Toward an Autonomic and Adaptive Load Management Strategy for

Reducing Energy Consumption under Performance Constraints in

Data Centers

Abdulrahman Nahhas

, Sascha Bosse

, Matthias Pohl

and Klaus Turowski

Very Large Business Applications Lab, Faculty of Computer Science, Otto-von-Guericke University Magdeburg, Germany

Keywords: IT Resources Management, Adaptive Load Distribution Strategies, Virtual Machines Live Migration, Energy-

aware Virtual Machines Allocation, Heuristic and Metaheuristic Optimization, Data Center Management.

Abstract: The future vision of IT-industry is shifting toward a utility-based offering of computing power using the

concepts of pay-per-use. However, the elasticity and scalability characteristics of cloud computing massively

increased the complexity of IT-system landscapes, since market leaders extensively expanding their IT-

infrastructure. Accordingly, the carbon-footprint of data centers operations is estimated to be the fastest

growing footprint among different IT fields. The majority of contribution in the examined literature that

address IT resources management in data centers exhibits either a specific or a generic nature. The specific

solutions are designed to solve specific problems, but yet neglecting the dynamic nature of IT-systems. The

design of generic solutions usually overlooks many details of the investigated problems that have an impact

on the possible optimization potential. One can argue that an optimized combination of different algorithms

used during a specified time span would outperform a single specific or generic algorithm for the management

of IT recourses in data centers. Therefore, a conceptual design for an autonomic and adaptive load

management strategy is presented to investigate the aforementioned hypothesis. Our initial experimental

results showed considerable improvement when multiple algorithms are used for the allocation of virtual

machines.

1 INTRODUCTION

Virtualization strategies have changed the traditional

design and deployment of IT-system landscapes. The

future vision of IT-industry is shifting toward a

utility-based offering of computing power using the

concepts of cloud computing. Therefore, market

leaders are massively expanding their IT-system

landscapes (Kushida et al., 2011), in which the

optimization of the IT-system design and engineering

is not significantly important for decision makers to

announce investments worth millions of euro for new

IT-infrastructure. However, the elasticity and

scalability features of the cloud computing model

have a major impact on the complexity of IT-system

landscapes, since the incoming workload becomes

much harder to predict. Consequently, the massive

expansions of IT-system landscapes in addition to the

https://orcid.org/0000-0002-1019-3569

https://orcid.org/0000-0002-2490-363X

https://orcid.org/0000-0002-6241-7675

aforementioned characteristics of cloud computing

radically complicated the management process of

those landscapes.

The energy costs will keep increasing, which

poses a necessity for IT-service provider to

investigate the efficiency of their operations to reduce

costs while holding their Service Level Agreements

(SLA). The efficiency of utilizing IT-resources

becomes a market competitive advantage for IT-

service provider to offer reliable but yet sustainable

IT-services with reasonable costs in the market. The

main fraction of costs is encountered through the

energy consumption of physical servers, which is

estimated to reach up to 50 % of the overall costs. In

addition, statistical analysis on the worldwide energy

consumption triggered an alarm on a governmental

level since numbers suggest a total growth of roughly

hundred percent reported by data center industry

Nahhas, A., Bosse, S., Pohl, M. and Turowski, K.

Toward an Autonomic and Adaptive Load Management Strategy for Reducing Energy Consumption under Performance Constraints in Data Centers.

DOI: 10.5220/0007754004710478

In Proceedings of the 9th International Conference on Cloud Computing and Services Science (CLOSER 2019), pages 471-478

ISBN: 978-989-758-365-0

471

between 2005 and 2010 (Koomey, 2011). Obviously,

the associated CO

emissions of data center’s

operations reported, accordingly, tremendous growth

and estimated to be the fastest growing carbon-

footprint among different IT fields (Avgerinou et al.,

2017). On a European Union (EU) level, some

initiatives, research, and further regulations have

been introduced to suppress the impact of data center

footprint as, for instance, the EU Data Center

Code of Conduct (Avgerinou et al., 2017). Those

facts motivated data centers operators to revision the

management strategies of data centers to achieve a

higher level of sustainability in service offering and

management.

In this research, we will present an overview on

the current advances of load management strategies

targeting sustainable management of IT resources in

data centers. The second section is dedicated to shed

a light on the usual formulation of static and dynamic

Virtual Machines (VMs) placement problems. Based

on the initial findings, we present our intermediate

analysis from our systemic literature analysis. We

discuss the adopted solution approaches from an

algorithmic point of view to present our hypothesis

and research question. In the third section, a

conceptual design for an autonomic and adaptive load

management strategy is presented. The fourth section

is dedicated to present the initial computational

results to answer the posed research before closing

the paper with a conclusion.

2 STATE OF THE ART AND

LITERATURE ANALYSIS

Many efforts and investigations have been dedicated

in the last two decades to propose efficient but yet

specific solutions for data center management. The

majority of the static virtual machines placement or

virtual machines consolidation problems are

formulated in different forms of bin-packing

problems (Lopez-Pires and Baran, 2015). The

simplest form is the single dimension bin-packing

problem taking the CPU as the main resource to

allocate the virtual machines. The goal is eventually

to place the existing virtual machines modelled as

items into the minimum number of active physical

hosts modelled as bins to reduce the overall energy

consumption. Many similar problems have been

intensively addressed in the literature in the fields of

scheduling and operations research as for instance,

the identical and non-identical parallel machines

scheduling problems and different forms of bin-

packing problems (Pinedo, 2012; Skiena, 1998).

Unfortunately, the majority of those problems have

been proven to be NP-Hard. In addition, the

complexity of a considered problem is further

increased when it is formulated in form of multi-

dimensional bin packing problem. In such more

realistic problem formulation, three recourses

dimensions can be taken into consideration as for

instance, CPU, memory and storage.

Therefore, the majority of the research conducted

on the virtual machines placement problems is

inspired by heuristic approaches. They are usually

adopted when the solution space of a problem cannot

be investigated entirely with the current

computational power in polynomial time. More

profoundly, heuristic approaches comprise two main

categories: constructive and improvement

approaches. The constructive approaches are simple

straightforward algorithms, in which the decision for

allocation is taken instantly after conducting some

calculations without searching in the solution space

of the problem. They are intuitive to implement and

exhibit a light execution time to take decision for a

new allocation. However, they are not robust against

major modifications in the problem formulation. In

essence, if the underlying infrastructure or the

incoming workload patterns witness a major change,

their performance usually massively degrades and

their internal design has to be adjusted accordingly

(Keller et al., 2012).

Therefore, IT- research has been for decades

relying on improvement and metaheuristic

approaches for solving static virtual machine

consolidation problems. Improvement heuristics are

conceptually more sophisticated heuristic procedures

in comparison to the constructive ones since the

construction of a solution is the first step in their

internal functionality. Thereafter, based on a solution,

an improvement heuristic seeks to conduct single or

several changes on the constructed allocation to find

a so-called neighbour solution, which hopefully

yields to a better investigated objective function. The

modification process is then iteratively conducted

until some breaking criterion is met. Finally, the

metaheuristic approaches are the most powerful

optimization techniques that fall under heuristic

procedures. The majority of them are inspired by

some natural phenomena, as for instance, Genetic

Algorithms (evolution theory) (Holland, 1992) or

Simulated Annealing (annealing process of metals)

(Kirkpatrick et al., 1983). They are fundamentally

based on an improvement heuristic and an overall

control strategy that attempt to guide the

CLOSER 2019 - 9th International Conference on Cloud Computing and Services Science

472

improvement procedure to achieve better

optimization results and avoid false or local optima.

However, the adoption of those optimization

techniques is associated with significantly higher

computational effort to find good solutions in

comparison to constructive approaches. Therefore,

their adoption is strictly subject to whether the

allocation decision needs to be taken instantly or not.

In the static virtual machine consolidation problem

case, the required computational effort to find a very

good to a near optimal solution does not have to be

necessarily instant since the migration of the virtual

machines is conducted anyways in an offline mode.

In addition to the rapid evolvement of

virtualization strategies, the introduction of virtual

machines live migration algorithms shifted the focus

of academia from the classical static virtual machines

consolidation problems to the so-called dynamic

Virtual Machines Placement problems (VMP).

Dynamic virtual machine placement implies that

virtual machines are subject to reallocation processes

during operational time based on the dynamic state of

the system to meet various goals, as for instance, to

reduce energy consumption. The virtual machines

live migration algorithms address the migration

process of virtual machines during the operational

time, in which the goal is to migrate a virtual machine

from an active physical host to another one with the

minimum downtime (Clark et al., 2005; Jin et al.,

2014). Thus, to reduce the impact of the migration

process on the associated hosted services in order to

avoid violations in the signed SLA while reducing the

overall energy consumption. This major advance in

virtualization strategies led to the introduction of a

new research stream under the term “energy-aware”.

In the past decade, many algorithms have been

presented to schedule virtual machines or tasks taking

into consideration the increase in the energy

consumption of the underlying infrastructure. In

addition, the popularity of metaheuristic approaches

for virtual machines allocation is significantly

decreased with major domination of heuristics

approaches. Obviously, the reason can be traced back

to the relatively high required computational effort of

them to find suitable allocation. Based on our initial

analysis on the prominent publications on science

direct database the majority of the found articles are

presenting energy-aware solutions such as the

contributions of (Zheng and Cai, 2011; Goiri et al.,

2012; Bodenstein et al., 2012; Beloglazov et al.,

2012; Luo et al., 2013; Zhang et al., 2014; Tesfatsion

et al., 2014; Khani et al., 2015; Dupont et al., 2015;

Kumar and Raghunathan, 2016; Carli et al., 2016;

Vafamehr and Khodayar, 2018; Marotta et al., 2018;

Malekloo et al., 2018; Kaur and Chana, 2018; Han et

al., 2018). Energy-aware heuristics are specially

designed algorithms to reduce energy power in data

centers and usually based on a core power model that

highly determines their behavior. The majority of the

found articles are based on constructive heuristic

procedures since the light execution time is of major

importance for solutions with live migration

capabilities. For instance, in (Beloglazov et al., 2012;

Dupont et al., 2015; Han et al., 2018; Zhang et al.,

2014) the core functionality of the presented

heuristics is based on the designed or adopted power

model.

However, IT industry requires solutions that are

able to adapt to the dynamic nature of those systems

with minimal human intervention. Therefore, the

complexity of our current and future IT-systems

requires a deep analysis of the current understanding

of artificial intelligence techniques and its advances

for automation proposes. Therefore, based on our

initial literature analysis, we identified two main

research streams that explicitly deal with the virtual

machines placement problem based on some machine

learning approaches to reduce energy consumption in

data centers. In the first stream, adaptive approaches

are presented such the contributions of (Jeyarani et

al., 2012; Xu et al., 2012; Vitali et al., 2015; Suresh

and Sakthivel, 2017; Yoon et al., 2017; Zhou et al.,

2018; Kumar and Singh, 2018). Adaptive solutions

are algorithms with monitoring capabilities that are

designed to react or adapt to specific scenarios such

upper and lower threshold of server’s workload or

statistical analysis on workload to rely on some

predictions (Yoon et al., 2017). The internal design of

the presented solutions is definitely more

sophisticated than the energy-aware solutions and the

majority of them have been presented in form of

frameworks. In essence, in the adaptive solutions, the

algorithm is a component that relays on some

prediction model to derive predictions for the upper

and lower threshold of servers to take the allocation

decision as in (Jeyarani et al., 2012; Kumar and

Singh, 2018; Suresh and Sakthivel, 2017; Vitali et al.,

2015; Yoon et al., 2017; Zhou et al., 2018).

The second stream presents autonomic

frameworks as for instance the contributions of

(Wang et al., 2008; Xu et al., 2012; Tchana et al.,

2013; Amoretti et al., 2013; Delaval et al., 2015).

Autonomic strategies are self-organizing strategies

that exhibit sophisticated features usually targeting

the management of landscape on an application level

as for instance, application scalability (Tchana et al.,

2013; Wang et al., 2008; Delaval et al., 2015). Based

on the conducted analysis, the majority of

Toward an Autonomic and Adaptive Load Management Strategy for Reducing Energy Consumption under Performance Constraints in Data

Centers

473

contributions in the examined literature exhibit either

a specific or a generic nature. The specific solutions

are designed to solve specific problems, but yet

neglecting the dynamic nature of IT-systems

especially in a cloud-computing context. The design

of generic solutions usually overlooks many details of

the investigated problems that have an impact on the

desired optimization potential and thus, do not

achieve the possible optimization potential.

Therefore, we aim to answer the following research

question: Will a combination of heuristic and

metaheuristic approaches to present a hybrid

framework for the management of data center

operation overcome the aforementioned drawbacks in

the analyzed literature?

The question is based on the argument that an

optimized combination of different algorithms used

during a specified time span would outperform a

single specific or generic algorithm for the

management of IT recourses in data centers. To

exactly know how the combination should be built,

we need to rely on some overall optimization

mechanisms, as for instance, a metaheuristic

approach. The main idea is to exploit the light

execution time of constructive approaches to take

instant decision for allocation and the robustness of

metaheuristic approaches to achieve a higher

optimization potential. In the course of the next two

sections, we present a conceptual design of an

adaptive and autonomic concept for the management

of data centre operations based on multiple

algorithms to answer the research question and

validate the aforementioned hypothesis.

3 ADAPTIVE AND AUTONOMIC

LOAD DISTRIBUTION

STRATEGY

The concept under design is presented in Figure 1.

The framework consists of three main components:

the workload Monitoring and Prediction component

(MP), the Adaptive component (AD) and the

Artificial Intelligence component (AI). The MP

component is designed to deliver likely future

workload distributions of the considered Virtual

Machine (VMs) types. Based on the analyzed

literature, one can rely on statistical analysis or

machine learning approaches on the workload

demand to predict the incoming workload for a

specific time span (Kumar and Singh, 2018). In some

studies, it is even suggested to conduct statistical

analysis on the power consumption requirements on

an application level to derive power consumption

profiles of applications. For instance, Bartalos et al.

(2016) presented an aggregated model to predict the

power demand of an application running on specific

servers using multiple linear regression models. It is

of interest to study the behavior of the optimization

model if one combines both prediction approaches to

derive workload profiles as well as energy power

profiles.

The AD component contains an optimization and

evaluation models. In our prototypical analysis and

implementation, we relied on Genetic Algorithms

(GA) to design the optimization model, which is

dedicated for finding the best combination of

heuristics that should be used for load management

depending on the system state over time. The

optimization model might be further fed with

different algorithms, performance models,

operational constraints and finally different sensitive

parameters that are collected through feedback loops

from the AI component. As for operational

constraints, different forms of Service Level

Agreements (SLAs) can be modeled to suppress the

impact of live migrations on the associated possible

penalties. Sensitive parameters include performance-

based and workload-based measurements, as for

instance, the physical server’s upper and lower

thresholds and the global threshold of the system.

Such measurements have a major impact on the

design and the functionality of the scalability

mechanisms of the systems. The evaluation model

might be based on a simulation model.

The AD component provides a solution that

contains a combination of different algorithms to be

used during a defined time span for load management

in addition to a set of sensitive parameters (e.g.

Threshold of servers workload) to control the

migration policies. The AI component is dedicated to

learn from the optimization results, pass the solution

to the underlined infrastructure and provide a

feedback loop to continuously adjust the performance

of the AD component to achieve better results in the

next optimization interval. The goal of the feedback

loops is mainly to reduce the deviation of the inquired

predictions on the workload and other measurements

from the actual ones and systematically achieving a

higher accuracy of the optimization model. Fuzzy sets

have been, for instance, applied to address different

problems, especially, in the field of supply chain

management (Ganga and Carpinetti, 2011). They are

a powerful approach to model uncertainties of some

phenomena and incorporating expert’s knowledge.

Thus, they can be adapted to model the highly

dynamic behavior of IT landscapes and its associated

CLOSER 2019 - 9th International Conference on Cloud Computing and Services Science

474

Workload monitoring and prediction component

Artificial intelligence componentAdaptive component

Workload predictor

Evaluation model

(e.g. Simulation)

Predictecd interarival rates per VM-Types T i

Execution model

(Hypervisor scripting)

Knowledge base

Actual interarival rates T i

Performance Metrics

Server

Monitoring

database

Optmization model

(Metaheuristic e.g. GA)

Adaptive

execution

plans

Solution

Load distribution

strategies

Performance and

costs models

Data analytic

models

Data analytic

models

Fuzzy-based

rules

Actual Workload i

SLA Constraints

Figure 1: A conceptual model for an autonomic and adaptive load management strategy.

uncertainties. The overall goal is to detach the

optimization component as soon as the AI component

is trained and ready to work without the AD

component. This goal can be systemically achieved

through collecting data on the applied solutions and

their deviation from the real data to derive

measurements and sensitive parameters and apply

data analytics approaches to extract knowledge. The

obtained knowledge will be then further reflected in

forms of rules and actions that must be applied to

react to different phenomena.

4 INITIAL ANALYSIS ON THE

PRESENTED CONCEPT

The initial analysis is dedicated to investigate the

validity of the aforementioned hypothesis and answer

the research question. Therefore, we relied on

collected information through interviews with experts

to mimic the MP component to derived different

workload distribution for different VMs types of a

real system. In addition, we did not extend our

experiment to investigate the role of the AI

component since the research question is profoundly

based on the functionality of the adaptive component.

In the course of the next section, we present a brief

problem formulation to investigate the functionality

of the adaptive component before presenting our

initial findings.

We relied on a simple problem formulation to

draw some conclusions on whether the concept

achieves the desired optimization potential or not

(Nahhas et al., 2018). One can assume that the

optimization potential tends to increase with the

increase in the complexity of a considered problem.

The adaptive component is designed to optimize the

functionality of the system, in which a simplified set

of algorithms (load -concertation and -balancing) and

sensitive parameters (Thresholds of physical servers)

are passed to the optimization model. In our analysis,

the optimization model is based on Genetic

Algorithms (GA), while a simulation model is built to

evaluate the fitness of the solution candidates. In this

Toward an Autonomic and Adaptive Load Management Strategy for Reducing Energy Consumption under Performance Constraints in Data

Centers

475

initial analysis, we set the optimization model to

investigate whether we need to change our allocation

algorithm every hour. This implies that a solution

candidate in the population of the GA comprises 24

integer values that represent the codes of the

modelled allocation algorithms every hour and

thresholds of physical servers.

We simulated five days of operations and relied

on the expert’s interviews to derive mathematical

distribution that describes the behavior of the

considered virtual machines as shown in Table 1. The

IT landscape of the considered system consists of

eight homogeneous servers, which host five different

types of virtualized systems deployed in 290 VMs.

The capacity of the main memory of the servers is 500

GB. Unlike many problem formulations in the

literature, the bottleneck of the considered system is

not the CPU capacity but rather the main memory,

since the majority of the offered virtual machines

servers as desktops. This implies that sharing the

main memory is not allowed. We formulated the

problem to take into consideration the number of

migrated virtual machines as well as the total number

of online hours of all servers during a time span.

Table 1: Descriptive information of the virtual machines in

the considered IT landscape.

VMs Type

Main

memory

Online time

Offline time

Assistant

VMs

Triangular

[1, 6, 3]

Triangular

[22, 30, 24]

Researcher

VMs

Triangular

[6, 14, 8]

Triangular

[14, 18, 16]

SAP system

access 1

Triangular

[2, 8, 5]

Triangular

[16, 22, 19]

SAP system

access 2

Triangular

[2, 8, 5]

Triangular

[16, 22, 19]

SAP system

access 3

Triangular

[2, 8, 5]

Triangular

[16, 22, 19]

Given a data centre, that consists of a set of

physical machines, which are serving customer

requests to deploy various types of virtual machines.

The problem under investigation might be formalized

in the following:

 Let P = {p

, …, p

}: be a set of m physical

machines.

 Let V = {v

, …, v

}: be a set of n online virtual

machines.

 Let R = {r

, …, r

}: be a set of o resources required

for each v



 Let D

i,y

: be the required resource for v



V from

resource type y



 Let C

j,y

: be the total capacity of h



H of the

resource type y



 Let A (A



{1, 2}): denote the codes that describe

the algorithms that can be used for allocation of

virtual machines.

 Let S = {s

, …, s

}: be the set of m values, which

represent the online hours of the physical machines

P = {p

, …, p

} during a time span T.

 Let   



denote the number of migrated virtual

machines over the time interval T.

Let



denote the set of all possible combinations

of the considered set of algorithms A during a defined

time interval T. It is desired to find the combination

of the algorithms H





to allocate the set of VMs V

on the hosts dynamically. This combination is then

subject to the minimization of γ

refers to the total

online hours of all servers and minimization of γ

refers to the total number of migrated virtual

machines over a time interval T as shown in equation

(1). Those are to reduce total energy consumption

taking into consideration the impact of live migration

on the performance of the system in a simple

formulation.





  













 





(1)

 





























(2)

For solving the problem, we adopted a weighted-

sum approach to formulate the objective function in

formula (3) to obtain formula (4).









 









  









(3)













 







 













 



 

(4)

The simulation has been set to consider a time

interval of 120 hours, which correspond to five days

of operations. For the hybrid approach, 10 to 20

replications were recorded during the optimization

before drawing any conclusion on the fitness of a

solution candidate. Finally, after acquiring the

solution, 200 replications are recorded to ensure the

quality of the obtained results and eliminate the bias

from the system for each simulated scenario. A 95 %

confidence interval has been applied to all observed

measurements to observe the possible deviation and

obtain the margin of error. The results showed that the

CLOSER 2019 - 9th International Conference on Cloud Computing and Services Science

476

hybrid approach significantly outperforms both

algorithms in terms of minimizing the number of

migrated virtual machines over the optimization

interval. With a slight deviation from the load

concentration algorithm, nearly the same

performance in terms of minimizing the total online

hours of physical servers is observed. The

computational results of the experiments are

presented in Figure 2. The 960 hours refers to the total

online hours over all servers in the considered time

interval. The obtained margin of error on the collected

results in terms of the total migrated virtual machine

ranged between (±2.07, ±5.56). While more stable

results are obtained in terms of the total initiated

migrations that ranged between (± 0.21, ±0.67) and

the total online hours (± 4.29, ±1.82).

960

295

253

272

200

400

600

800

1000

Online hours

Average (hr)

275

342

157

100

150

200

250

300

350

400

Total migrated VMs

Average (number)

18 18

Total ititiated migrations

Average (number)

Is situation Load balancing Load concentration Hybrid approach

Figure 2: Experimental result on the presented hypothesis.

5 CONCLUSION AND FUTURE

WORK

Our future work will be concentrated on finalizing a

systematic literature analysis and further presenting

taxonomy for the virtual machine live migration

problems. Our preliminary analysis in a small use-

case showed that the framework can achieve

considerable improvements in minimizing the

objective values. In addition, we are designing large-

scale experiments based on collecting different

information on the operational procedures of IT-

service providers. Moreover, we are expecting to

achieve a higher optimization potential with the

increase of the problem complexity since the

performance of the constructive approaches usually

reasonable for solving simple problems. In the

problem formulation, we addressed only the number

of migrated virtual machine as an operational

constraint, which might have an impact on the service

level agreement. Therefore, in the final experimental

analysis, we aim to address operational constraints by

IT service provider more profoundly. In addition, in

the formulated objective function we aim to address

not only the minimization of the total online hours but

also different power states of server based on

different workload levels to reduce energy

consumption.

REFERENCES

Amoretti, M., Zanichelli, F. and Conte, G. (2013),

“Efficient autonomic cloud computing using online

discrete event simulation”, Journal of Parallel and

Distributed Computing, Vol. 73 No. 6, pp. 767–776.

Avgerinou, M., Bertoldi, P. and Castellazzi, L. (2017),

“Trends in Data Centre Energy Consumption under the

European Code of Conduct for Data Centre Energy

Efficiency”, Energies, Vol. 10 No. 10, p. 1470.

Bartalos, P., Wei, Y., Blake, M. B., Damgacioglu, H.,

Saleh, I. and Celik, N. (2016), “Modeling energy-aware

web services and application”, Journal of Network and

Computer Applications, Vol. 67, pp. 86–98.

Beloglazov, A., Abawajy, J. and Buyya, R. (2012),

“Energy-aware resource allocation heuristics for

efficient management of data centers for Cloud

computing”, Future Generation Computer Systems,

Vol. 28 No. 5, pp. 755–768.

Bodenstein, C., Schryen, G. and Neumann, D. (2012),

“Energy-aware workload management models for

operation cost reduction in data centers”, European

Journal of Operational Research, Vol. 222 No. 1, pp.

157–167.

Carli, T., Henriot, S., Cohen, J. and Tomasik, J. (2016), “A

packing problem approach to energy-aware load

distribution in Clouds”, Sustainable Computing:

Informatics and Systems, Vol. 9, pp. 20–32.

Clark, C., Fraser, K., Hand, S., Hansen, J. G., Jul, E.,

Limpach, C., Pratt, I. and Warfield, A. (2005), “Live

Migration of Virtual Machines”, in Proceedings of the

2Nd Conference on Symposium on Networked Systems

Design & Implementation - Volume 2, USENIX

Association, Berkeley, CA, USA, pp. 273–286.

Delaval, G., Gueye, S. M. -K. and Rutten, É. (2015),

“Distributed Execution of Modular Discrete Controllers

for Data Center Management”, IFAC-PapersOnLine,

Vol. 48 No. 7, pp. 139–146.

Dupont, C., Hermenier, F., Schulze, T., Basmadjian, R.,

Somov, A. and Giuliani, G. (2015), “Plug4Green: A

flexible energy-aware VM manager to fit data centre

particularities”, Ad Hoc Networks, Vol. 25, pp. 505–519.

Ganga, G. M. D. and Carpinetti, L. C. R. (2011), “A fuzzy

logic approach to supply chain performance

management”, International Journal of Production

Economics, Vol. 134 No. 1, pp. 177–187.

Goiri, Í., Berral, J. L., Fitó, J. O., Julià, F., Nou, R., Guitart,

J., Gavaldà, R. and Torres, J. (2012), “Energy-efficient

and multifaceted resource management for profit-

driven virtualized data centers”, Future Generation

Computer Systems, Vol. 28 No. 5, pp. 718–731.

Han, G., Que, W., Jia, G. and Zhang, W. (2018), “Resource-

utilization-aware energy efficient server consolidation

Toward an Autonomic and Adaptive Load Management Strategy for Reducing Energy Consumption under Performance Constraints in Data

Centers

477

algorithm for green computing in IIOT”, Journal of

Network and Computer Applications, Vol. 103, pp. 205–

214.

Holland, J. H. (1992), Adaptation in natural and artificial

systems: an introductory analysis with applications to

biology, control, and artificial intelligence, MIT press.

Jeyarani, R., Nagaveni, N. and Vasanth Ram, R. (2012),

“Design and implementation of adaptive power-aware

virtual machine provisioner (APA-VMP) using swarm

intelligence”, Future Generation Computer Systems,

Vol. 28 No. 5, pp. 811–821.

Jin, H., Deng, L., Wu, S., Shi, X., Chen, H. and Pan, X.

(2014), “MECOM. Live migration of virtual machines

by adaptively compressing memory pages”, Future

Generation Computer Systems, Vol. 38, pp. 23–35.

Kaur, T. and Chana, I. (2018), “GreenSched: An intelligent

energy aware scheduling for deadline-and-budget

constrained cloud tasks”, Simulation Modelling

Practice and Theory, Vol. 82, pp. 55–83.

Keller, G., Tighe, M., Lutfiyya, H. and Bauer, M. (2012),

“An analysis of first fit heuristics for the virtual

machine relocation problem”.

Khani, H., Latifi, A., Yazdani, N. and Mohammadi, S.

(2015), “Distributed consolidation of virtual machines

for power efficiency in heterogeneous cloud data

centers”, Computers & Electrical Engineering, Vol. 47,

pp. 173–185.

Kirkpatrick, S., Gelatt, C. D. and Vecchi, M. P. (1983),

“Optimization by simulated annealing”, Science (New

York, N.Y.), Vol. 220 No. 4598, pp. 671–680.

Koomey, J. (2011), “Growth in data center electricity use

2005 to 2010”, A report by Analytical Press, completed

at the request of The New York Times, Vol. 9.

Kumar, J. and Singh, A. K. (2018), “Workload prediction

in cloud using artificial neural network and adaptive

differential evolution”, Future Generation Computer

Systems, Vol. 81, pp. 41–52.

Kumar, M. R. V. and Raghunathan, S. (2016),

“Heterogeneity and thermal aware adaptive heuristics for

energy efficient consolidation of virtual machines in

infrastructure clouds”, Journal of Computer and System

Sciences, Vol. 82 No. 2, pp. 191–212.

Kushida, K. E., Murray, J. and Zysman, J. (2011),

“Diffusing the Cloud: Cloud Computing and

Implications for Public Policy”, Journal of Industry,

Competition and Trade, Vol. 11 No. 3, pp. 209–237.

Lopez-Pires, F. and Baran, B. (2015), “Virtual machine

placement literature review”, arXiv preprint arXiv:

1506.01509.

Luo, L., Wu, W., Tsai, W. T., Di, D. and Zhang, F. (2013),

“Simulation of power consumption of cloud data

centers”, Simulation Modelling Practice and Theory,

Vol. 39, pp. 152–171.

Malekloo, M. -H., Kara, N. and El Barachi, M. (2018), “An

energy efficient and SLA compliant approach for

resource allocation and consolidation in cloud

computing environments”, Sustainable Computing:

Informatics and Systems, Vol. 17, pp. 9–24.

Marotta, A., Avallone, S. and Kassler, A. (2018), “A Joint

Power Efficient Server and Network Consolidation

approach for virtualized data centers”, Computer

Networks, Vol. 130, pp. 65–80.

Nahhas, A., Bosse, S. and Turowski, K. (2018), “Load

distribution strategies for a sustainable IT resources

management”, in Drews, P., Funk, B., Niemeyer, P. and

Xie, L. (Eds.), Multikonferenz Wirtschaftsinformatik

2018, 2018-March, Leuphana Universität Lüneburg

Institut für Wirtschaftsinformatik, Lüneburg.

Pinedo, M. (2012), Scheduling: Theory, algorithms, and

systems, 4th ed., Springer, New York.

Skiena, S. S. (1998), The algorithm design manual: Text,

Springer Science & Business Media.

Suresh, S. and Sakthivel, S. (2017), “A novel performance

constrained power management framework for cloud

computing using an adaptive node scaling approach”,

Computers & Electrical Engineering, Vol. 60, pp. 30–44.

Tchana, A., Son Tran, G., Broto, L., DePalma, N. and

Hagimont, D. (2013), “Two levels autonomic resource

management in virtualized IaaS”, Future Generation

Computer Systems, Vol. 29 No. 6, pp. 1319–1332.

Tesfatsion, S. K., Wadbro, E. and Tordsson, J. (2014), “A

combined frequency scaling and application elasticity

approach for energy-efficient cloud computing”,

Sustainable Computing: Informatics and Systems, Vol.

4 No. 4, pp. 205–214.

Vafamehr, A. and Khodayar, M. E. (2018), “Energy-aware

cloud computing”, The Electricity Journal, Vol. 31 No.

2, pp. 40–49.

Vitali, M., Pernici, B. and O’Reilly, U.-M. (2015),

“Learning a goal-oriented model for energy efficient

adaptive applications in data centers”, Information

Sciences, Vol. 319, pp. 152–170.

Wang, X., Du, Z., Chen, Y. and Li, S. (2008),

“Virtualization-based autonomic resource management

for multi-tier Web applications in shared data center”,

Journal of Systems and Software, Vol. 81 No. 9, pp.

1591–1608.

Xu, C. -Z., Rao, J. and Bu, X. (2012), “URL. A unified

reinforcement learning approach for autonomic cloud

management”, Journal of Parallel and Distributed

Computing, Vol. 72 No. 2, pp. 95–105.

Yoon, M. S., Kamal, A. E. and Zhu, Z. (2017), “Adaptive

data center activation with user request prediction”,

Computer Networks, Vol. 122, pp. 191–204.

Zhang, Q., Metri, G., Raghavan, S. and Shi, W. (2014),

“RESCUE: An energy-aware scheduler for cloud

environments”, Sustainable Computing: Informatics

and Systems, Vol. 4 No. 4, pp. 215–224.

Zheng, X. and Cai, Y. (2011), “Energy-aware load

dispatching in geographically located Internet data

centers”, Sustainable Computing: Informatics and

Systems, Vol. 1 No. 4, pp. 275–285.

Zhou, H., Li, Q., Choo, K. -K. R. and Zhu, H. (2018),

“DADTA: A novel adaptive strategy for energy and

performance efficient virtual machine consolidation”,

Journal of Parallel and Distributed Computing, Vol.

121, pp. 15–26.

CLOSER 2019 - 9th International Conference on Cloud Computing and Services Science

478