Cache mechanism to avoid dulpication of same thing in hadoop system to speed up the extension

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 11 | Nov-2014, Available @ http://www.ijret.org 299
CACHE MECHANISM TO AVOID DULPICATION OF SAME THING IN
HADOOP SYSTEM TO SPEED UP THE EXTENSION
Ritu A.Mundada1
, Aakash.A.Waghmare2
1
M. E Student, Department of Computer Science, Ssbt College, Jalgaon, India
2
Associate Professor, Department of Computer Science, Ssbt College, Jalgaon, India
Abstract
Cloud computing provides a proper platform for hosting large-scale data-intensive applications. MapReduce is a programming
model as well as a framework that supports the model. The main idea of the MapReduce model is to hide details of parallel
execution and allow users to focus only on data processing strategies. Hadoop is an open-source implementation for MapReduce.
For storage and analysis of online or streaming data which is big in size. Most organization are moving toward Apaches Hadoop
HDFS. Applications like log processors, search engines etc. ueses hadoop Map reduce for computing and HDFS for storage.
Hadoop is popular for analysis, storage and processing of very large data but require to make changes in hadoop system. There is
no mechanism to identify duplicate computations which increase processing time and unnecessary data transmission .To co-locate
related files by considering content and using locality sensitive hashing algorithm. By storing related files in same cluster using
cache mechanism which improve data locality mechanism and avoids repeated execution of task, both helps to speed up execution
of hadoop.
Keywords-Distributed file system, Datanode, Locality Sensitive Hashing
-------------------------------------------------------------------***-------------------------------------------------------------------
1. INTRODUCTION
In present scenario with the internet of things a lot of data is
generated and is analyzed mainly for business intelligence.
There are various sources of Big Data like social networking
sites, sensors, transactional data from enterprise
applications/databases, mobile devices, machine generated
data, huge amount of data generated from high definition
videos and many more sources. Some of the sources of this
data have vital value that is helpful for businesses to
develop. Hadoop is a distributed computing platform written
in Java which incorporates features similar to those of the
Google File System and MapReduce programming
paradigm. Apaches Hadoop is open source implementation
of Google Map/Reduce framework, it enables data intensive,
distributed and parallel applications by diving massive job
into smaller tasks and massive data sets into smaller
partition such that each task processes a different partition in
parallel. Map tasks that process the partitioned data set using
key/value pairs and generate some intermediate result.
Reduce tasks merged all intermediate values associated with
keys. Hadoop uses Hadoop Distributed File System (HDFS)
which is distributed file system, used for storing large data
files. Each file is divided into numbers of blocks and
replicated for fault tolerance. HDFS cluster is based on
master/slave architecture. Name Node work as master which
manages and store the file system namespace and provide
access to the client. The slaves are number of Data Nodes.
HDFS provides a file system name space and allows user
data to be stored in files. File is divided into number of
block; size of block is normally 64MB which is too large.
The default placement of Hadoop does not consider any data
characteristics during placement. If related files are kept in
same set of data nodes, the access latency and efficiency
will be increased. The file similarity will be calculated by
comparing content of it and to reduce comparison, a
Locality Sensitive Hashing will be used. Hash function hash
the points using different hash function in such way that
probability of collision will be higher for similar points.
Client is controlling overall process and providing sub-
clusterid where file will be placed otherwise default
placement strategy is used. Data aware cache is introduced
for avoiding execution of repeated task, which requires each
data object indexed by its content and also implement cache
request and reply protocol.
In section II we discuss in more about some basic types of
schedulers used in Hadoop and scheduler improvements.
Further section III talks about related work. section IV
actual mechanisms of cache. Finally section V concludes the
paper after which references follow.
2. SCHEDULING IN HADOOP
The default Scheduling algorithm is based on FIFO where
jobs were executed in the order of their submission. Later on
the ability to set the priority of a Job was added. Facebook
and Yahoo contributed significant work in developing
schedulers i.e. Fair Scheduler and Capacity Scheduler
respectively which subsequently released to Hadoop
Community.
2.1 Default FIFO Scheduler
The default Hadoop scheduler operates using a FIFO queue.
After a job is partitioned into individual tasks, they are
loaded into the queue and assigned to free slots as they
become available on TaskTracker nodes. Although there is

_______________________________________________________________________________________
support for assignment of priorities to jobs, this is not turned
on by default. Typically each job would use the whole
cluster, so jobs had to wait for their turn. Even though a
shared cluster offers great potential for offering large
resources to many users, the problem of sharing resources
fairly between users requires a better scheduler. Production
jobs need to complete in a timely manner, while allowing
users who are making smaller ad hoc queries to get results
back in a reasonable time.
2.2 Fair Scheduler
The Fair Scheduler was developed at Facebook to manage
access to their Hadoop cluster and subsequently released to
the Hadoop community. The Fair Scheduler aims to give
every user a fair share of the cluster capacity over time.
Users may assign jobs
to pools, with each pool allocated a guaranteed minimum
number of Map and Reduce slots. Free slots in idle pools
may be allocated to other pools, while excess capacity
within a pool is shared among jobs. The Fair Scheduler
supports preemption, so if a pool has not received its fair
share for a certain period of time, then the scheduler will kill
tasks in pools running over capacity in order to give the slots
to the pool running under capacity. In addition,
administrators may enforce priority settings on certain
pools. Tasks are therefore scheduled in an interleaved
manner, based on their priority within their pool, and the
cluster capacity and usage of their pool. As jobs have their
tasks allocated to Task Tracker slots for computation, the
scheduler tracks the deficit between the amount of time
actually used and the ideal fair allocation for that job.
2.3 Capacity Scheduler
Capacity Scheduler originally developed at Yahoo addresses
a usage scenario where the number of users is large, and
there is a need to ensure a fair allocation of computation
resources amongst users. The Capacity Scheduler allocates
jobs based on the submitting user to queues with
configurable numbers of Map and Reduce slots. Queues that
contain jobs are given their configured capacity, while free
capacity in a queue is shared among other queues. Within a
queue, scheduling operates on a modified priority queue
basis with specific user limits, with priorities adjusted based
on the time a job was submitted, and the priority setting
allocated to that user and class of job. When a Task Tracker
slot becomes free, the queue with the lowest load is chosen,
from which the oldest remaining job is chosen. A task is
then scheduled from that job. Overall, this has the effect of
enforcing cluster capacity sharing among users, rather than
among jobs, as was the case in the Fair Scheduler.
2.4 Delay Scheduling
Fair scheduler is developed to allocate fair share of capacity
to all the users. Two locality problems identified when fair
sharing is followed are – head-of-line scheduling and sticky
slots. The first locality problem occurs in small jobs (jobs
that have small input files and hence have a small number of
data blocks to read). The problem is that whenever a job
reaches the head of the sorted list for scheduling, one of its
tasks is launched on the next slot that becomes free
irrespective of which node this slot is on. If the head-of-line
job is small, it is unlikely to have data locally on the node
that is given to it. Head-of-line scheduling problem was
observed at Facebook in a version of HFS without delay
scheduling. The other locality problem, sticky slots, is that
there is a tendency for a job to be assigned the same slot
repeatedly. The problems aroused because following a strict
queuing order forces a job with no local data to be
scheduled.
2.5 Dynamic Priority Scheduling
Thomas Sandholm et al .proposed Dynamic Priority
Scheduler that supports capacity distribution dynamically
among concurrent users based on priorities of the users.
Automated capacity allocation and redistribution is
supported in a regulated task slot resource market. This
approach allows users to get Map or Reduce slot on a
proportional share basis per time unit. These time slots can
be configured and called as allocation interval. It is typically
set to somewhere between 10 seconds and 1 minute. For
example a max capacity of 15 Map slots gets allocated
proportionally to three users. The central scheduler contains
a Dynamic Priority Allocator and a Priority Enforcer
component responsible for accounting and schedule
enforcement respectively. This model appears to favor users
with small jobs than users with bigger jobs. However
Hadoop MapReduce supports scaling down of big jobs to
small jobs to make sure that fewer concurrent tasks runs by
consuming the same amount of resources.
3. RELATED WORK
Performance of Hadoop system will be improved if related
files are placed in similar set of nodes. Considering past
work some techniques are used which provides some degree
of colocation but needs lots of changes in framework.
Colocating related file in HDFS, Co-Hadoop[1],provides
solution, which helps the application to control data
placement at file system level.
Considering this ADAPT[3], works to achieve high
reliability without need of additional data replication facility
. Based on availability of host ADAPT distribute data blocks
dynamically which improves data locality and reduces
network traffic. ADAPT guarantees that all host finishes
processing of their allocated job at same time which
improve execution time. When there are large numbers of
small files, each having less than size of block size of HDFS
then that file size become block size.
DARE[2]Adaptive DataReplication for Efficient Cluster
Scheduling .DARE improves data locality without network
overhead but requirementof storage is very high.
Down analyzes the performance of widely used hadoop
schedulers including FIFO and Fair sharing and compares
them with the COSHH (Classi_cation and Optimization

_______________________________________________________________________________________
based Scheduler for Heterogeneous Hadoop) scheduler. The
scheduler is a combination of the three analyzed algorithms.
The selector chooses an appropriate scheduler as the number
of jobs and resources scale up or down. However, the
overall solution is to use the COSHH algorithm when the
system is overloaded (e.g., during peak hours), the FIFO
algorithm for under loaded systems (e.g., after hours), and
the Fair Sharing algorithm when the system load is
balanced. A combination of the FIFO, Fair Sharing, and
COSHH schedulers is e_ective, where the selection is based
on the load on the system and available system resources. A
job scheduler is an essential component of every hadoop
system.
4. PROPOSED SOLUTION
Data is everywhere now. The amount of information
available now is very huge for analysis, storage and process
such huge information Apaches Hadoop tool is much
popular. Hadoop uses HDFS for storage and MapReduce for
analysis. If default placement of Hadoop is considered then
it places file anywhere in cluster. Hadoop is uses principal
of data locality means tasks are executed where data are
placed but in practices this will not be true for all data files.
If needed files are placed on different nodes then that files
need to be copied to worker node for task execution. But
when placing files data characteristics are considered then
related files are stored in same node which improve data
locality and reduce network traffic. MapReduce does not
have any mechanism to find whether task is executed before
so that result can be re used but result of executed task is not
stored so if task is executed again then there is no
mechanism is available to find task and reuse the result. The
purpose of the experiment is to extend Hadoop system by
improving the data placement and execution policies of
Map/Reduce. The file similarity will be calculated based on
its content and similar file will be placed in same data node
or nearby data nodes (subcluster). The result of
Map/Reduce will be stored in cache as framework to access
them again if same task is executed on same data repeatedly.
To find cluster client executes following modules:
1. Preprocessing File: File contain collection of words,
file is pre-process means words like stop words are
removed, stop word are word like „a ‟, „of ‟, „the‟ etc.
and also stemming (historical is replace with history )
and many techniques are used to pre-process a file.
After preprocessing file will contain collection of
word which related to particular file and which can be
use to represent that file.
2. File Vector: after preprocessing file which contains
collection of words from that words which are
presenting that file need to find, this is done using
TFIDF technique. TFIDF(Term Frequencies-Inverse
Document Frequencies) technique finds words in file
that come many times compare to all remaining files,
which indicate that word is representinga file and it is
important word in file. If word is representing that
file then that word can be use to find similar files.
3. Create Signature - To find similar file it should be
compared with content of each and every files
available but there are millions of files which makes
process time consuming. So to make process faster
compact bit representation of each file vector is
created, Signature. To create Signature f bit vector is
used and this vector initialized to zero first then it
hashed with file vector and comparing value is 0 or 1
weight of word will be incremented or decremented.
Advantage of Signature is that similar file will have
same Signature which makes process faster.
4. Use Locality Sensitive Hashing to find nearest
neighbor- In large clustering environment to compare
file Signature to each and every cluster is time
consuming to avoid comparing each and every cluster
locality sensitive hashing technique is used which
ensures that only nearest neighbour need to be
checked to place file. For this hashing function is
used which query file Signature to find nearest
neighbour and m number of neighbour is returned to
client.
5. Store file with related files- If m neighbour is return
to client then only that m neighbour will be compared
and after finding cluster where file will be placed, this
subclusterid will be given to Name Node. Name
Node maintains subclustertable which store
subclusterid and file placed on that cluster. If Name
Node finds entry then that file will be placed on
subclusterid but if subclusterid is not found then new
subcluster will be created, file will be stored on newly
created cluster and file and cluster Signature will be
calculated and this information will be updated to
subcluster table.Now suppose client want execute
map task and system should not execute repeated map
task for this, cache will be implemented. Cache table
will be created which stores file name, operation
perform on that file and result file name.
1. Map task execution: when client wants to
execute any map task first then it request cache
manager to find file name and operation. If file
name and operation performed on that file is
same then result file name will be given to
directly to reduce phase which completely save
execution time of task.
2. Lifetime of cache item fixed size cache will be
used and if cache is full then older entry will be
deleted.
The project requires various data structure to perform
various modules in proposed system which is listed below:
Data structure for locality sensitive hashing function, Data
structure for SubCT, Data structure for storing mapping
information, Data Structure for CacheTable, Data Structure
for storing intermediate result. All above structure will be
either array of structure or linked list or object of classes.
The internal data structures will be used to store result
obtained by map task it need to store locally because it
improves data locality. The data structure to create mapping
information at client will contain 1) clusterid, 2) Signature
of cluster ids 3) cluster centroids 4) file name 5) Signature

_______________________________________________________________________________________
of file. The data structure require to store hash table at client
side which store cluster Signature as key and cluster
information as value. The global data structures use for
maintaining Subclusterid table for indexing of Sub-cluster id
and file which is having same subclusterid. etc. The
structure of Sub-cluster is as follow:
Table 1 Sub-Cluster Table
Sub Clusterid Files
1 A,B
2 C
N 0
The data structures will be structure used to data structure to
create CacheTable will contain 1) name of file, 2) type of
operation performed on file 3) result file name.
5. CONCLUSION
“Speedup extension to Hadoop system” is the modification
in the input format and task management of the map reduce
framework. The applications, using this modified Hadoop
need not to change at all. The proposed system shows that it
can eliminate all the duplicate tasks and new approach for
incremental file clustering is proposed for HDFS which will
cluster similar files in the same set of data nodes with
minimal changes to the existing framework. For faster
clustering operations bit wise representation of the feature
vectors called Signature are used. To reduce the number of
cluster centroid comparisons, only the nearest neighbours
are considered using the technique of Locality Sensitive
Hashing. In this experiment two modules are combined first
is data placement modified with clustering and second is
map/reduce tasks execution time is reduce and processing is
done faster. So this experiment speeds up hadoop system by
changing data placement and task execution.
REFERENCES
[1] Yongqiang He, Rubao Lee, Yin Huai, Zheng Shao,
“RCFile: A fast and space-efficient data placement
structure in MapReduce-based warehouse systems,”
Data Engineering (ICDE), 2011 IEEE 27th
International Conference on , vol., no., pp.1199,1208,
11-16 ,April 2011.
[2] Abad, C.L., Yi Lu, Campbell, R.H.,
“DARE:Adaptive Data Replication for Efficient
ClusterScheduling,” Cluster Computing (CLUSTER),
2011IEEE International Conference on , vol., no.,
pp.159,168, 26-30 Sept. 2011.
[3] Hui Jin; Xi Yang; Xian-He Sun; Raicu, I., “ADAPT:
Availability-Aware MapReduce Data Placement for
Non-dedicated DistributedComputing,” Distributed
Computing Systems(ICDCS), 2012 IEEE 32nd
18-21 June 2012.
[4] Shvachko, K.; Hairong Kuang; Radia, S.; Chansler,
R., “The Hadoop Distributed File System,”Mass
Storage Systems and Technologies (MSST), 2010
IEEE 26th Symposium on , vol., no., pp.1,10, 3-7
May 2010.
[5] Kala Karun,„„A review on hadoop HDFS
infrastructure extensions,”Information and
Communication Technologies (ICT), 2013 IEEE
Conference on , vol., no., pp.132,137, 11-12 April
2013.
[6] Kala, K.A., Chitharanjan,„„Locality Sensitive
Hashing based incremental clustering for creating
affinity groups in Hadoop HDFS – An infrastructure
extension,”Circuits, Power and
ComputingTechnologies (ICCPCT), 2013
20-21 ,March 2013.
[7] Yaxiong Zhao, Jie Wu, „„Dache: A data aware
caching for big-data applications using the Map
Reduce framework,” INFOCOM, 2013 Proceedings
IEEE , vol., no., pp.35,39, 14-19 April 2013.
[8] Hui Jin; Xi Yang; Xian-He Sun;,„„ADAPT:
Availability-Aware MapReduce Data Placement for
Non-dedicated Distributed Computing,”
DistributedComputing Systems (ICDCS), 2012 IEEE
32nd International Conference on , vol., no.,
pp.516,525, 18-21 June 2012

Cache mechanism to avoid dulpication of same thing in hadoop system to speed up the extension

More Related Content

Cache mechanism to avoid dulpication of same thing in hadoop system to speed up the extension