded
and Communication Systems (ICIIECS)
2) Sehrish
MunawarCheema,Nadeem
978-1-4673-8207-6 , 2016.
Sarwar, Fatima Yousaf, “Contrastive
Analysis of Bubble & Merge Sort
7) Prof. Chandu Vaidya, “An Approach for
Proposing Hybrid Approach”, IEEE,
Processor Utilization in Master Slave
371–375, Aug.2016.
Environment” International Conference
on Advanced Material Technologies
3) Jing-meiLi, JieZhang, “The performance
analysis
and
research
of
sorting
(ICAMT)December
2016.
2016.
27th
and
28th
Dadi Institute of
algorithm based on Open MP”, IEEE,
Engineering and Technology (DIET),
3281 – 3284, July 2011.
Visakhapatnam, Andhra Pradesh, India.
4) Monika Gurbhele ,Chandu Vaidya and
8) C. D. Vaidya and M. B. Chandak,
Hemant Turkar , “An approch for share
"Efficient Parallel Process Migration
resource allocation for load sharing"
Algorithm Using Statistical Approach,"
International
Computational
Intelligence
Communication
Networks
Research
Journal
of
Innovative
and
in
Computer
and
Communication
Engineering
DOI:
2012 Fourth International Conference
10.15680/IJIRCCE.2015. 0309001 Vol.
on, Mathura, 2012, pp. 525-529.doi:
3, Issue 9, September 2015.
10.1109/CICN.2012.106.
8
Page 1-9 © MANTECH PUBLICATIONS 2017. All Rights Reserved
(CICN),
Journal of Computer Aided Parallel Programming
Volume 2 Issue 1
9) Xiong Xiao, Shoichi Hirasaa, Hiroyuki
Takizawa, and Hiroaki Kobayashi, “The
Importance of Dynamic Load Balancing
amongOpen MP Thread Teams for
Irregular Workloads”,IEEE, 529 - 535,
Nov. 2016.
10) Rohit Singhal, Gwan Choi, Rabi N.
Mahapatra,
“Programmable
LDPC
Decoder Based on the Bubble-Sort
Algorithm”, IEEE,6, Jan. 2006.
11) Hadi
Sutopo,
“Multimedia
based
Instructional development: bubble sort
Visualization”, IEEE, 791–794, Sept.
2015.
12) Lun-Min
Shih,
“Faulttolerant
Connectivity
Jimmy
Maximal
on
the
J.M.
Tan,
Local-
Bubble-sort
Graphs”, IEEE,564 – 569, April 2009.
9
Page 1-9 © MANTECH PUBLICATIONS 2017. All Rights Reserved