GREEN CLOUD GREEN CLOUD

www.seminarsTopics.com 1
Abstract
Cloud computing is offering utility oriented IT services to users worldwide. It enables
hosting of applications from consumer, scientific and business domains. However data centers
hosting cloud computing applications consume huge amounts of energy, contributing to high
operational costs and carbon footprints to the environment. With energy shortages and global
climate change leading our concerns these days, the power consumption of data centers has
become a key issue. Therefore, we need green cloud computing solutions that can not only save
energy, but also reduce operational costs. The vision for energy efficient management of cloud
computing environments is presented here. green scheduling algorithm which works by
powering down servers when they are not in use is also presented.
Green Cloud computing is envisioned to achieve not only efficient processing and
utilisation of computing infrastructure, but also minimise energy consumption. This is essential for
ensuring that the future growth of Cloud computing is sustainable. !therwise, Cloud computing
with increasingly pervasive front"end client devices interacting with back"end data centres will
cause an enormous escalation of energy usage.
www.seminarsTopics.com #
Introduction
In 1$%$, &eonard 'leinrock , one of the chief scientists of the original dvanced (esearch
)ro*ects gency +etwork ,()+-T. which seeded the Internet, said/ As of now, computer
networks are still in their infancy, but as they grow up and become sophisticated, we will probably
see the spread of computer utilities which, like present electric and telephone utilities, will
service individual homes and offices across the country. This vision of computing utilities based
on a service provisioning model anticipated the massive transformation of the entire computing
industry in the #1st century whereby computing services will be readily available on demand, like
other utility services available in today0s society. 1imilarly, users ,consumers. need to pay
providers only when they access the computing services. In addition, consumers no longer need to
invest heavily or encounter difficulties in building and maintaining comple2 IT infrastructure.
In such a model, users access services based on their re3uirements without regard to where
the services are hosted. This model has been referred to as utility computing, or recently as Cloud
computing . The latter term denotes the infrastructure as a 4Cloud5 from which businesses and
users can access applications as services from anywhere in the world on demand. Hence, Cloud
computing can be classified as a new paradigm for the dynamic provisioning of computing
services supported by state"of"the"art data centers that usually employ 6irtual 7achine ,67.
technologies for consolidation and environment isolation purposes . 7any computing service
providers including 8oogle, 7icrosoft, 9ahoo, and I:7 are rapidly deploying data centers in
various locations around the world to deliver Cloud computing services.
Cloud computing delivers infrastructure, platform, and software ,applications. as services, which
are made available to consumers as subscription"based services under the pay"as"you"go model. In
industry these services are referred to as Infrastructure as a 1ervice ,Iaa1., )latform as a 1ervice
,)aa1., and 1oftware as a 1ervice ,1aa1. respectively. recent :erkeley report stated 4Cloud
Computing, the long"held dream of computing as a utility, has the potential to transform a large
part of the IT industry, making software even more attractive as a service5.
Clouds aim to drive the design of the ne2t generation data centers by architecting them as
networks of virtual services ,hardware, database, user"interface, application logic. so that users
can access and deploy applications from anywhere in the world on demand at competitive costs
depending on their ;o1 ,;uality of 1ervice. re3uirements .
www.seminarsTopics.com <
Need of Cloud Computing
The need of cloud computing can be e2plained with the help of an e2ample. The following graph
shows the number of users who log on to the ustralian !pen web page.
fig !monthly page views of australian open official website
The spikes correspond to the month of =anuary during which the tournament is going on. The site
remains almost dormant during the rest of the year. It would be wasteful to have servers which can
cater to the ma2imum need,as they wont be needed during the rest of the year. The concept of
cloud computing comes to the rescue at this time. >uring the peak period, cloud providers such as
8oogle,9ahoo,7icrosoft etc.can be approached to provide the necessary server capacity.
In this case, Infrastucture is provided as a service,Iaa1. through cloud computing.
&ikewise,cloud providers can be approached for obtaing software or platform as a service.
>evelopers with innovative ideas for new Internet services no longer re3uire large capital outlays
in hardware to deploy their service or human e2pense to operate it . Cloud computing offers
significant benefits to IT companies by freeing them from the low"level task of setting up basic
www.seminarsTopics.com ?
hardware and software infrastructures and thus enabling focus on innovation and creating business
value for their services.
www.seminarsTopics.com @
8reen Computing
8reen computing is defined as the study and practice of designing , manufacturing, using,
and disposing of computers, servers, and associated subsystemsAsuch as monitors, printers,
storage devices, and networking and communications systemsAefficiently and effectively with
minimal or no impact on the environment.B The goals of green computing are similar to green
chemistryC reduce the use of haDardous materials, ma2imiDe energy efficiency during the productEs
lifetime, and promote the recyclability or biodegradability of defunct products and factory waste.
(esearch continues into key areas such as making the use of computers as energy"efficient as
possible, and designing algorithms and systems for efficiency"related computer technologies.
There are several approaches to green computing,namely
)roduct longetivity
lgorithmic efficeincy
(esource allocation
6irtualisation
)ower management etc.
www.seminarsTopics.com %
+eed of green computing in clouds
7odern data centers, operating under the Cloud computing model are hosting a variety of
applications ranging from those that run for a few seconds ,e.g. serving re3uests of web
applications such as e"commerce and social networks portals with transient workloads. to those
that run for longer periods of time ,e.g. simulations or large data set processing. on shared
hardware platforms. The need to manage multiple applications in a data center creates the
challenge of on"demand resource provisioning and allocation in response to time"varying
workloads. +ormally, data center resources are statically allocated to applications, based on peak
load characteristics, in order to maintain isolation and provide performance guarantees. Fntil
recently, high performance has been the sole concern in data center deployments and this demand
has been fulfilled without paying much attention to energy consumption. The average data center
consumes as much energy as #@,GGG households H#GI. s energy costs are increasing while
availability dwindles, there is a need to shift focus from optimising data center resource
management for pure performance to optimising for energy efficiency while maintaining high
service level performance. ccording to certain reports,the total estimated energy bill for data
centers in #G1G is J11.@ billion and energy costs in a typical data center double every five years.
>ata centers are not only e2pensive to maintain, but also unfriendly to the environment. >ata
centers now drive more in carbon emissions than both rgentina and the +etherlands . High
energy costs and huge carbon footprints are incurred due to massive amounts of electricity needed
to power and cool numerous servers hosted in these data centers. Cloud service providers need to
adopt measures to ensure that their profit margin is not dramatically reduced due to high energy
costs. Kor instance, 8oogle, 7icrosoft, and 9ahoo are building large data centers in barren desert
land surrounding the Columbia (iver, F1 to e2ploit cheap and reliable hydroelectric power .
There is also increasing pressure from 8overnments worldwide to reduce carbon footprints, which
have a significant impact on climate change. Kor e2ample, the =apanese government has
established the =apan >ata Center Council to address the soaring energy consumption of data
centers . &eading computing service providers have also recently formed a global consortium
known as The 8reen 8rid to promote energy efficiency for data centers and minimise their
environmental impact.
&owering the energy usage of data centers is a challenging and comple2 issue because computing
applications and data are growing so 3uickly that increasingly larger servers and disks are needed
www.seminarsTopics.com L
to process them fast enough within the re3uired time period. Green Cloud computing is
envisioned to achieve not only efficient processing and utilisation of computing infrastructure, but
also minimise energy consumption. This is essential for ensuring that the future growth of Cloud
computing is sustainable. !therwise, Cloud computing with increasingly pervasive front"end
client devices interacting with back"end data centers will cause an enormous escalation of energy
usage.
To address this problem, data center resources need to be managed in an energy"efficient manner
to drive 8reen Cloud computing. In particular, Cloud resources need to be allocated not only to
satisfy ;o1 re3uirements specified by users via 1ervice &evel greements ,1&., but also to
reduce energy usage.
rchitecture of a green cloud computing platform
www.seminarsTopics.com M
"ig #! architecture of a green cloud computing environment
Kigure # shows the high"level architecture for supporting energy"efficient service
allocation in 8reen Cloud computing infrastructure. There are basically four main entities
involved/
a) Consumers/Brokers: Cloud consumers or their brokers submit service re3uests from anywhere
in the world to the Cloud. It is important to notice that there can be a difference between Cloud
consumers and users of deployed services. Kor instance, a consumer can be a company deploying
a Web application, which presents varying workload according to the number of users accesing
it.
b) Green Resource Allocator: cts as the interface between the Cloud infrastructure and
consumers. It re3uires the interaction of the following components to support energy"efficient
resource management/
N $reen %egotiator/ +egotiates with the consumersObrokers to finaliDe the 1& with specified
prices and penalties ,for violations of 1&. between the Cloud provider and consumer depending
www.seminarsTopics.com $
on the consumer0s ;o1 re3uirements and energy saving schemes. In case of Web applications, for
instance, ;o1 metric can be $@P of re3uests being served in less than < seconds.
N &ervice Analyser/ Interprets and analyses the service re3uirements of a submitted re3uest before
deciding whether to accept or re*ect it. Hence, it needs the latest load and energy information from
67 7anager and -nergy 7onitor respectively.
N Consumer 'rofiler/ 8athers specific characteristics of consumers so that important consumers
can be granted special privileges and prioritised over other consumers.
N 'ricing/ >ecides how service re3uests are charged to manage the supply and demand of
computing resources and facilitate in prioritising service allocations effectively.
N (nergy )onitor/ !bserves and determines which physical machines to power onOoff.
N &ervice &cheduler/ ssigns re3uests to 67s and determines resource entitlements for allocated
67s. It also decides when 67s are to be added or removed to meet demand.
N *) )anager/ 'eeps track of the availability of 67s and their resource entitlements. It is also
in charge of migrating 67s across physical machines.
N Accounting/ 7aintains the actual usage of resources by re3uests to compute usage costs.
Historical usage information can also be used to improve service allocation decisions.
c. VMs: 7ultiple 67s can be dynamically started and stopped on a single physical machine to
meet accepted re3uests, hence providing ma2imum fle2ibility to configure various partitions of
resources on the same physical machine to different specific re3uirements of service re3uests.
7ultiple 67s can also concurrently run applications based on different operating system
environments on a single physical machine. In addition, by dynamically migrating 67s across
physical machines, workloads can be consolidated and unused resources can be put on a low"
power state, turned off or configured to operate at low"performance levels ,e.g., using >6K1. in
order to save energy.
d) Physical Machines: The underlying physical computing servers provide hardware
infrastructure for creating virtualised resources to meet service demands.
www.seminarsTopics.com 1G
7aking cloud computing more green
7ainly three approaches have been tried out to make cloud computing environments more
environmental friendly. These approaches have been tried out in the data centres under
e2perimental conditions. The practical application of these methods are still under study. The
methods are/
Dynamic Voltage frequency scaling technique(DV!)/" -very electronic circutory will
have an operating clock associated with it. The operatin fre3uency of this clock is ad*usted
so that the supply voltage is regulated. Thus, this method heavily depends on the hardware
and is not controllabale according to the varying needs. The power savings are also low
compared to other approaches. The power savings to cost incurred ratio is also low.
Resource allocation or "irtual machine migration techniques/" In a cloud computing
environment,every physical machine hosts a number of virtual machines upon which the
applications are run. These virtual machines can be transfered across the hosts according to
the varying needs and avaialble resources.The 67 migration method focusses on
transferring 67s in such a way that the power increase is least. The most power efficient
nodes are selected and the 67s are transfered across to them. This method is dealt in detail
later.
Algorithmic a##roaches/" It has been e2perimently determined that an ideal server
consumes about LGP of the power utilised by a fully utilised server. ,1ee figure <..
"ig +! 'ower consumption under different work loads.
www.seminarsTopics.com 11
Fsing a neural network predictor,the green scheduling algorithms first estimates re3uired
dynamic workload on the servers. Then unnecessary servers are turned off in order to minimiDe
the number of running servers, thus minimiDing the energy use at the points of consumption to
provide benefits to all other levels. lso,several servers are added to help assure service"level
agreement. The bottom line is to protect the environment and to reduce the total cost of ownership
while ensuring 3uality of service.
www.seminarsTopics.com 1#
67 7igration
The problem of 67 allocation can be divided in two/ the first part is admission of new
re3uests for 67 provisioning and placing the 67s on hosts, whereas the second part is
optimiDation of current allocation of 67s.
!ptimiDation of current allocation of 67s is carried out in two steps/ at the first step we
select 67s that need to be migrated, at the second step chosen 67s are placed on hosts using
7:K> algorithm. We propose four heuristics for choosing 67s to migrate. The first heuristic,
ingle !"res"old #!), is based on the idea of setting upper utiliDation threshold for hosts and
placing 67s while keeping the total utiliDation of C)F below this threshold. The aim is to
preserve free resources to prevent 1& violation due to consolidation in cases when utiliDation by
67s increases. t each time frame all 67s are reallocated using 7:K> algorithm with additional
condition of keeping the upper utiliDation threshold not violated. The new placement is achieved
by live migration of 67s .
The other three heuristics are based on the idea of setting upper and lower utiliDation
thresholds for hosts and keeping total utiliDation of C)F by all 67s between these thresholds. If
the utiliDation of C)F for a host goes below the lower threshold, all 67s have to be migrated
from this host and the host has to be switched off in order to eliminate the idle power
consumption. If the utiliDation goes over the upper threshold, some 67s have to be migrated from
the host to reduce utiliDation in order to prevent potential 1& violation. We propose three policies
for choosing 67s that have to be migrated from the host.
N 7inimiDation of 7igrations ,77. Q migrating the least number of 67s to minimise
migration overhead.
N Highest )otential 8rowth ,H)8. Q migrating 67s that have the lowest usage of C)F
relatively to the re3uested in order to minimise total potential increase of the utiliDation and 1&
violation.
N (andom Choice ,(C. Q choosing the necessary number of 67s by picking them
according to a uniformly distributed random variable.
$%#erimental !etu#
www.seminarsTopics.com 1<
s the targeted system is a generic Cloud computing environment, it is essential to evaluate it on a
large"scale virtualised data center infrastructure. However, it is difficult to conduct large"scale
e2periments on a real infrastructure, especially when it is necessary to repeat the e2periment with
the same conditions ,e.g. when comparing different algorithms.. Therefore, simulations have been
chosen as a way to evaluate the proposed heuristics. The Cloud1im toolkit has been chosen as a
simulation platform as it is a modern simulation framework aimed at Cloud computing
environments. In contrast to alternative simulation toolkits ,e.g. 1im8rid, 8and1im., it supports
modeling of on"demand virtualiDation enabled resource and application management. It has been
e2tended in order to enable power"aware simulations as the core framework does not provide this
capability. part from the power consumption modeling and accounting, the ability to simulate
service applications with variable over time workload has been incorporated.
There are a few assumptions that have been made to simplify the model of the system and enable
simulation"driven evaluation. The first assumption is that the overhead of 67 migration is
considered as negligible. 7odeling the cost of migration of 67s is another research problem and
is being currently investigated . However, it has been shown that application of live migration of
67s can provide reasonable performance overhead. 7oreover, with advancements of
virtualiDation technologies, the efficiency of 67 migration is going to be improved. nother
assumption is that due to unknown types of applications running on 67s, it is not possible to
build the e2act model of such a mi2ed workload . Therefore, rather than simulating particular
applications, the utiliDation of C)F by a 67 is generated as a uniformly distributed random
variable. In the simulations we have defined that 1& violation occurs when a 67 cannot get
amount of 7I)1 that are re3uested. This can happen in cases when 67s sharing the same host
re3uire higher C)F performance that cannot be provided due to consolidation. To compare
efficiency of the algorithms we use a characteristic called 1& violation percentage, or simply
1& violation, which is defined as a percentage of 1& violation events relatively to the total
number of measurements.
data center that comprises 1GG heterogeneous physical nodes was simulated. -ach node
is modeled to have one C)F core with performance e3uivalent to 1GGG, #GGG or <GGG 7illion
Instructions )er 1econd ,7I)1., M 8b of (7 and 1 T: of storage. ccording to this model, a
host consumes from 1L@ W with GP C)F utiliDation and up to #@G W with 1GGP C)F utiliDation.
-ach 67 re3uires one C)F core with #@G, @GG, L@G or 1GGG 7I)1, 1#M 7: of (7 and 1 8: of
storage. The users submit re3uests for provisioning of #$G heterogeneous 67s that fills the full
www.seminarsTopics.com 1?
capacity of the simulated data center. -ach 67 runs a web"application or any kind of application
with variable workload, which is modeled to create the utiliDation of C)F according to a
uniformly distributed random variable. The application runs for 1@G,GGG 7I)1 that e3uals to 1G
minutes of e2ecution on #@G 7I)1 C)F with 1GGP utiliDation. Initially, 67s are allocated
according to the re3uested characteristics assuming 1GGP utiliDation. -ach e2periment has been
run 1G times and the presented results are built upon the mean values.
!imulation Results
Kor the benchmark e2perimental results we have used a +on )ower ware ,+). policy. This
policy does not apply any power aware optimiDations and implies that all hosts run at 1GGP C)F
utiliDation and consume ma2imum power. The second policy applies >6K1, but does not perform
any adaptation of allocation of 67s in run"time. Kor the simulation setup described above, using
the +) policy leads to the total energy consumption of $.1@ 'Wh, whereas >6K1 allows
decreasing this value to ?.? 'Wh.
The simulation resuts of various policies are e2plained in the ne2t sections.
www.seminarsTopics.com 1@
-nergy Consumption and 1& vioaltion of 1T policy
ig &'simulation results of !( #olicy
To evaluate 1T policy we conducted several e2periments with different values of the
utiliDation threshold. The simulation results are presented in Kigure ?. The results show that energy
consumption can be significantly reduced relatively to +) and >6K1 policies Q by LLP and @<P
respectively with @.?P of 1& violations. They show that with the growth of the utiliDation
threshold energy consumption decreases, whereas percentage of 1& violations increases. This is
due to the fact that higher utiliDation threshold allows more aggressive consolidation of 67s,
however, by the cost of the increased risk of 1& violations.
-nergy consumption and 1& violations of other policies
We have compared 77 policy with H)8 and (C policies varying e2act values of the
thresholds but preserving ?GP interval between them. The results ,Kigures @ R % . show that these
policies allow the achievement of appro2imately the same values of energy consumption and 1&
violations. Whereas the number of 67 migrations produced by 77 policy is reduced in
www.seminarsTopics.com 1%
comparison to H)8 policy by ma2imum of @LP and ?GP on average and in comparison to (C
policy by ma2imum of ?$P and #LP on average.
ig )'energy consum#tion of *ifferent #olicies
ig+'!,A "iolations of *ifferent #olicies un*er *ifferent threshol*s
www.seminarsTopics.com 1L
Com#arison -ith res#ect to CP. utilisation

ig /'com#arison of *iff #olicies un*er *ifferent -orkloa*s -ith res#ect to c#u utilsation

s can be seen,the +on )ower ware ,+). policy which is currently being followed
results in higher power loss,even though it doesnt result in 1& violations. The optimal case
occurs when 77 policy under a workload of <G"LGPis used.
Conclusion
pplying green technologies is highly essential for the sustainable development of cloud
computing. !f the various green methodologies en3uired, the >6K1 technology is a highly
hardware oriented approach and hennce less fle2ible. The reuslt of various 67 migration
simulations show that 77 policy leads to the best energy savings/ by M<P, %%P and #<P less
energy consumption relatively to +), >6K1 and 1T policies respectively with thresholds <G"
LGP and ensuring percentage of 1& violations of 1.1PC and by MLP, L?P and ?<P with
thresholds @G"$GP and %.LP of 1& violations. 77 policy leads to more than 1G times less 67
migrations than 1T policy. The results show fle2ibility of the algorithm, as the thresholds can be
ad*usted according to 1& re3uirements. 1trict 1& ,1.11P. allow the achievement of the energy
www.seminarsTopics.com 1M
consumption of 1.?M 'Wh. However, if 1& are rela2ed ,%.%$P., the energy consumption is
further reduced to 1.1? 'Wh. 1ingle threshold policies can save power upto #GP,but they also
cause a large number of 1& violations. 8reen scheduling algorithms based on neural predictors
can lead to a LGP power savings. These policies also enable us to cut down data centre energy
costs, thus leading to a strong,competitive cloud computing industry. -nd users will also benefit
from the decreased energy bills.

www.seminarsTopics.com 1$
(eferences
1. 4-nergy efficient management of data centre resources for cloud computing/
vision,architectural elements and !pen Challenges5 (a*kumar :uyya,nton
:eloglaDov,=emal bawa*y 'roc. of ,th -((( -nternational &ymposium on Cluster
Computing and the $rid .CC$rid #//,0, (io >e =aneiro, :raDil, 7ay #GG$.
#. 4)erformance -valuation of a green 1cheduling algorithm for energy savings in cloud
computing5 Troung 6inh Troung >uy,9ukinori 1ato,9ashushi Inoguchi -((( 1plore,
)arch #//
<. www.wikipedia.comOgreencomputing.
?. www.ibm.comOdeveloperworksOwebsphereODonesOhipods
www.seminarsTopics.com #G