International Journal of Computer Trends and Technology (IJCTT) volume 9 number 2 Mar 2014

ISSN: 2231-2803 http://www.ijcttjournal.org Page58

Optimizing Resource Allocation in IAAS Clouds
Y.Rajesh Kumar
1
, K.Sirisha
Student, CSE, QIS College of Engineering & Technology, Ongole, India
1

Assistant Professor, CSE, QIS College of Engineering & Technology, Ongole, India
2

Abstract--Cloud computing is a phenomenon which bestows new model of computing. It has various service
modes. The service model which provides on-demand access to computing resources is known as infrastructure-as-
a-service (IaaS). The IaaS allows cloud users to access resources in pay per use fashion. The problem with this
service model in the existing solutions is that the cloud provider may either overprovision the resources or not in a
position to serve large number of requests. At any given time, all users might not need on-demand access. Keeping
this in mind a strategy is required in order to optimize resource allocation. Many applications exists that take care of
high-throughput computing (HTC) where more resources are expected to be allocated. For instance Condor is such
application that assigns available resource to jobs and then the resources are released once job is completed.
Recently Marshall et al. proposed cloud infrastructure that has provision for on-demand allocation of resources
using backfill virtual machines. For this purpose Nimbus is used. In this paper we have implemented the concept of
optimal resource allocation using the notion of HTC users and on-demand users. We built a prototype application
that demonstrates the proof of concept. The empirical results are encouraging.

Index Terms Cloud computing, high throughput computing, infrastructure as a service

I. INTRODUCTION
Cloud computing has become famous in recent years.
For infrastructure services the infrastructure as a
service (IaaS) [1] has become an attractive and
viable alternative. This service of cloud provides on
demand allocation of resources. It also allows to
increase or decrease resource availability of cloud
users. This makes it useful as it supports flexible
allocation of resources. Commercial clouds started
using this using virtual machine images and related
interfaces. Open source implementations of IaaS kind
like Eucalyptus [2] and Nimbus [3] configured such
private clouds. The problem with such
implementation is the proper utilization. Utilization
reflects either over provisioning or inability to
support resources. At any point of time there are
resources not utilized or there are number of cloud
users who could not get in time services. This needs
to be improved. On-demand computing is essential
for the success of IaaS. In case of scientific
computing, it results in higher utilization [4]. Low
utilization is also a potential problem here that needs
to be addressed. Users of IaaS are basically of two
types. They are the users who need on-demand
services while other users need high throughput
computing. Many applications came into existence in
order to ensure good resource utilization. Condor [5]
is one such example where HTC is supported.
Scientific applications utilize the HTC service
provided by IaaS.

In this paper we implement a cloud infrastructure that
combines opportunistic provisioning of idle cloud
nodes with on-demand resource allocation. Our
implementation serves well both HTC users and on-
demand users. In case of HTC users backfill VMs are
used. The VMs are terminated when they need to
serve on-demand users. Backfill VMs are used by
HTC users but when on-demand users need them
they are automatically terminated in order to serve
on-demand users. Moreover it is possible to provide
cycles of VMs to users with lesser cost when
compared to that of on-demand users. This is made
possible because the VMs can be closed based on the
requirement. Backfill VMs thus provide good service
to HTC users and also help the on-demand users if
required. This enables the design to achieve two
goals such as offering high utilization of cloud
infrastructure and also supports resource lease which
is much cheaper. In fact the cloud computing is
realized through virtualization only. Virtual Machine
Monitors (VMM) [6] and Nimbus toolkit [3] can be
used to deploy IaaS clouds as they are flexible and
suitable to many cloud projects. In this paper we
build a prototype application that demonstrates the
resource allocation and optimal utilization of cloud
resources using the notion of two users such as on-
demand user and HTC user.

International Journal of Computer Trends and Technology (IJCTT) volume 9 number 2 Mar 2014
ISSN: 2231-2803 http://www.ijcttjournal.org Page59

The remainder of the paper is structured as follows.
Section II provides review of literature available on
utilization of infrastructure clouds and relevant
topics. Section III provides details of the proposed
system. Section IV presents experimental results
while section V concludes the paper.

II. RELATED WORK
There is clear difference between the resource
utilization concept followed in large super computers
[7] and our work. In our work backfill virtual
machines are used to achieve utilization of resources
provisioning with respect to IaaS infrastructure. On
super computers, scheduling does not use any kind of
backfill jobs to be preempted when on-demand
requests arise. Instead of this algorithms try to use
backfill unused resources only in certain cases unlike
the procedures followed in this paper. There are
preemption based backfill solutions as explored in
[8]. These authors focused on supers computers that
make use of queue-based approach while our solution
uses virtual machines. BOINIC [9] which is a
volunteer computing system can make use of idle
cycles of distributed systems across Internet. Other
examples of such systems include Folding@Home
[10] and SET1@Home [11]. These applications are
best used to withstand interruptions in computers that
involve in distributed computing. Distributed
computing is the computing that takes place over
multiple servers. When a request is made by a client
that request is processed by multiple servers together.
These servers have program to program or machine
to machine interaction in order to serve the clients
request. There is much research on the failure
prediction, efficiency and usefulness of the volunteer
computing [12], [13], [114], and [15]. In this paper,
our work focuses on making the IaaS to optimize
cycles that are idle with respect to HTC. Volunteer
computing is best used and achieved through backfill
virtual machines.

Recovery techniques were focused in [26] that will
help in suspending and reusing virtual machines for
high utilization of infrastructure clouds. Condor
application is considered as an example for making
such experiments. Other areas that are related to our
work include spot pricing which has been explored
by Amazon [17]. Users are allowed to bid for virtual
machine instances and the based on the commercial
benefits, the VMs are allocated to users. In this paper
we use the VMs based on the current demand of the
on-demand user and the resource availability with
respect to backfill VMs.

Our work uses the current demand for on-demand
user VMs to determine the availability for backfill
VMs, whereas Amazon bases availability of spot
instances on a spot price. The prototype
implementations of infrastructure clouds used in this
work are Nimbus toolkit [3] and GitHub [18]. There
are some green computing techniques that optimize
the allocation of resources as explored in [19]. The
better cases it uses include prices, auction and have
built a model to utilize infrastructure clouds
optimally that bestows mutual benefits.

III. PROPOSED RESOURCE
UTILIZATION FRAMEWORK
The proposed resource utilization framework for
infrastructure clouds is based on the work done by
Marshall et al. [20]. In our approach we built a
prototype application to simulate the environment
presented in [20]. Our approach has two users. They
are on-demand user and HTC user. We consider two
types of leases for resource provisioning. They are
on-demand flexible leases and opportunistic pre-set
leases. In the former approach the resource allocation
is made on-demand. The users can have provision to
use any compatible virtual machine. Therefore it is
suitable for on-demand users. The latter is suitable
for HTC users who need more resources but non on-
demand nature. Backfill virtual machines are used to
process requests of HTC users but the on-demand
users requests are given priority thus utilizing
resources optimally.

Backfill Termination Policies

Backfill virtual machines are used to improve
utilization of infrastructure clouds. They are deployed
in idle VMM nodes in order to utilize them
optimally. The backfill VMs are usually deployed to
serve HTC users who have multiples jobs to be
carried out. When this is happening, on-demand users
might make request. In this case, the running backfill
virtual machines (one or more based on the
requirement) are terminated in order to allow the
resources to on-demand users. This is because on-
demand users should not be denied service and they
have to continue utilizing infrastructure clouds in pay
per use fashion.

Backfill Configuration Options

In the prototype we built we simulated the concept of
cloud and Condor application. For this reason we
built an administrative option where users can
configure the backfill features. The options to
configure backfill include Backfill.disabled to enable
International Journal of Computer Trends and Technology (IJCTT) volume 9 number 2 Mar 2014
ISSN: 2231-2803 http://www.ijcttjournal.org Page60

or disable backfill, Max.instances to specify
maximum number of instances, Disk.image to
specify path to VM, Memory.MB to specify the
amount of memory, VCPUs to specify number of
VCPUs, Duration.seconds to specify the duration of
backfill VM, Termination.policy to specify
termination approach, Retry.period to specify waiting
time while retry, and Network to specify private or
public network usage. More technical details about
the system can be found in [20].

IV. EXPERIMENTAL RESULTS
We built a prototype application to make experiments
on the resource utilization of infrastructure clouds in
Java platform. The environemtn used to built the
application is a PC with 4 GB RAM, core 2 dual
processor running Linux operating system. The
expreiments are made in terms of usage or non usage
of backfill virtual machines, with Condor simulation
to demonstrate the resource utlization with respect to
the users such as on-demand user and HTC user.


Figure 2 - Service response time for on-demand user
VM requests
As shown in the above Figure 2, the horizontal axis
represents Virtual Machine ID while vertical axis
represents the time in seconds. Less time is taken
when backfill is not enabled while more time is
utilized when backfill is enabled.


Figure 3 - Condor job queued time when the job first
begins executing, using the most recent backfill
termination policy (Condor simulation).
As shown in the figure 3 horizontal axis represents
Condor Job ID while vertical axis represents time in
seconds. When backfill is enabled less time is taken
to process Condor jobs while more time is consumed
when no user VMs are used for the same purpose.

V. CONCLUSIONS
In this paper we studied the utilization of
infrastructure clouds. With respect to cloud
computing technology, it is essential to have
infrastructure cloud that provides benefits to users of
cloud computing. The users will be able to utilize
infrastructure in pay per use fashion. As resources are
costly, the cloud computing has to optimize the
resource allocation. It has to make optimum use of
resources without having idle cycles. This is the
intention behind this work. In this paper we consider
Condor application and improved resource
utilization. We considered the notion of having two
users that make use of infrastructure clouds. They are
on-demand user who needs to be given priority and
the other user is known as HTC (High Throughput
Computing) users who need huge resources to
compute their jobs. We have built an approach that
satisfied on demand users and HTC users. The
HTC users are served by utilizing backfill VMs that
make use of available cycles effectively. When on-
demand users need more resources the HTC users
backfill VMs are terminated and the on-demand users
are served thus making the maximum utilization of
infrastructure clouds. We built a prototype
application that demonstrates the proof of concept.
The empirical results revealed that the proposed
approach is useful and able to optimize resource
utilization.
0
2
4
6
8
10
12
14
0 10 20 30 40 50 60 70
S
e
c
o
n
d
s
VM ID
Backfill
Enabled
No Backfill
0
200
400
600
800
1000
1200
1400
100 200 300 400 500 600
S
e
c
o
n
d
s
Condor Job ID
No User VMs
Backfill
Enabled
International Journal of Computer Trends and Technology (IJCTT) volume 9 number 2 Mar 2014
ISSN: 2231-2803 http://www.ijcttjournal.org Page61

REFERENCES
[1] Michael A, et al., Above the Clouds: A Berkeley View of
Cloud Computing, EECS Department, University of California,
Berkeley, Tech. Rep. UCB/EECS-2009-28, Feb 2009.
[2] Nurmi D, Wolski R, Grzegorczyk C, Obertelli G, Soman S,
Youseff L, and Zagorodnov D. Eucalyptus opensource cloud-
computing system. In CCA08: Cloud Computing and Its
Applications, 2008.
[3] Nimbus. [Online]. Retreived December 6, 2010, from:
http://www.nimbusproject.org/
[4] Woitaszek, M. and Tufo, H., Developing a cloud computing
charging model for high-performance computing resources, in
Computer and Information Technology (CIT), 2010 IEEE 10th
International Conference on, J uly 2010, pp. 210217.
[5] TannenbaumT, Wright D, Miller K, Livny M. Condor - A
Distributed J ob Scheduler, in Thomas Sterling, editor, Beowulf
Cluster Computing with Linux, The MIT Press, 2002.
[6] Smith, J E. and Nair, R. Virtual machines: versatile platforms
for systems and processes. Morgan Kaufmann Publishers, San
Francisco, CA, USA, 2005.
[7] Feitelson DG, Rudolph L. Parallel job scheduling: Issues and
approaches. Lecture Notes in Computer Science: J ob Scheduling
Strategies for Parallel Processing, 949, 1995.
[8] Snell Q, Clement M, and J ackson D. Preemption based backfill.
In Feitelson, Rudolph, and Schwiegelshohn, editors, J ob
Scheduling Strategies for Parallel Processing, pages 2437.
Springer Verlag, 2002.Lect. Notes Comput. Sci. vol. 2537.
[9] Anderson, D. BOINC: A System for Public-Resource
Computing and Storage, 5th IEEE/ACM Workshop on Grid
Computing, Nov. 2004.
[10] Larson SM, Snow CD, Shirts M, Pande VS. Folding@Home
and Genome@Home: Using distributed computing to tackle
previously intractable problems in computational biology.
Computational Genomics, Horizon Press, 2002.
[11] Anderson DP, Cobb J , Korpela E, Lebofsky M, Werthimer D.
SETI@home: An Experiment in Public-Resource
Computing.Communications of the ACM, 45(11), November
2002, 56-61.
[12] Acharya A, Edjlali G, and Saltz J . The Utility of Exploiting
Idle Workstations for Parallel Computation, SIGMETRICS 97,
pp. 225-34.
[13] Anderson D and Fedak G. The Computational and Storage
Potential of Volunteer Computing, CCGRID06, 2006, p. 73-80.
[14] Ren X, Lee S, Eigenmann R, and Bagchi S. Resource Failure
Prediction in Fine-Grained Cycle Sharing System, HPDC 06,
2006.
[15] Ryu K and Hollingsworth J . "Unobtrusiveness and Efficiency
in Idle Cycle Stealing for PC Grids," in Proceedings of IPDPS04,
2004, p. 62a.
[16] Sotomayor B, Keahey K, Foster I. Combining Batch
Execution and Leasing Using Virtual Machines. the 17th
International ACM Symposiumon High-Performance Parallel and
Distributed Computing (HPDC) Boston, MA. J une 2008.
[17] Amazon Web Services. Amazon.com, Inc. [Online]. Retreived
December 6, 2010, from: http://www.amazon.com/aws/
[18] Nimbus 2.6 Backfill (prototype). GitHub. [Online]. Retreived
December 6, 2010, from:
https://github.com/pdmars/nimbus/tree/backfill-2.6
[19] Lefe`vre, L. and Orgerie, AC. Designing and evaluating an
energy efficient cloud, The J ournal of Supercomputing, vol. 51,
pp. 352373,2010,
[20] Paul Marshall, Kate Keahey and TimFreeman, Improving
Utilization of Infrastructure Clouds, pp. 1-10.
Authors
Rajesh Kumar.Y, he is pursuing
M.Tech (CSE) in QIS College of Engineering &
Technology, Ongole, AP, INDIA. He has received
B.Tech Degree in Computer Science and
Engineering. His main research interest includes
Cloud Computing and Datamining.
K.Sirisha She is currently with the
Department of Computer Science and Engineering, QIS
College of Engineering & Technology, Andhra Pradesh,
India. Her main research interest includes Cloud
Computing and Data Mining.