DYNAMIC ENERGY MANAGEMENT IN CLOUD DATA CENTERS: A SURVEY

International Journal on Cloud Computing: Services and Architecture (IJCCSA) ,Vol.3, No.4, August 2013
DOI : 10.5121/ijccsa.2013.3402 13
DYNAMIC ENERGY MANAGEMENT IN CLOUD DATA
CENTERS: A SURVEY
T.Veni1
and S. Mary Saira Bhanu2
Department of Computer Science and Engineering
National Institute of Technology Tiruchirappalli
Tiruchirappalli-620015, India
{406111001, msb}@nitt.edu
ABSTRACT
Cloud data centers have become indispensable infrastructure for computing and data storage that facilitate
the development of diversified services offered by the cloud. These data centers consume enormous
amounts of electrical energy to process the cloud services resulting in large amount of CO2 emissions, high
operational cost, and affecting the lifetime of hardware equipments. This necessitates the development of
efficient energy management techniques in cloud data centers for both economic and environmental
standpoints. Energy management at data centers can be static or dynamic. Energy consumption can be
reduced by employing dynamic energy management techniques both in hardware and software levels at
cloud data centers. This paper surveys various issues related to dynamic energy management in cloud data
centers.
KEYWORDS
Cloud Data Center, Virtualization, Dynamic Energy Management, & Server Consolidation
1. INTRODUCTION
Data Centers have emerged as a back-bone infrastructure, housing large number of IT equipments
such as servers, data storage, network devices, power and cooling devices etc. that facilitate the
development of wide variety of services offered by the cloud [1] Currently, several service
providers such as Amazon, Google, Yahoo, Microsoft, IBM and Sun, have their own data centers
to provide the scalable services to a large customer base [2-3].With the rapid development of IT
industry and increasing demand for cloud services, the number of data centers have increased.
These data centers consume enormous amount of energy to process its services resulting in
increased energy consumption. The surging energy consumption of these data centers has become
a serious concern from both economic and environmental standpoints.
According to McKinsey report [4], the energy consumption of data centers is $11.5 billion in
2010 and it doubles every five years. Gartner [5] also estimated that world wide IT infrastructures
are responsible for 2% of global CO2 emissions and energy related costs account for the 12% of
the total economical expenditures. The excessive energy consumption at data centers leads to
high operational cost, large amount of CO2 emission and falling lifetime of hardware equipments.

14
Hence, it is necessary to design energy-efficient data centers not only for ensuring system
reliability but also reducing environmental impact and operational cost.
Energy management techniques at the data centers can be static or dynamic. The static energy
management techniques fail to address the run time adaptation of data centers in response to
workload changes. The dynamic energy management techniques configure the data center at both
hardware and software levels dynamically based on workload variability.
Further, the energy conservation can be achieved by efficient utilization of data center resources.
Virtualization technology is one such powerful technology to address this energy inefficiency by
increasing resource utilization [6].This technology allows multiple virtual machines (VMs) to
share the resources on a single physical machine (PM). The features such as VM isolation and
VM migration along with dynamic resource provisioning can be used either to consolidate virtual
machines on fewer physical servers or to balance the load across physical servers in data centers,
thereby ensuring applications’ performance. This paper surveys the various techniques and issues
related to dynamic energy management in cloud data centers.
The rest of this paper is organized as follows. Section 2 explains taxonomy of energy
management mechanisms in virtualized data centers; section 3 presents an overview of dynamic
energy management techniques in cloud data centers and finally, section 4 presents the
conclusion.
2. TAXONOMY OF ENERGY MANAGEMENT TECHNIQUES IN CLOUD DATA
CENTERS
It is necessary to understand the background terminologies involved in the context of energy
management techniques. The energy consumption (E) at cloud data center is deﬁned as a total
amount of power (P) consumed over a period of time (T) while performing the work [7], i.e.,
E = P * T (1)
Thus, the energy conservation at the cloud data centers can be achieved by controlling average
power consumption (P) over the period of time. As per taxonomy [7], the energy management
approaches in cloud data center can be classified into static and dynamic energy management
techniques as shown in Fig.1.

15
Fig .1. Taxonomy of Energy Management Techniques in Cloud Data Centers
Static Energy Management (SEM) technique [8] uses low-power hardware components at the
data centers for energy savings. The request for the cloud services running in the data center is
volatile in nature. This technique fails to address the runtime adaptation of data centers in
response to the service demand to avoid resource wastage and it is also a very expensive
technique for energy conservation.
3. OVERVIEW OF DYNAMIC ENERGY MANAGEMENT TECHNIQUES
Dynamic energy management (DEM) techniques dynamically reconfigure the system based on
current resource requirements of the requested services. It utilizes the power scalable hardware
components and software approaches to optimize the energy consumption at data centers.DEM
techniques are classified into hardware and virtualization assisted techniques based on the level of
their applicability.
3.1. Hardware Level Energy Management Techniques
The energy management techniques need to be applied on each hardware components such as the
processors, network equipments, and storage devices to reduce overall hardware level energy
consumption at cloud data centers.
3.1.1. Processor Level
Energy consumption of a processor consists of two components such as static power and dynamic
power depends on the usage of various resources at server [7]. The dynamic power consumption
of a CMOS-based processor is given by
Dynamic Power = A ∗ F ∗ C ∗ V2
(2)
Where A is the percentage of active gates, F is clock frequency of the processor, C is capacitance
load of the processor, and V is the voltage supplied. The techniques such as Dynamic Voltage and
Frequency Scaling (DVFS), Network Level Clock Gating and Supply Shutdown are used for
power management in processor level.

16
3.1.2. Network Level
As the size of the data center and network infrastructure is exploding, it is necessary to address
the energy consumed by the network devices such as routers, switches, NIC, etc. Switches form
the basis of interconnection fabric that delivers job requests to the computing servers for
execution. The energy consumption of a switch depends on the type of switch, number of ports,
port transmission rates and employed caching solutions. The energy consumed by a switch can be
generated by the following equation;-
P Switch = P Chassis + N Line cards * P Line card + ∑R
i=0 N Ports * Pr (3)
Where P Chassis is the power consumed by the switch base hardware, P Line card is the power
consumed by the active line card and Pr is the power consumed by the active ports. The
techniques such as Link State Adaptation (LSA), Idle Elements Shutdown are used for power
savings in network level.
3.1.3. Storage Level
The storage devices constitute a significant fraction of the overall energy budget. The storage
level energy management techniques [38] can be divided as follows:-
 Hardware based techniques increase the disk power conservation by maintaining
storage hierarchy to strike the right balance between performance and power consumed
by storage resources
 Disk management techniques introduce the new disk management layer on top of the
file system, which controls disk configuration and data layout to achieve power optimal
disk access patterns
 Caching techniques reduce the power consumption at storage level by allowing large
fractions of the storage system to remain idle for longer periods of time and switch to
lower power modes

17
3.1.4. Summary of Hardware Level Energy Management Techniques
System
Resources
Considered
References Techniques used Description
Processor
Level
Benini et al
[9],
Beloglazov
et al[7]
Dynamic Voltage
and Frequency
Scaling (DVFS)
In DVFS, the voltage and frequency of
the processor can be scaled
dynamically depending on the request
that is being processed.
Benini et al
[9]
Network Clock
Level Gating
In Network Level Clock Gating, the
supplied voltage or the clock frequency
is reduced for idle components.
Benini et al
[9]
Supply Shutdown In Supply Shutdown, the idle
components are powered off to
decrease power dissipation as there is a
leakage of current even if all the clocks
are halted.
Network
Level
Nordmen et
al[12],
Link State
Adaptation
(LSA)
In LSA, the link speed of network can
be changed dynamically to reduce
energy consumption in networks.
M. Gupta et
al [35]
Switch
Shutdown
In Switch Shutdown, the Ethernet
switches are shutdown dynamically
based on the traffic arrivals, buffer
capacity and bounded maximum
packet delay for energy conservation.
Heller et al
[36]
LSA + Idle
elements
shutdown
This hybrid approach uses both LSA
and Switch shutdown methods to
conserve energy on network level.
Storage
Level
Gurumathi et
al [10]
Dynamic
Rotations Per-
Minute(DRPM)
In DRPM, the speed of disk rotation
can be changed dynamically per
minute (RPM) for controlling the
power explored in the spindle motor
driving the platters.
Colarelli et al
[11]
Massive Array of
Idle Disks
(MAID)
In MAID, only those disk drives in
active use are spinning at any given
time. MAID reduces power
consumption and prolongs the lives
of the drives.
Bianchini et
al[37]
Popular Data
Concentration
(PDC)
This method selects the most popular
files based on frequency of disk
access and least popular disks are
shutdown for energy conservation.
Table.1. Summary of Hardware Level Energy Management Techniques

18
For efficient energy conservation, the hardware energy management techniques should consider
all resources such as Processor, Network and Storage etc. The above mentioned hardware
techniques are applicable at individual server level for energy conservation. Hence, these
hardware level energy management techniques do not contribute much in total energy
conservation at cloud data centers. The energy consumption can be further reduced by applying
energy management techniques at software (virtualization level) at cloud data centers.
3.2. Virtualization Level Energy Management Techniques
The energy consumption problem has been addressed partially by making improvements in the
physical infrastructure of modern data centers. According to Open Compute project report [33],
93% of the data center’s energy consumption depends upon efficient utilization of computing
resources at data centers. Virtualization is a key technology that facilitates the better use of
available data center resources using the technique called server consolidation [13]. This
technique involves consolidating multiple physical server workloads into single physical server to
increase the resource utilization. Thus, it allows for reduction in the total number of physical
server used, minimizes the server sprawl as well as total data center space requirements. The
server consolidation can be performed either statistically or dynamically.
 In static consolidation, VMS are placed on physical servers for a long time period and
not migrated even if workload changes.
 In dynamic consolidation, VMs are placed on physical servers at runtime and migrations
of VMs performed automatically in response to the current workload demands. This
helps in utilizing the data centers resources efficiently.
The steps for server consolidation (Fig.2.) are as follows:
 Determining the capacities of VM (VM sizing) for running applications and placing of
VMs on PM based on their requirements
 Monitoring and profiling the resource utilization for hotspot detection (overloaded and
under loaded PMs)
 Resize and remap the VM to another PM.
The VM placement and VM migration act as backbone to VM consolidation process. The
challenges for efficient VM consolidation are:-
 Finding the proper VM- to-PM mappings to minimize the acceptable cost function
 Dynamic hotspot detection
 Performing VM migration process with minimal service downtime and resource
consumption during migration process

19
Fig.2. Server Consolidation Steps
3.2.1. Virtual Machine Placement
VM placement [14] is the process of mapping VMs to PMs. The placement approach should
consider multiple resources such as CPU, memory, disk storages and network bandwidth to
reduce the energy consumption at data centers and also needs to maintain the energy performance
tradeoff.
The VM placement involves two main steps:
 Provisioning of resources for the virtual machines according to the capacity requirements
of corresponding applications (VM sizing)
 Actual placement of VMs onto PMs.
VM placement problem is a NP –Hard problem [34] and no optimal solution exists for it. A set
of heuristics is used to solve VM placement problem.
3.2.1.1. State- of-the Art in VM Placement
The energy conservation at data centers increases by maintaining optimal placement of the VMs
on the PM. The existing VM placement heuristics are classified into the following:-
 Availability aware VM placement - The main challenge of the VM placement
algorithms is to identify suitable VM-PM mapping so as to satisfy workload demands of
various applications. Once the hotspot is detected in source PM, the VM placement
algorithm finds the workload demands of each VM and identifies the target PM based on
the resources availability such as CPU, memory, network and disk [13-17].
 Affinity aware VM placement can be achieved by considering both availability of VM’s
resources and relationship among VMs. This approach improves the response time of the
applications and optimizes the usage of resources. If a hotspot has occurred because of
network resource crunch among PMs, then placing two communicating VMs on the same
PM will reduce the network overload. In [15] [20], the intra and inter-PM network traffic
and also network topology has been considered for VM placement. In [17] [19], the inter

20
memory sharing VMs are identified and placed on a single PM. This reduces the number
of servers and memory resource used.
 Workload aware VM placement- The efficient VM consolidation can be accomplished
by considering workload interference between co-located VMs to avoid performance
degradation. In workload aware VM placement mechanism, the different combination of
workload mix such as CPU, I/O intensive, etc., are investigated in order to find optimal
VM placement. The benefits of workload mixing can be proactively predicted by using
machine learning algorithms which can be used for VM placement decisions [21] [22].
3.2.1.2. Analysis
The VM placement algorithms are analyzed based on the following parameters:-
 Resources availability: the availability of resources (CPU, memory, network and disk of
PM determines the number of VMs on it.
 Heterogeneity of PMs
 Workload type of VMs
An efficient VM placement algorithm should consider all the above mentioned parameters during
VM placement, but most of the existing VM placement algorithms have not considered the all
parameters.
3.2.2. Virtual Machine Migration
VM migration [23] has become an indispensable tool that facilitates resource management at
cloud data centers for a wide variety of key scenarios such as load balancing, fault- tolerance, hot
spot mitigation and server consolidation. The VM migration can be either offline or live. The
offline VM migration follows traditional suspend and resume mechanism which increases service
down time, causes performance degradation of applications running in VM. Live migration
relocates the VM from source to target PM without disturbing the applications running in the VM
resulting in minimal downtime. In live migration process, the virtual disks are stored on some
shared storage such as Network File Systems, which is accessible from entire network. Hence,
virtual disks are not transferred. The steps of Live migration are as follows (Fig. 3):-
1. Once migration process is triggered, source PM transfers the VM’s memory pages to
target PM without suspending the VM (and hence it is called live migration). The dirtied
memory pages during the migration process are transferred to target PM which leads to
multiple iterations. This is known as iterative precopy-phase.
2. This iterative process ends and enters into suspend -and- copy phase when either one of
the following conditions occurs:
a. The number of pre-copy iterations are greater than the fixed threshold value (n)
b. The network traffic generated is greater than the multiples of VM size
c. The memory page dirty rate is greater than a memory transmission rate
d. The number of memory pages dirtied during current iteration becomes smaller
than a predefined threshold

21
3. In suspend-and-copy phase, the VM being migrated is suspended in the source PM and
the remaining dirty memory pages as well as the state of the CPU register are transferred
to the target PM. Once the migration process is completed, the target PM resumes the
VM.
Fig. 3. Live Virtual Machine Migration Mechanism
3.2.2.1. State-of- the Art in Live Migration Research
It is hard to provide a quick live VM migration with less resource overhead because large amount
of data transfer during migration results in performance degradation of VM applications. The
response time and throughput of the data center applications heavily depends on the VM
migration process. Hence, it is essential to improve the performance of migration process to
facilitate efficient resource management at data centers.
3.2.2.1.1. Optimizations of Live Migration
Many methods have been proposed to improve live migration and optimize its performance
metrics such as total migration time and downtime to provide uninterruptible services to
applications running in VMs.
 Memory compression: Live migration performance is improved by minimizing the amount of
data transfer to the destination using the technique called memory page compression [24]. This
technique compresses the source PM memory pages to minimize the amount of data during
live migration process and decompresses the memory pages in target PM. However, the
performance of application running in VM depends on selection of suitable compression and
decompression algorithms, resources availability and workload characteristics. The changes in
workload characteristics introduce significant resource overhead during compression and
decompression process leads to performance degradation of applications running in VM.
Therefore, it is necessary to ensure trade-off between resource overhead and compression
effects to achieve efficient migration.

22
 Delta page transfer: This technique reduces the network bandwidth consumption by
maintaining a cache of previously transferred memory pages. This optimizes the transmission
of dirtied pages by sending the difference between cached page content and page content that is
going to be transferred. This improves the live migration process with reduced risk of service
interruption [25][32].
 Data deduplication: This technique finds identical data inside the memory and disk of a single
VM and removes that duplicates during the live migration process. The performance of this
technique depends on selection of suitable techniques for finding identical data on memory and
disk contents [31-32].
 Post-Copy Approach: This approach postpones the memory page transfer phase until after the
VM’s CPU state has been transferred to the target PM and resumed there. The performance of
post-copy approach depends on the way and which the memory pages are fetched from source
PM during live migration process. There are three main ways the memory pages are fetched
from source PM [26].
 Post-copy via demand paging- The page faults are serviced by requesting the
referenced page over the network from the source node
 Post-copy via active pushing- This technique, proactively pushes the memory
pages from source to the target, while VM continues executing at the target PM
 Post-copy via prepaging- This approach extends the active pushing technique by
estimating the spatial locality of the VM’s memory access pattern in order to
anticipate the occurrence of major page faults
The post-copy approach ensures that each memory page is transferred at most once, hence
avoiding the duplicate transmission overhead of pre-copy approach. However, this approach
does not have the reliability as Pre-Copy approach because VM cannot restart in the source
PM, if the destination PM crashes.
 Hybrid pre and post copy: This hybrid approach performs single round of pre-copying which
precedes the virtual CPU state transfer. This is followed by post-copying of the remaining dirty
pages from the source PM. Therefore, it gets the benefits of both approaches and improves live
migration process [27].
3.2.2.1.2. Multiple VM Migrations
Whenever there is a need to shift entire virtual clusters to different locations, it requires large
amount of data to be transferred over the network which results in network and CPU overhead.
This leads to performance degradation of applications running in the PM. Hence, the identical
contents of co-located VMs in the physical machine can be pro-actively tracked and transferred
only once to the target PM. This improves the performance of multiple VM migration. This is
known as “Gang Scheduling” of virtual machines [27-32]. It optimizes both memory and network
overhead of migration.
3.2.2.1.3. Migration in Network
Several enterprises have data centers that are spread across the world to offer efficient services to
worldwide users. The network can be LAN or WAN. Live migration of VMs across worldwide
data centers entails transferring virtual CPU, memory state and disk state to target PM and also it

23
necessitates network re-configuration as the VM shifts into a new subnet where a new IP address
is assigned to the VM [29-30] [32].
3.2.2.2. Analysis
The live migration process consumes resources such as CPU and network which can seriously
impact the performance of both the VM being migrated as well as other VMs in the PM. The
performance of the migration approach depends on the parameters such as
 VM size consumes significant amount of CPU and network resources during live
migration process. Thus, an optimal VM size needs to be selected to reduce resource
overhead during migration process.
 Page dirty rate indicates the rate at which memory pages are dirtied in source PM
during migration process. If page dirty rate increases, it results in more data being sent
per iteration over the network, results in increasing of total migration time
 Available network capacity between source and target PM
 Available CPU resource at both source and target PM
The following performance metrics should be considered to evaluate performance of live
migration process;-
 Migration time- the time between initiation of migration and successful completion
 Service downtime – the time for which service is unavailable during migration
 CPU and network resource consumption during migration process
Table .2. Comparison of state-of-art existing live migration approaches

24
From Table 2, it is observed that the two important parameters such as VM size and available
CPU cycles, have not been taken into consideration during migration process. For efficient
migration, all the mentioned parameters should be taken into account and also the performance of
the migration algorithm needs to be measured.
4. CONCLUSION
The cloud data centers are prominent hosting infrastructure for provisioning of its services. The
surging energy consumption of these data centers has become a critical problem from both the
economic and environmental standpoints. This paper presents an overview about various dynamic
energy management techniques both in hardware and virtualization levels and highlights issues
and its significance. However, the hybrid approach that covers both hardware and resource
management levels (virtualization) needs to be developed to provide continuous and efficient
energy management in virtualized data centers.
REFERENCES
[1] K.Kant,(2009) “ Data Center Evolution A Tutorial on State of the Art, Issues, and Challenges”,
Journal of Computer Networks, Vol. 53, pp 2939–2965.
[2] Amazon Elastic Compute Cloud (Amazon EC2), http://aws.amazon.com/ec2/
[3] Microsoft’s Cloud Platform (Windows Azure), http://www.windowsazure.com/
[4] J.Kaplan, W.Forrest, & N.Kindler, (2008) “Revolutionizing Data Center Energy Efficiency.
Technical report”, McKinsey & Company
[5] Gartner Says Energy-Related Costs Account for Approximately 12 Percent of Overall Data Center
Expenditures,(2011) http: //www.gartner.com/it/ page. Jsp? Id = 1442113
[6] P.Barham, B.Dragovic, K.Fraser, S.Hand, T.Harris, A.Ho, (2003) “Xen and the Art of
Virtualization”, 19th ACM Symposium on Operating Systems Principles, pp. 164-177,USA
[7] A.Beloglazov, R. Buyya, Y.Lee, A.Zomaya, (2011) “A Taxonomy and Survey of Energy Efficient
Data Centers and Cloud Computing”, Journal of Advances in Computers. 82, 47-111
[8] G.Luigi, W.Lassonde, S.Khan, G.Valentini, et al, (2011) “An Overview of Energy Efficiency
Techniques in Cluster Computing Systems”, journal of Cluster computing, Springer
[9] L.Benini, A.Bogliolo, G.Micheli, (2000) “A Survey of Design Techniques for System-Level Dynamic
Power Management”, IEEE Transactions on Very Large Scale Integration (VLSI) Systems. 8, 299-
316
[10] S.Gurumurthi, A.Sivasubramaniam, M.Kandemir, et al, (2003) “DRPM: Dynamic Speed Control for
Power Management in Server Class Disk”, 30th Annual International Symposium on Computer
Architecture, pp 169–181. USA
[11] D.Colarelli, D.Grunwald, (2002) “Massive Arrays of Idle Disks for Storage Archives”, Proceedings
of the ACM/IEEE Conference on SuperComputing. pp 1–11. USA
[12] C.Gunaratne, K. Christensen, B. Nordman, et al, (2008) “Reducing the Energy Consumption of
Ethernet with Adaptive Link Rate (ALR)”, Journal of IEEE Trans Computer, Vol. 57, 448-461
[13] S. Srikantaiah, A.Kansal, F.Zhao, (2008) “Energy Aware Consolidation for Cloud Computing”,
Proceedings of the Conference on Power Aware Computing and Systems, USA
[14] C.Hyser, B.McKee, R.Garner, B.Watson, (2008) “Autonomic Virtual Machine Placement in the Data
Center”, HP laboratories, HPL-2007-189
[15] N.Bobroff, A.Kochut, K.Beaty, “Dynamic Placement of Virtual Machines for Managing SLA
violations”, IBM T.J Watson Research Center, USA.
[16] A.Beloglazov, R.Buyya, (2010) “Energy Efficient Resource Management in Virtualized Cloud Data
Centers”, International Conference on Cluster, Cloud and Grid Computing, pp. 826-831

25
[17] T.Wood, G.Levin, P.Shenoy, et al, (2009) “Memory Buddies: Exploiting Page Sharing for Smart Co-
location in Virtualized Data Centers”, International Conference on Virtual Execution Environments,
pp 31-40, USA
[18] X.Meng, V.Pappas, L.Zhang, (2010) “Improving the Scalability of Data Center Networks with Traffic
Aware Virtual Machine Placement”, International Conference on Information Communications, pp.
1154-1162, USA
[19] M.Sindelar, R.Sitaraman, P.Shenoy, (2011) “Sharing Aware Algorithms for Virtual Machine Co-
location”, 23rd ACM Symposium on Parallelism in Algorithms and Architectures, pp 367-378, USA
[20] R.Reutiman, J.Sonnek, J.Greensky, A.Chandra, (2009) “Starling Minimizing Communication
Overhead in Virtualized Computing Platforms using Decentralized Affinity Aware Migration”,
Technical Report, University of Minnesota.
[21] A.Corradi, M.Fanelli, L.Foschini, (2012) “VM consolidation: A Real Case Based on Open Stack
Cloud”, Journal of Future generation of computer system
[22] U.Kousiouris, T.Cacinotta, T.Varvarigou, (2011) “The Effects of Scheduling, Workload type and
Consolidation Scenarios on Virtual Machine Performance and their Prediction through Optimized
Artificial Neural Networks”, Journal of system software, vol. 84, pp. 1270-1291
[23] C.Clark, K.Fraser, S.Hand, J.Hnasen, E.Jul, et al, (2005) “ Live Migration of Virtual Mahines”,
Proceedings of the 2nd Conference on Symposium on Networked Systems Design and
Implementation , PP. 273-286, USA
[24] H.Jin, L.Deng, S.Wu, X.Shi, X.Pan, (2009) “Live Virtual Machine Migration with Adaptive Memory
Compression”, In: IEEE CLUSTER, New Orleans, LA
[25] S.Hacking, B.Hudzia, (2009), “Improving the Live Migration Process of Large Enterprise
Applications”, International Workshop on Virtualization Technologies in Distributed Computing”, pp
51-58, USA
[26] M. Hines, K. Gopalan, (2009) “Post-copy Based Live Virtual Machine Migration Using Adaptive
Pre-paging and Dynamic self ballooning”, International Conference on Virtual Execution
Environments, PP. 51-60, USA
[27] .Sahni, V.Varma, (2012) “A Hybrid Approach to Live Migration of Virtual Machines”, International
Conference on Cloud Computing in Emerging Markets, India
[28] U.Deshpande, X.Wang, K.Gopalan, (2011) “Live Gang Migration of Virtual Machines”, International
Conference on High Performance Distributed Computing, pp. 135-146, USA
[29] P.Riteau, C.Morin.et al, (2011) “Shringer: Efficient Wide Area Live Virtual Machine Migration
Using Distributed Content Based Addressing”, 17th International Conference on Parallel Processing,
pp. 431-442, Heidelberg
[30] R.Bradford , E. Kotsovinos , et al,(2007) “Live Wide Area Migration of Virtual Machine including
Local Persistent State”, 3rd International Conference on Virtual Execution Environments, pp. 169-
179, USA
[31] X.Zhang, Z.Huo, J.Ma, et al, (2010) “Dan: Exploiting data duplication to accelerate live virtual
machine migration” International conference on Cluster Computing, pp. 88-96, Crete.
[32] T.Wood, P. Shenoy, et al, (2011) “Cloudnet: Dynamic Pooling of Cloud Resources by Live WAN
Migrations of Virtual Machines”, 7th ACM SIGPLAN/SIGOPS International Conference on Virtual
Execution Environments, pp. 121-132, USA
[33] Open compute project, http://opencompute.org/2012/11/08/building-a-better-server-chassis
[34] Skiena, S.: The Algorithm Design manual, ISBN 0-387-94860-0ssss
[35] M. Gupta, S. Singh, (2007) “Using Low-Power Modes for Energy Conservation in Ethernet LANs”,
26th IEEE International Conference on Computer Communications”, pp 2451-2455, AK
[36] B.Heller, S. Seetharaman, P. Mahadevan, (2010) “Elastic Tree: Saving Energy in Data Center
Networks”, 7th USENIX conference on networked systems design and implementation, pp 1-17,
Berkeley, USA
[37] E.Pinheiro, R.Bianchini, (2004) “Energy Conservation Techniques for Disk Array-Based Servers”,
18th Annual International Conference on Supercomputing (ICS), pp 68-78, USA

26
Authors
Veni T received the B.E Degree in Computer Science and Engineering from Anna
University, Chennai in 2008 and the M.E Degree in Computer Science and Engineering
from Anna University, Chennai in 2011. Currently, she is pursuing the PhD Degree at the
Department of Computer Science and Engineering in National Institute of Technology,
Tiruchirappalli, India. Her research interests include Operating System and Cloud
Computing.
Mary Saira Bhanu S received the B.E Degree in Electronics and communication from
Madurai Kamaraj University in 1986, the M.E Degree in Computer Science from
Bharathidasan University in 1989 and the Ph.D. Degree from the Department of Computer
Science and Engineering from National Institute of Technology, Tiruchirappalli in
2009.Currently, she is a Associate Professor at the Department of Computer Science and Engineering in
National Institute of Technology, Tiruchirappalli, India. Her research interests include OS, Real Time
Systems, Distributed Computing, Grid Computing and Cloud C omputing.

DYNAMIC ENERGY MANAGEMENT IN CLOUD DATA CENTERS: A SURVEY

More Related Content

What's hot (16)

Similar to DYNAMIC ENERGY MANAGEMENT IN CLOUD DATA CENTERS: A SURVEY (20)

Recently uploaded (20)

DYNAMIC ENERGY MANAGEMENT IN CLOUD DATA CENTERS: A SURVEY