A Survey on Neural Network Based Minimization of Data Center in Power Consumption

127
E. Brindha and S. Sobitha Ahila, “A survey on neural network based minimization of data center in power consumption,” International
Journal of Scientific and Technical Advancements, Volume 2, Issue 1, pp. 127-130, 2016.
International Journal of Scientific and Technical Advancements
ISSN: 2454-1532
A Survey on Neural Network Based Minimization of
Data Center in Power Consumption
E. Brindha1
, S. Sobitha Ahila2
1, 2
Easwari Engineering College, Tamilnadu, India
Email address: 1
brindha235@gmail.com, 2
sobithaahila@gmail.com
Abstract— Cloud computing is the current technology used for sharing and accessing resources via internet. Reducing power consumption
is an essential requirement for cloud resource providers to decrease operating costs. We employ the predictor to predict future load demand
based on historical demand. According to the Prediction, the algorithm turns off unused servers and restarts them to minimize the number of
running servers. Power consumption can be regulated by using proper load balancing technique. Load balancing is done so to distributing
the load fairly amidst the servers and also a scheduling technique is followed to selectively hibernate the servers to optimize the energy
consumption. The load balancing is based on load prediction and server selection policy. Neural Network is used for load prediction, which
predicts the future load based on past historical data. The servers can be monitored and given ranking based on their reliability record and
this information is used as a criterion while performing load balancing.
Keywords— Cloud computing; eucalyptus; load balancing; neural network.
I. INTRODUCTION
loud computing is a recent advancement where in
the Information Technology (IT) infrastructure and
applications are provided as “services” to end-users
under a usage-based payment model. The cloud makes it
possible to access the information from anywhere at any time.
While a traditional computer setup requires the user to be in
the same location where the data storage is present, the cloud
makes the storage and retrieval location independent.
Cloud computing offers delivery of on-demand computing
resources everything from application to data centre resources
over the Internet. It offers guaranteed services. It allows
customers to scale up and down their resources based on the
dynamic needs.
One of the most pressing issues in cloud computing is the
resource management. Resource management problems
include allocation, provisioning, requirement mapping,
adaptation, discovery, brokering, estimation and modeling.
Resource management in cloud computing offers the benefits
like scalability, quality of service, optimal utility, reduced
overheads, improved throughput, reduced latency, specialized
environment, cost effectiveness and simplified interface. This
research focuses on some of the important resource
management techniques such as resource provisioning,
resource allocation, resource mapping and resource adaptation.
It brings out an exhaustive survey of such techniques in cloud
computing and also put forth the open challenges for further
research.
A. Cloud Service Models
The services offered by the Cloud service provider are
generally classified into three types of service models.
Software as Services (SaaS):- End-user application is
delivered as a service. Platform and infrastructure is
abstracted, and can be deployed and managed with less effort.
Cloud-based applications run on remote servers in the cloud
that are owned and operated by others and that connect to
users’ computers via the Internet and a web browser. Platform
as a Services: - Services is provided as application platform
onto which custom applications and services can be deployed.
PaaS services can be built and deployed more inexpensively,
although services need to be supported and managed. Platform
as a Service provides a cloud-based environment with
everything required to support the complete life cycle of
building and delivering web-based cloud applications, without
the cost and complexity of buying and managing the
underlying hardware, software, provisioning and hosting. With
PaaS one can develop applications and get to market faster,
deploy new web applications to the cloud in minutes and
reduce complexity with middleware. Infrastructure as a
Services (IaaS):- Consumers control and manage the systems
in terms of the processors, memory, and storage and network
connectivity but do not have control over the cloud
infrastructure. Infrastructure as a Service provides companies
with computing resources on a pay-per-use basis.
B. Deployment Models
Cloud computing offers three types of deployment models,
namely, Private cloud:-The cloud infrastructure is deployed
and maintained for a specific organization. The datacenter
may be in-house or on a third party premises. Public cloud:-
The cloud infrastructure is available to the public on a
commercial basis by a cloud service provider. This enables a
consumer to develop and deploy a service in the cloud with
little financial outlay compared to the capital expenditure
requirements normally associated with other deployment
options. Hybrid cloud: - The cloud infrastructure consists of a
number of clouds of any type, but the clouds have the ability
through their interfaces to allow data and/or applications to be
moved from one cloud deployment model to another. This can
be a combination of private and public clouds that support the
requirement to retain some data within an organization and
also the need to offer services to the general public.
C

128
ISSN: 2454-1532
C. Neural Network
Neural networks, with their remarkable ability to derive
meaning from complicated or imprecise data, can be used to
extract patterns and detect trends that are too complex to be
noticed by either humans or other computer techniques. A
trained neural network can be thought of as an "expert" in the
category of information it has been given to analyze. This
expert can then be used to provide projections given new
situations of interest and answer "what if" questions. Other
advantages include: Adaptive learning: An ability to learn how
to do tasks based on the data given for training or initial
experience. Self-Organization: An ANN can create its own
organization or representation of the information it receives
during learning time. Real Time Operation: ANN
computations may be carried out in parallel, and special
hardware devices are being designed and manufactured which
take advantage of this capability. Fault Tolerance via
Redundant Information Coding: Partial destruction of a
network leads to the corresponding degradation of
performance. However, some network capabilities may be
retained even with major network damage. An artificial
neuron network is a data processing system consisting large
number of highly interconnected processing elements as
artificial neuron in a network structure. By using cloud
computing service for neural networks we can design any type
of neural networks.
D. Load Analysis and Prediction
Load analysis is the process of analyzing the load of the
server in the datacenter on time period basis. The load analysis
will be done more than once to provide a better result.
Fig. 1. A three layer neural network prediction.
Load prediction will be carried out based on the analyzed
data using neural network predictor. The neural network
perceptrons are applied to predict the future load of the
applications. The perceptrons are trained in a supervised
manner with a back propagation algorithm. Perceptrons uses
the values: Number of external inputs, one internal input.
Threshold and single output are prediction. The figure 1 as
three layer neural network predictor in operation with a time
series input is plotted. Load prediction is necessary for
improving power grid management. In recent years, several
methodologies have been advanced to accurately predict
power load. However, most of them cannot handle the
prediction well during unanticipated situations, which may
escalate into serious power failures. We introduce a
methodology seeking improvements in power load analysis
and prediction for unanticipated situations. The new
methodology exploits the synergisms found between artificial
neural networks and fuzzy logic to detect load transients in a
quickest manner. It subsequently identifies the transients and
provides on line prediction for abnormal conditions.
When the network is run, each layer performs the
calculation on the input and transfers the result yn+1 to the next
layer.
Yn+1=h(( (1)
The above equation (1) is used for prediction of future load
based on the input value. Where Yn+1 provides the output of
the current node and n is the number of nodes in the previous
layer, xi is the input of the current node from the previous
layer, b is the bias value and wi is the modified weight based
on the mean square error. Here the neural predictor is
developed and the experiment is performed to prove its highly
accurate prediction ability which fits in dynamic real time.
II. LITERATURE SURVEY
A. Power-Aware Scheduling of Virtual Machines in DVFS-
Enabled Clusters
Gregor von Laszewski et al 2012, Scheduling virtual
machine in a compute cluster to reduce power consumption
via the technique of dynamic voltage frequency scaling.
Specifically, we design of an efficient scheduling algorithm to
allocate virtual machine in DBFS-enabled cluster. A computer
cluster provides various virtual machines and when a job
arrives at the cluster, the cluster scheduler allocates the job
with preconfigured virtual machine then starts it on proper
compute nodes. The job is executed in the virtual machine.
After the job is executed, the virtual machine is shut-down.
Cloud computing technologies, the needs for efficient
algorithm to minimize wasted server energy becomes
increasingly important. As such, the field of green computing
provides a way to prevent unnecessary emission from
contributing to global warming to save large amount of money
on operational cost.
B. Performance Evaluation of a Green Scheduling Algorithm
for Energy Saving in Cloud Computing
Truong vinh Truong Duy et al 2010, we employ the
prediction to predict future load demand based on historical
demand. According to the prediction, the algorithm turns off
unused servers, thus minimizing the energy use at the points of
consumption to benefits all other levels. For evolution, we
simulation with two load traces. The result show that the P20
mode can save up to 46.3% of power consumption with a drop
rate of 0.03% on one load trace, and a drop rate of 0.12% with
a power reduction rate of 46.7% on the other.

129
ISSN: 2454-1532
Green scheduling Algorithm which makes use of neural
network based predictor for energy saving in cloud computing.
The prediction is exploited to predict future load demand
based on collected historical demand. The algorithm used the
prediction in making turning off/on decision to minimize the
number of running servers.
C. A Study on Server Sleep State Transition to Reduce Power
Consumption in a Virtualized Server Cluster Environment
V. K. Mohan Raj et al 2012, Reducing power consumption
is an essential requirement for cloud resources providers to
decrease operating costs. One of the options to reduce power
consumption is to reduce the number of servers in IDLE
(unused) state as these IDLE state can be reduced by turning
off IDLE servers or transitioning these IDLE servers to low
power SLEEP state. With virtualization being the backbone to
provision cloud computing services, we use simulation to
study and report the impact of using SLEEP state on the server
with its virtual machine servicing application workload
requests. The average response time per request, our
simulation result are using SLEEP state server level we
achieve a 2% saving in average power usage and around 27%
saving in average response time per request.
The work deal with physical SLEEP state transition impact
to overall power consumption and response times in a physical
server environment. In this work heuristic approaching for
SLEEP state at server level and using VM”s to processing
application requests the virtualized environment. To show
saving to overall average per application request response time
and marginal workload specific power consumption saving.
D. The Eucalyptus Open Source Cloud-Computing System
D. Nurmi et al 200, Cloud computing that implement what
is commonly referred to as infrastructure as service systems
that give users the ability to run and control entire virtual
machine instances deployed across a variety physical
resources. The operational aspects of the system and
architectural trade-off that we have made in order to allow the
eucalyptus to be portable, modular and simple to use on
infrastructure commonly found within academic settings, we
provides evidence the eucalyptus enables users familiar with
existing grid and HPC systems to explore new cloud
computing functionality while maintaining access to existing,
familiar application development software and grid middle
ware.
Eucalyptus system built to allow administrators and
researchers the ability to deploy an infrastructure for user-
controlled virtual machine creation and control a top existing
resources. It hierarchical design targets resource commonly
found within academic and setting including but not limited to
small and medium Linux clusters, workstation pools, and
server farms. We use a virtual networking solution that
provides VM isolation, high performance and a view of
network that is simple and flat.
E. Genetic Load and Time Prediction Technique for Dynamic
Load Balancing in Grid Computing
Z. Akhtar et al 2007, a genetic new algorithm based task
scheduling has been proposed and its performance has been
proved on many applications on the grid. This addresses the
problem with Priority based Load balancing in cloud for Data
intensive applications. The requests are classified based on a
certain parameter like source IP address, bandwidth of the
cloud consumer, affordability of consumer, SLA required etc.
Normally the data is retrieved from a distributed database
environment since it provides high availability and
redundancy. The scheduling of resources has been classified
from three Cloud Service Model perspectives. The first model
is based on Software as a Service (SaaS) perspective which
aims at providing efficient resource allocation using cost
optimization and priority models. Cost optimization is done by
consolidating the hardware and software resources required to
run the datacenters on specified SLAs and also aim to provide
maximum profit to the cloud service provider. In a datacenter
during the period when resources are available in excess, the
resources can be auctioned. A bidding policy is defined to
ensure a fair and optimum profit generated for the cloud
service provider. Auctioning is generally the technique used to
increase the price of an object in which multiple bids are
compared to determine the highest bid and the object.
F. Load Balancing Incoming IP Request Across a Farm of
Clustered my SQL Server
M. Kaitsa, et al 2009, MySQL Cluster Database provides
services with full capability to cover the peak demands. In
Cloud environment based on demand, requests are assigned to
the different VM’s. Requests are sent across a farm of MySQL
cluster servers in which the data is replicated to avoid single
point failure and also does load balancing, there-by a better
response time is achieved. Advanced eager scheduling
achieves fault tolerance and load balancing by dynamically
breaking down the tasks and by performing parallel
computing.
The last perspective model is based on Infrastructure
which aims at energy saving and power-aware cloud based on
past data and current data. A green scheduling algorithm with
neural based prediction technique, predicts the future load
based on the past history. However this approach causes heavy
overhead due to stop/start of the servers dynamically and
hence leads to strong performance degradation. This thesis
provides a power optimized solution based on neural network
trained with back propagation algorithm. The future load is
predicted using the trained set which has less mean square
error and also high reliability.
III. CONCLUSION
This growing crisis in power shortage has brought a
concern in existing and towards reliable optimal load
balancing with the help of load prediction using neural
network model. The sample loads are gathered from a
datacenter and used an input for prediction. Using past input
data the future load is predicated using the trained data set
with less mean square error and high reliability. From the
predication value the required number of server with high

130
ISSN: 2454-1532
reliability is selected to perform load balancing to optimize
power consumption with higher reliability.
REFERENCES
[1] Z. Akhtar, “Genetic load and time prediction technique for dynamic load
balancing in grid computing,” Infomation Technology Journal, vol. 6,
pp. 978-986, 2007.
[2] V. K. M. Raj, “A study on server sleep state transition to reduce power
consumption in a virtualized server cluster environment,” IEEE Fourth
International Conference on Communication Systems and Networks
(COMSNETS), pp. 1-6, 2012.
[3] D. Nurmi, “The eucalyptus open sources cloud computing system,”
Department of Computer Science, University of California, 2000.
[4] G. V. Lazewski, “Power aware scheduling of virtual machines in DVFS
enables clusters,” Pervasive Technology Institute, Indiana University,
Boomington, 2012.
[5] T. V. T. Duy, Y. Sato, and Y. Inoguchi, “Performance evaluation of a
green scheduling algorithm for energy saving in cloud computing,”
IEEE International Sysmposim on Parallel and Distributed Processing,
Workshops and Phd Forum (IPDPSW), pp. 1-8, 2010.
[6] M. Kaitsa, I. Stavrakas, T. Kontogiannis, I. Daradimos, M. Panaousis,
and D. Triantis, “Load balancing incoming IP request across a farm of
clustered MySQL servers,” Proceeding of the International Conference
on Computer as a Tool, Warsaw, pp. 546-550, 2009.

A Survey on Neural Network Based Minimization of Data Center in Power Consumption

Recommended

More Related Content

What's hot (20)

Similar to A Survey on Neural Network Based Minimization of Data Center in Power Consumption (20)

More from IJSTA (20)

Recently uploaded (20)

A Survey on Neural Network Based Minimization of Data Center in Power Consumption