You are on page 1of 5

Dynamic Resource Allocation Using

Virtualized Smartphones

ANAND MAHALAHA
VIT UNIVERSITY
anand.mahalaha@yahoo.com
BHUPESH DHAKAD
VIT UNIVERSITY
bhupeshdhakad@gmail.com
RAMA KRISHNA RAJU
VIT UNIVERSITY
ramakrishna.v90@gmail.com



Abstract In Todays life cloud
computing provides a flexible usage of
resources to all the users. Many
advantage comes from the resource
multiplexing technique. In this paper we
are going to present a system that uses the
virtualization technology to assign the
resources dynamically on the basis of the
requirements of the different
applications. This also helps to improve
green computing by the utilization the
resources in multidimensional way.
In this paper we also explains the
algorithm which measures the term,
coined as skewness. The value of
skewness measures the efficient
utilization of the dynamic resources. By
minimizing the skewness value we can
manage the load balancing of resources
among the different applications.

Keywords Virtual Machine, Hypervisor,
Cloud Computing, Skewness, Green
Computing, Resource Management, Load
Balancing


I. INTRODUCTION
Most Cloud platforms are facing various
problems with balancing resources on
physical hardware, not utilizing existing
Datacenters when the server cant avoid
overload for peak stage. Virtualization is a
technology, giving effective utilizing of load
balancing and live Migration for virtualized
smart phones in cloud computing. Most of
the applications can reduce the costs by
reallocating resources, when load balancing
occurs in cloud computing. VM screens
(Xen) gives mapping virtual resources with
physical machines (PMs) running on mobile
applications, the information is hidden to
clients.
Cloud supplier has to check existing
physical machines have enough resources
based on requirements. The current
challenges in datacenters are load balancing
and migration, if more number of requests
has given to one region, it will effect on the
whole system. The work exceeds the limit
then hot spot are created because of
overloading of requests, while cold spot may
create, if server not utilizing the resources so
we can migrate the resources, where they are
required in order to balance the work load.
Here we discuss the two techniques cold
and hot spot, cold spot shows more
availability of resources, collects from
physical nodes. To discover the cold spots in
system is invoking periodically, checking to
the requests and ranking to the nodes. Hot
spot occurs when sever reaches above the
point, it shows the resources are overloaded
in server, we have to migrate the virtual
machines.
The resources allocated to virtual
machines, which are run on hardware having
unbalance of load will affect the server at
cold and hot spot, based on prediction,
inactiveness of virtual machine and overload
of resources, we can migrate the virtual
Machines (VMs).


Fig. 1 Virtual Machine Monitor


2. SYSTEM OVERVIEW
The architecture of the whole system is
shown in figure 2. Each Physical Machine
(PM) runs on the Xen hypervisor [4], which
supports two domain i.e. zero and one or
more domain U. We are assuming that all
the Physical Machines (PMs) shares the back
end storage. The Usher framework [3]
manages the multiplexing of the VMs to
PMs [1].
The main concentration of our cloud
system is implemented as a set of plugins to
usher. Each of them runs on the usher local
node manager on the domain zero [1], which
stores the usage statistics of the resource for
each and every Virtual Machine.


Fig. 2 System Architecture
The statistics collected of each and every
Physical Machines are forwarded to the
Usher central controller. At regular interval
of time the Virtual Machine scheduler is
invoked and then receives from the local
node machine.
There is a prediction for the future
resource demands of the Virtual machines
and the future load of the Physical Machines
based on the past statistics [5].

3.THE SKEWNESS ALGORITHM
Here we introduce the term called
skewness which calculates the value of
unevenness of the utilization of resources at
server [1] [2]. Let n be the total number of
resources and r
t
be the value of utilization of
the i
th
resources. Now the skewness factor of
the server p can be given as:
Skewness (p) =

where r is nothing but the average utilization
of all the resources of the server p. Now
according to our algorithm only few
resources are there that need to be
considered for the calculations.
We have to minimize the value of the
skewness factor to increase the performance
of the server [1] [2]. After the detection of
the value of skewness factor now we have to
determine the hot threshold and cold spots
for better performance of the server. The
concept of hot threshold and cold spot is
explained in details in the later section.

4. HOT AND COLD POINT
The algorithm which we introduce
executes at regular interval of time and
checks the resource allocation status as
demands by the Virtual Machines running on
a virtualizes smartphones. In this paper we
also define a term, hot spot and cold spot in
context to server. We define server as a hot
spot if any resource skewnesss the server
beyond a certain point.
At this time the mobile devices shows the
critical message to the users and the virtual
machines which are running currently should
be migrated away. We are calculating the
temperature of a hot spot p which is
calculated as:

Temperature(p) = ( )

,

where R is a set of all the overloaded
resources in the server p and rt is the hot
cutoff point for the resource r. The one thing
we have to remember here that only
overloaded resources are to be considered for
the calculation. The temperature gives the
degree of overload [2]. If the value of
temperature is zero then the server is not a
hot spot.
We have also define a term cold spot if
the utilization of the resources at server are
below the certain point. At this stage the
server is seems to be running in idle state.
Finally we also define a term warm spot,
which tells that the resource utilization is
sufficiently high so that to maximize the
efficiency and minimize the risk factor of
becoming hot spot.
We can define thresholds points for
different resources as per their capabilities.
For example, we can define a hot spot for
CPU to be 90 percentage and that of memory
resources to be 80 percentage. Therefore we
can conclude that a server is gone beyond
hot spot if either of above two cross their
limits.

5. HOT SPOT MIGRATION
To overcome the problem of hot spot we
have prepared a list of hot spots in a system
in their descending temperature, i.e. the
hottest spot will be the first one. Our mission
is to bring down all the hot spot of the
system. To do so first we have to decide that
which virtual machine of that particular
server should migrate away.
Here we again prepared a list of resultant
temperature of server after migrating a
particular virtual machine. Our aim is to
migrate that virtual machine which decreases
the temperature of server as much as
possible. If there is tie between two virtual
machines then we go for their other
consequences. The one which leads to
greater stability of the server will migrate
first.
For every virtual machine which is being
migrated, we have to find out appropriate
destination server so that after migration of
that virtual machine the server will not
become hot spot. Among all the other servers
we have to select that server which will not
affected much more after the migration of
virtual machine.
Now if there is no such destination server
found then we keep track of that server and
update the load of that server. Otherwise we
have to move to the next virtual machine in
the list and try to find out appropriate
destination for that. We have to run our
algorithm as long as the virtual machine
finds their proper destination.
After the successful completion of
algorithm, we have to move to next hot spot.
Here we have to note that at each run of our
algorithm at most one virtual machine should
migrate from the overloaded server.
This algorithm does not guarantee the
elimination of the hot spot from the
overloaded server but can reduce the
temperature of that server. If by chance, the
hot spot remains as it is then the algorithm
executes again to eliminate it.

6. LOAD BALANCING
Several operating systems are running on
sever of physical machine and several
servers run on Data centers through network
with fast connectivity. Some servers are
overloaded and remaining may not skewness
properly. Our aim is to distribute resources
to carry out the load using various
algorithms.
In order to reach the goal, we introduce the
skewness algorithm to balance the
unevenness of various resources and to
migrate virtual machines from one system to
other, so hot spot decreases the intensive
workloads and maintains certain temperature
without any workloads to servers.
The algorithm predicts the load, it will
take necessary steps to avoid workload and
enhances utilization of server. Load balancer
detects the overloaded servers and migrate
the resources of virtual machines to
destination server, but it shouldnt be
overloaded and based on algorithm we can
migrate the virtual machines to available
servers.

7. PREDICTING FUTURE NEEDS
We should predict resources of virtual
machines for future purpose. Altering the
virtual machines for mobile application,
sometimes its not possible. We have to
predict physical nodes and past character of
tables of virtual machines. Cloud clients can
scale up and scale down the needs of
resources in cloud computing. We can avoid
unevenness of load of physical machine and
reallocate the resources. To predict the
resources for future needs using formula:

E(t) = * E( t- 1 ) + ( 1 - ) * O(t) ,
0 1+

where, E (t) is estimated time and O (t)
Observed time for load. is a constant for
utilizing stability and physical
responsiveness of resources.
This equation is used to predicting
processor load on domain name server, for
each minute , it will show utilizing resources
for current minute and predicting work load
for next successive minute. In the diagram, if
constant value is 0.7 and we represented as
dot and curve, which are observed and
predicted values from the formula. In order
to get exact prediction, the curve goes
through dots at certain domain.


We can predict the value but sometimes it
doesnt match with climbing trends. For
instance, take observed time values O (t) as
40, 50, 60 and 70, we can predict 80 as next
element. Sometimes the value (0, 1) the
predicted value should be in between past
and observed value. We have to set from
negative value and W is size of window.

8. CONCLUSION
In these day we are living the digital
world and this world the smartphone emerge
as a powerful tool in our everyday life. In
this paper we discussed about how we
provide the avaibility, accessibility and
utility of resources in an efficient manner
using the smartphones. This paper is based
on how to distribute the resource to the user
in efficient manner to attain the maximum
performance.
We also take care about the green
computing technology hot spot and cold
spot. In this paper we successfully describe
how to implement the dynamic resource
allocation algorithm in smartphone in
virtualize manner and application migration
technique in the smartphone.

9. REFERENCES

[1] Zhen Xiao, Weijia Song, Qi Chen, Dynamic
Resource Allocation Using Virtual Machine for Cloud
Computing Environment, IEEE Transactions on
Parallel and Distributed Systems, Vol.24, 2013.

[2] T. Kamalakar Raju, A. Lavanya, Dr. M.
Rajanikanth, Virtualization Technology using Virtual
Machines for Cloud Computing, International
Journal of Modern Engineering Research, Vol.4,
2014.

[3] M. McNett, D. Gupta, A. Vahdat, G. M. Voelker,
Usher: An Extensible Framework for Managing
Clusters of Virtual Machines, Large Installation
System Administration Conference, Nov. 2007.

[4] P. barham, B. Dragovic, K. Fraser, S. hand, T.
harris, A. Ho, R. neugebauer, I. Pratt, A. Warfield,
Xen and the Art of Virtualization, Operating
Systems Principles, Oct. 2003.

[5] T. Wood, P. Shenoy, A. Venkataramani, M.
Yousif, Black-Boxand gray-Box Strategies for
Virtual Machine Migration, Networked Systems
Designe and Implementation, Aug. 2002.

You might also like