Вы находитесь на странице: 1из 5

Imperial Journal of Interdisciplinary Research (IJIR)

Peer Reviewed – International Journal


Vol-3, Issue-9, 2017 (IJIR)
ISSN: 2454-1362, http://www.onlinejournal.in

Framework for Dynamic Resource Allocation


and Efficient Scheduling strategies to enable
Cloud for HPC platforms.
Tavhare Kisan Sopan, Kamble B.S, Hule K.A.
1
Batchlar of Computer Engineering From Universal College of Engineering Pune.
2.
Assistant Professor Universal College of Engineering, Pune. 3. Assistant Professor
Universal College of Engineering, Pune

Abstract: Resource scheduling and Algorithm: - Genetic Algorithm.


allocation is very critical issue in Introduction: The cloud computing
cloud computing. Any computation is has great potential of providing robust
carried out when there is sufficient or computational power to the society at
proportionate resources available. reduced cost. The dynamic resource
Services are provided to the customers scheduling model for a public cloud
or end users with the proper analysis which has numerous nodes with
of resources. Infrastructure as a distributed computing environments
service in cloud grabs much attention with many different geographic
in Cloud computing. To utilize locations. To be competitive,
resources more efficiently an corporations must minimize
optimized scheduling algorithm is nefficiencies and maximize
used to address cloud scheduling productivity. In manufacturing,
problems. By deploying virtual productivity is inherently linked to
machines in appropriate locations to how well you can optimize the
improve the speed of locating best resources you have, reduce waste and
allocation method which intern permit increase efficiency. Finding the best
maximum utilization of resources way to maximize efficiency in a
available. The scheduling resources to manufacturing process can be
overcome unbalance in assignment extremely complex. Even on simple
problem, in this parallel genetic projects, there are multiple inputs,
algorithm concept is used which is multiple steps, many constraints and
much faster than traditional genetic limited. In general a resource
algorithm Cloud computing plays a constrained scheduling problem
vital role is a model for enabling consists of: A set of jobs that must be
ubiquitous network access to a shared executed, finite set of resources that
pool of configurable computing can be used to complete each job,
resources. Any cloud provides associated with set of constraints that
services mainly three ways software must be satisfied. Constraints
as a service (SaaS), platform as a basically of two types in Temporal
service (PaaS) and infrastructure as a Constraints the time window to
service (IaaS). complete the task, Procedural
Keywords:-Cloud computing; Constraints is the order each task must
Resource Scheduling; VRaaS (virtual be completed and Resource
resource as a service); VMware; Constraints is the resource available
Parallel genetic with set of objectives to evaluate the

Imperial Journal of Interdisciplinary Research (IJIR) Page 1


Imperial Journal of Interdisciplinary Research (IJIR)
Peer Reviewed – International Journal
Vol-3, Issue-9, 2017 (IJIR)
ISSN: 2454-1362, http://www.onlinejournal.in

scheduling performance. Clouds can designed for large static clusters.


be used to provide on-demand Recently there was lot of research went
capacity as a utility, although the on parallel data processing and its
realization of this idea can differ implications and possibilities. Many
among various cloud providers, the systems came into existence for
most flexible approach is the processing MTC applications where
provisioning of virtualized resources as a parallel processing of data is essential.
service (VRaaS). Cloud computing
emerges as a new computing Challenges and Opportunities:
paradigm which aims to provide They consider the number of
reliable, customized and QoS (Quality available machines to be constant,
of Service) guaranteed computing especially when scheduling the
dynamic environments for end-users. processing job’s execution. While
Cloud computing is the delivery of IaaS clouds can certainly be used to
computing as a service rather than a create such cluster-like setups, much
product, whereby shared resources, of their flexibility remains unused. To
software and information are provided integrate the cloud computing task
to users over the network. such as portfolio access these services
Related work: - All such systems have a and to deploy their programs for
common goal such as fault tolerance and efficient parallel processing. Each
parallel execution of tasks and they are vertex in the graph represents process
being used in different fields. Open flow edges define the communication
source version of MapReduce i.e. between these tasks also decided to
Hadoop or the MapReduce were use Directed Acyclic Graph is relevant
designed to run jobs in parallel in cost to Nephele. The user must write the
effective manner using commodity program code for external task must
servers. For simplicity an example be assigned to a vertex and must be
framework is MapReduce. Once job is connected by edges to define the
given to it, it automatically takes care of communication paths of the job. In
dividing the given job into tasks and this paper we have discussed the
spreading them across the available challenges and opportunities for
servers. There are two programs involved efficient parallel data processing in
namely Map and Reduce for specific cloud environments and presented.
functionality. There are many other Resource Allocation and Its
programs that coordinate with the jobs of Significance:
MapReduce nature. MapReduce is 1) Resource contention arises when
designed to run data analysis jobs on a two applications try to access the
large amount of data, which is expected same resource at the same time.
to be stored across a large set of share- 2) Scarcity of Resource - Scarcity
nothing commodity servers. MapReduce of resource arises when there are
is highlighted by its simplicity: Once a limited resources and the demand for
user has fit his program into the required resources is high.
map and reduce pattern, the execution 3) Resource Fragmentation -
framework takes care of splitting the job Resource fragmentation arises when
into subtasks, distributing and executing the resources are isolated. There
them. A single MapReduce job always would be enough resources but cannot
consists of a distinct map and reduce allocate it to the needed application
program. MapReduce has been clearly due to fragmentation into small

Imperial Journal of Interdisciplinary Research (IJIR) Page 2


Imperial Journal of Interdisciplinary Research (IJIR)
Peer Reviewed – International Journal
Vol-3, Issue-9, 2017 (IJIR)
ISSN: 2454-1362, http://www.onlinejournal.in

entities. Over Provisioning - Over A. Model:-


provisioning arises when the In this a scheduler is available
application gets surplus resources than at each node in the cloud. Mainly
the demanded one three activities are carried out in the
4) Under Provisioning - Under scheduling. At first the system
provisioning of resources occurs when maintains the list of idle resources, list
the application is assigned with fewer of virtual machines availability will be
numbers of resources than it updated all at regular intervals of time
demanded. and each time new VM requests may
Problem definition:- come or VMs are in shutdown state or
Schedulers used in PGA are any change in the physical resources
unique and separate this can access are being detected. Then use the PGA
information’s on all requests received to find the optimal allocation
by the core, based on the particular sequence pattern. Later allocate the
request which keep tracks current specified physical machines to the
allocations, creating and updating a VM requests. Several important
resource schedule and sending the factors we must consider in the
appropriate deployment commands to process of solving unbalanced
the cloud Core. There is request is assignment problem with our PGA.In
referred as instance requests (IRs) assigning the requested resources
and m idle or partially idle computing need to consider different genetic
nodes available in the cloud. At this algorithms parameters such as
stage the main problem is to find the chromosome representation, suitable
allocation sequence which makes the fitness function design and use of
utilization rate of the resource to be proper migration method.
achieved as maximum. VMs are B. Query Processing:-
allocated on physical resources Processing is based on
according to a ranking algorithm implementation of the theorem uses
even though which fails to achieve (network-based) search operations as
higher VMs utilization rate. To off the shelf building blocks. Thus, the
achieve an optimal VMs allocation NAP query evaluation methodology is
sequence, GA seems to be a good readily deployable on existing
choice, but it can be very demanding systems, and can be easily adapted to
in terms of computation load and different network storage schemes. In
memory. With this Parallel Genetic this case, the queries are evaluated in
Algorithm (PGA) has been promoted, a batch. We propose the network-
the parallel implementations of GA, based anonymization and processing
which can provide considerable gains (NAP) framework, the first system for
in terms of performance and anonymous query processing in road
scalability. PGA can easily be networks. NAP relies on a global user
implemented to address resource ordering and bucketization
scheduling problem. thatsatisfies reciprocity and
guarantees K-anonymity. We identify
the ordering characteristics that affect
subsequent processing, and
qualitatively compare alternatives.
System Architecture:- Then, we propose query evaluation

Imperial Journal of Interdisciplinary Research (IJIR) Page 3


Imperial Journal of Interdisciplinary Research (IJIR)
Peer Reviewed – International Journal
Vol-3, Issue-9, 2017 (IJIR)
ISSN: 2454-1362, http://www.onlinejournal.in

techniques that exploit these One of the flexibility in


characteristics. genetic algorithm is fitness function.
C. Parallelization In this system CPU the number of
One fundamental idea to refine cores, memory capacity and disk
the scheduling strategy for recurring capacity is considered to achieve
jobs is to use feedback data. We maximum utilization of resources.
developed a profiling subsystem for Java genetic algorithm package can be
Nephele+ which can continuously used which is a combination of
monitor running tasks and the genetic algorithm and genetic
underlying instances. Based on the programming component as a
Java Management Extensions (JMX) framework that can be easily used in
the profiling subsystem is, among evolutionary principles. GA is so
other things, capable of breaking powerful technology available to
down what percentage of its solve problems with a huge solution
processing time a task thread actually space having limited time and CPU
spends processing user code and what power. Java genetic algorithm
percentage of time it has to wait for package (JGAP) framework which
data. With the collected data also supports multiple threads concept
Nephele+ is able to detect both to implement and tested on windows
computational as well as I/O XP. Virtual machines in the system
bottlenecks. While computational offer different services to the users by
bottlenecks suggest a higher degree of accepting request as Information
parallelization for the affected tasks, request (IR). When the request arrives
I/O bottlenecks provide hints to at the system, scheduler will updates
switch to faster channel types. (Like the list of idle resources and assign the
in memory channels) and reconsider request to particular node, nodes
the instance assignment. Since associated with active server will
Nephele+ calculates a cryptographic provide services. The speed of the
signature for each task, recurring tasks parallel genetic algorithm is high with
can be identified and the previously genetic algorithm in finding best
recorded feedback data can be allocation scheme and utilization rate
exploited. factor for PGA is better than GA. Due
D. Advantages:- to multiple thread concepts in PGA
Major advantage of resource performance will be more acceptable
allocation is that user neither has to with GA.
install software nor hardware to Conclusion:-By using genetic
access the applications, to develop the algorithm to address scheduling of
application and to host the application resources problem in cloud this system
over the internet. Also there is no associated with parallel genetic algorithm
limitation of place and medium. We (PGA) to achieve the optimization and
can reach our applications and data sometime sub-optimization for cloud
anywhere in the world, on any system. scheduling problems. Genetic operators
Cloud providers can share their like mutation, cross over etc are
resources over the internet during deployed in the system to obtain desired
resource scarcity. performance with efficiency in resources
allocation. Representing scheduling
Fitness Function:- problem as unbalanced assignment
problem with server node concept and

Imperial Journal of Interdisciplinary Research (IJIR) Page 4


Imperial Journal of Interdisciplinary Research (IJIR)
Peer Reviewed – International Journal
Vol-3, Issue-9, 2017 (IJIR)
ISSN: 2454-1362, http://www.onlinejournal.in

proper scheduler to list the idle nodes to


assign incoming request for the specific
node to perform useful task. Compared
to traditional GA, this parallel genetic
algorithm improved the speed of locating
the best allocation. Utilization rate of the
resources is also increases with precision.

Future Work:-System can be designed


and deployed with dynamic load
balancing concept in cloud environment;
Bigdata concept on cloud is another
approach to consider, cloud environment
consists of numerous nodes to be
monitored properly to move the
incoming request dynamically to the idle
or normal nodes. Number of requests by
the users will be satisfied by reassign the
incoming request to other nodes to
process when previous node becomes
overloaded. Any nodes in the cloud are in
three states as Idle, busy, normal. Using
proper well defined algorithm can obtain
scalability and efficiency in providing
unbreakable services.

References:
1. Atsuo Inomata, TaikiMorikawa,
Minoru Ikebe, Sk.Md. MizanurRahman:
Proposal and Evaluation of Dynamic
Resource Allocation Method Based on
the Load of VMs on IaaS (IEEE, 2010),
978-1-4244-8704-2/11.
2. “Adaptive Management of Virtualized
Resources in Cloud Computing Using
Feedback Control,” in First International
Conference on Information Science and
Engineering, April 2010, pp. 99-102.
3. Buyya, R., Yeo, C. S., & Venugopal, S.
(2008). Market-oriented cloud
computing: Vision, hype, and reality for
delivering I services as computing
utilities. In HPCC '08: Proceedings of
the2008
4. 10th IEEE International Conference on
High Performance Computing and
Communications (pp. 5{13).
Washington, DC.

Imperial Journal of Interdisciplinary Research (IJIR) Page 5