Performance Improvement of Heterogeneous Hadoop Cluster using Ranking Algorithm

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 12 | Dec-2017 www.irjet.net p-ISSN: 2395-0072
Performance Improvement of Heterogeneous Hadoop Cluster Using
Ranking Algorithm
Shivani Soni1, Nidhi Singh2
1 M.tech Scholar, Dept. of Computer Science Engineering, L.N.C.T College, Bhopal (M.P), INDIA
2 Professor, Dept. of Computer Science Engineering, L.N.C.T College, Bhopal (M.P), INDIA
---------------------------------------------------------------------***---------------------------------------------------------------------
ABSTRACT: Enhancing technologies increases the use and
growth of information technology. As wecanseethegrowth of
data which is increasing rapidly in every single minute. This
exponential growth in data is one of the reason for the
generation of rapid data. Stored data is processed to extract
worth from inaccurate data to form a way for the paralleland
distributed processing for Hadoop. AllthenodesinHadoop are
assumed to be in homogeneous nature but it is not same as it
looks like, in cloud different configuration systems are used
which represents it logically. So data placement policy is used
to distribute data on the basis ofpowerofnode. Dynamicblock
placement strategy is used in Hadoop, this strategy work as
the distributing input data blocks to the nodes on the basis of
its computing capacity. The proposed approach balance and
reorganize the input data dynamically in accordance with
each node capability in a heterogeneous nature. Datatransfer
time is reduced in the proposed approach with the
improvement in performance. Block placement strategy, page
ranking algorithm andsamplingalgorithmstrategiesareused
in the proposed approach. The data placement strategy used
works as decreasing the execution time and improving the
performance of the clusters which are of heterogeneous
nature. Big data are handled using Hadoop. Small files are
handled using applications on Hadoop so that the issue of
performance can be reduced on the Hadoop platform. Better
performance improvement is shown in the proposed work.
Keywords: Hadoop; block placement strategy; page
ranking algorithm; sampling algorithm
1. INTRODUCTION
In the year 2006, Doug CuttingandMikeCafarella,developed
Hadoop as a framework which is an open source computing
and processing of large datasets in distributedenvironment.
System failure and data loss can be reduced in the proposed
work. Failure of any node does not matters because of the
availability of thousands of interconnected node.Thousands
of data and its frequent transfer is tackled among the nodes.
Big data is handled by the Hadoop.
Hadoop is widely known because of its increase in
popularity and handling of big data. To achieve better
performance several techniques are used.
1.1 Overview of Hadoop:
The Google Map Reduce programming model implemented
the open source framework called Apache Hadoop. Hadoop
distributed file system HDFS and Map Reduce model are the
two key parts of the Hadoop framework. HDFS stores
application data for Map Reduce and map Reduce process
the data using logics on map Reduce to store data on HDFS.
Versions available of Hadoop are Hadoop 1.x and Hadoop
2.x. YARN is the improvement feature of Hadoop 2.x. Node
Manager and Resource Manager are thetwocompositions of
YARN, where both works differently. For deploying and
managing resources, Resource Manager is responsible and
for managing and reporting data node status to resource
manager, Data node is responsible.
Figure 1. Hadoop Ecosystem
1.2 Big Data:
The collection of massive amount of data is called Big Data.
This data can be in any form either structured or
unstructured, relational ornon-relational.Unstructured data
is the audio, video, text, image or any different pattern of
data. In recent years, Big data has become very popular in
several different fields. This is a big opportunity in business
field. Large transmission and communication of data
generates large data from various sources. The need of data
mining algorithm is required to process big data. Earlier
production of large data is responsible due to corporate
world but in recent years users have become responsiblefor
its data.
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1339

1.3 Dynamic Block Placement Strategy:
Dynamic Block Placement Strategy works on two basis
Homogeneous cluster and Heterogeneous cluster.
Figure 2. Block Placement Strategy
Homogeneous Cluster:
Depending upon the availability of space in a cluster data is
distributed among the nodes in homogeneous cluster.
Hadoop has a feature of balancing the data, thisfunctionality
of balancing is called Balancer,whichbalancethedata before
running the applications. Whenever conditionsoccurlikeon
any node large data is accumulatedtheninthiscasebalancer
is an important functionality. Replication is an important
function, balancer is responsible for it to care for
replications. It is the key feature in data movement.
Figure 3. Homogeneous Cluster
Heterogeneous Cluster :
Data is transferred from one node to another node ideally in
an heterogeneous environment. The faster node faces
overheads while processing anddata transfer.Thisresultsin
exploring of the issues comes at the time of data transfer.
Data placement policy explores the arising consequences.
And this all occurs in the heterogeneous environment.
Implementation of data placement policy provides the
details of better goals.
Figure 4. Heterogeneous Cluster
2. RELATED WORK
Many of the author stated and researchedabouttheBigdata,
deal with them and also checks there functionality to work
on homogeneous and heterogeneous cluster of data.
Jeffrey Dean et al. In [1] described about Hadoop, Hadoop
process terabytes of data which is of large amount. The
Apache group written Hadoop using java technology. It
works as parallel processing to process large clusters.
Hadoop is attractive and open source framework. It process
the data and replicates it in reliable manner. It is designed in
a manner to run commodity cluster. Low cost low
performance working in parallel is preferred by commodity
computing. HDFS is an Apache project for Hadoop, it is
distributed, low-cost and have high fault-tolerance file
system. It is convenient for largedata andprovideswithhigh
throughput. Its deploymentisnotcostly.Single namenodein
each cluster maintains file system of Meta data and
application data is stored by multiple nodes. Map Reduce
analyzes large data and advantageous for various
organization. Machine learning, indexing, searching, mining
are some map Reduce applications. Traditional SQLareused
to implement these applications. Also helps in data
transformation,parallelization, network communicationand
handling fault tolerance.

Andrew Wang et al. In [2] explained about HDFS which is a
distributed file system. It stores files across cluster node
redundantly for the purpose of security. HDFS divides files
into blocks and replicates them depending on the factor of
replication. The block placement policy is the default in
HDFS, it works as distributing blocks across cluster node.
Many conditions like unnecessary load on cluster can be
possible at any time which results in reducing the overall
performance of cluster.
Konstantin Shvachko et al. In [3] proposed about block
placement which plays a vital role in performance and data
reliability terms. Reliability, availability and network
utilization are improved by this data block placement
strategy. At the time of creation of new block, the first
replicated block is assigned in first location of the block.And
the other replicates will be assigned randomly on different
nodes keeping in mind that only two replicates should be
placed in a single rack. Name node provides data node for
HDFS, which helps in reducing network traffic andimproves
performance.
Fang Zhou et al. In [4] describes that application master
generates input splits in a Hadoop map Reduce. One input
split is generated for a small file. One map container can use
only one input split, which explains that number of input
splits and number of map containers are equal, which is the
issue because it creates many map container for a small file.
If many containers are created then it require many
processes, resulting in many overheads. Similar overheads
are generated for reduce container
3. PROBLEM DOMAIN
Hadoop handles the big data, which is now become a great
deal to manage because of generation of data in every single
minute. Big data is also the opportunity for business. But
here we are looking forward on the issue of homogeneous
and heterogeneous cluster.
Homogeneous cluster where clusters nodes are of similar
form means they are homogeneous but loadallotedonevery
cluster are not similar. This overloading of any cluster
decreases the performance of cluster node and also reduces
overall performance of outcome.
Similar to heterogeneous cluster, where data nodes are of
different sizes and Hadoop works as distributing equal
amount of load on the nodes. In this case if one node of
larger size is experiencing low load then in that case that
particular node completes it work firstly but it waits for the
other node to complete their work because all these nodes
after processing computes on the result, so for it, it waits for
other nodes which results in reducing performance and
increasing computation time.
Problem in it comes as, there are number of cluster nodes of
different sizes like 2GB, 4GB, 6GB and 8GB. For example
these four nodes are taken and the master node here works
as dividing all the nodes with the number of nodes. Means
the master node will divide those nodes with four.
In this mitigation approach, we are working on the issue of
performance and computation time and will be achieved in
our solution domain.
4. SOLUTION DOMAIN
The solution provided to the above problem will be
described as using page ranking algorithm and sampling
algorithm.
Where page ranking algorithm works on the basis of
frequency. The one who is having better frequency will run
first. Page ranking is used for ranking purpose and on the
basis of frequency, ranking is allotted. Weightandfrequency
is calculated using page ranking algorithm and we used it in
work.
Figure 5. Page Ranking
Here, heterogeneous cluster is used and its performance is
increased using ranking algorithm which depends on the
frequency of occurrence. Whose frequency is maximum will
be run first.
And sampling algorithm works as randomly selecting the
nodes instead of mentioning all the possible samples.
Probability of selecting is the sum which is equal to the
sample size of data n. So, will result in increase in
performance with reducing computation time and
distributing the overall load in equal to all the data nodes.
5. RESULT ANALYSIS
Result analysis of the proposed work describes thatthetime
required in proposed work is less then the time required in
existing work. The below table shows the data of variant file
sizes at different time and time is taken in second.

TABLE 5.1. Comparison table of existing work and
proposed work
File Sizes Original System Proposed System
500 MB 249 sec 101 sec
1 GB 478 sec 226 sec
1.5 GB 633 sec 337 sec
2 GB 815 sec 462 sec
2.5 GB 1052 sec 560 sec
3 GB 1258 sec 652 sec
The below graph shows the graph representation of
proposed work for different file size at different time.
Figure 6. Graph representing the proposed work for
different file size
The below graph shows the comparison of existing work
with proposed work where the time taken for execution of
variant file sizes in proposed work is less in comparison to
existing work. Existing work requires more execution time.
Figure 7. Comparison graph
6. CONCLUSION
The mitigation approach concluded that the data nodes can
be of any size or of same size should not be overloaded and
each node should assign the equal amount ofloaddepending
on there size which will result in increasing performance.
This paper attempts to improve performance of
heterogeneous cluster in Hadoop using Ranking algorithm
which works on the basis of frequency,theone whoishaving
maximum frequency will be executed and run first. It
reduces the computation time and improving performance.
7. FUTUTRE WORK
Depending on the allotment of work load on every node this
issue of performance arises. In future implementation
performance can be more improved using different
techniques.
8. REFERENCE
1. Jeffrey Dean and Sanjay Ghemawat, Mapreduce:
simplified data processing on large clusters,
Communications of the ACM, vol. 51, no. 1, pp. 107113,
2008.
2. Andrew Wang. Better sorting in Network Topology
pseudo Sor tBy Distance when no local node is found.
https://issues.apache.org/jira/browse/HDFS-6268,
2014. [Accessed 28-April-2014].
3. Konstantin Shvachko, Hairong Kuang, Sanjay Radia, and
Robert Chansler, The hadoop distributed filesystem,in
Mass Storage Systems and Technologies (MSST), 2010
IEEE 26th Symposium on. IEEE, 2010, pp. 110.
4. Fang Zhou, Assessment of MultipleMapReduce Strategies
for Fast Analytics of Small Files, Ph.D. thesis, Auburn
University, 2015.
5. Fang Zhou, Hai Pham, Jianhui Yue, Hao Zou, and Weikuan
Yu, Sfmapreduce: An optimized mapreduceframework
for small files,” inNetworking,Architectureand Storage
(NAS), 2015 IEEE International Conference on. IEEE,
2015.
6. F. Ahmad, S. Chakradhar, A. Raghunathan, and T. N.
Vijaykumar, “Tarazu: optimizing mapreduce on
heterogeneous clusters,” in Proc. of Int’l Conf. on
Architecture Support for Programming Language and
Operating System (ASPLOS), 2012.
7. M. Zaharia, A. Konwinski, A. D. Joseph, R. Katz, and I.
Stoica, “Improving mapreduce performance in
heterogeneous environments,” in Proc. of USENIX
Symposium on Operating System Design and
Implementation (OSDI), 2008
8. H. Herodotou and S. Babu, “Profiling, what-if analysis,
and cost based optimization of mapreduce programs,”
in Proc. Int’ Conf. on Very Large Data Bases (VLDB),
2011

Performance Improvement of Heterogeneous Hadoop Cluster using Ranking Algorithm

More Related Content

What's hot (20)

Similar to Performance Improvement of Heterogeneous Hadoop Cluster using Ranking Algorithm (20)

More from IRJET Journal (20)

Recently uploaded (20)

Performance Improvement of Heterogeneous Hadoop Cluster using Ranking Algorithm