Assessment of the Environmental Impact of Applications in

Federated Clouds

Barbara Pernici

and Usman Wajid

Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milan, Italy

School of Computer Science, University of Manchester, Manchester, U.K.

Keywords: Environmental Impact of Applications, Federated Clouds, Adaptive Services, Eco-metrics.

Abstract: While in recent work the energy efficiency of cloud platforms has been emphasized and many approaches

have been proposed to reduce the energy consumption in data centres, the evaluation of the environmental

impact of applications running in cloud environments is still a research issue, as well as possible techniques

to lower their environmental footprint. In the paper the approach taken in the ECO

Clouds project towards

assessment of environmental impact of applications and the evaluation of its potential reduction based on

adaptive services is illustrated and discussed, including in the discussion the evaluation of alternative

possible uses of eco-metrics towards reducing the environmental impact of applications.

1 INTRODUCTION

Energy consumption in large scale data centres and

in clouds is a growing concern that has only recently

received necessary attention. Several techniques

have been developed to improve energy efficiency in

data centre and cloud environments (e.g. Kolodziej

et al., 2012); (Lindberg et al., 2012) and several

metrics have been proposed to assess the obtained

results (e.g. Kipp et al., 2012). In (Beloglazov et al.,

2011) a comprehensive survey is provided on the

existing approaches for achieving energy efficiency,

with particular reference to dynamic power

management approaches, both at the hardware and at

the software level. At the software level, approaches

range from management of workloads at the level of

the single servers, to overall workload management

at data centre and cloud level.

In cloud computing virtualization provides

necessary basis for not only dynamic management of

resources but also for making efficient use of

available resources while guaranteeing quality of

service. Lately research work in the area of cloud

computing has been focusing on allocating virtual

machines with the goal of improving energy

efficiency in cloud infrastructure and reducing

environmental impact of cloud computing

[Kolodziej et al., 2012); (Lindberg et al., 2012);

(Khosravi et al., 2013); (Wajid et al., 2013). On the

other hand, to carry out assessment of energy

efficiency techniques and approaches several metrics

have been proposed (Kipp et al., 2012);

(GreenPeace, 2012). In this respect, the GreenGrid

Consortium has proposed a Data Center Maturity

Model (DCMM) which, on the basis of computing,

networking, infrastructure, and management

variables enables users to determine the current

performance of data centres and identify ways to

achieve greater energy efficiency and sustainability

(Singh et al., 2011). In this work strong emphasis is

put on a better use of available resources based on

parameters such as for instance the percentage of

CPU usage, which, rather than being kept low in the

average data centre, should be increased to reach

60% on the average to achieve desired outcome

(envisioned by 2016).

Besides energy efficiency, a related stream of

research focuses on evaluating and reducing the

environmental impact of cloud computing by means

of minimizing CO

emissions. From the point of

view of environmental impact, many proposals are

based on deriving from the consumed energy the

equivalent estimation of CO

emissions. However,

in a (GreenPeace, 2012) report, several other factors

are highlighted that can also contribute towards

environmental impact of cloud computing.

In this respect, reporting of the Carbon

Utilization Effectiveness (CUE) is one of the

parameters which is starting to be considered by

256

Pernici B. and Wajid U..

Assessment of the Environmental Impact of Applications in Federated Clouds.

DOI: 10.5220/0004946802560261

In Proceedings of the 3rd International Conference on Smart Grids and Green IT Systems (SMARTGREENS-2014), pages 256-261

ISBN: 978-989-758-025-3

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

some of the public cloud providers. Another

contribution of GreenPeace report is the clean

energy index, which considers the energy mix for

generation of the utilized energy, siting, the adoption

of energy efficiency techniques, and reuse of energy

in datacentre or cloud environments.

Notwithstanding the many proposals there is not

a single agreed or standard metric or set of metrics

that clearly allow the assessment of the

environmental impact of cloud computing. In

particular, while the above-mentioned work focus on

the IT infrastructure (and in some cases also the

facility management, which is out of the scope of

this work), there is no clear indication on how

energy efficiency and environmental impact should

be evaluated at the level of application deployed or

running on cloud.

If we focus on the assessment of non-functional

parameters at application level, adaptive approaches

have been proposed ((e.g. in the S-Cube network,

(Papazoglou et al., 2010)) as a way to guarantee

quality of service in service-based systems where

services are provided in a distributed way to a

variety of users and to other service providers. In

such environments, a major emphasis is given to the

ability of assessing the context of execution, in order

to base adaptation on the available knowledge of the

context. In (Bucchiarone et al., 2012) such an

approach has been particularly studied for composed

services, where service is provided through a

combination of services to clients requesting it with

agreed Quality of Service constraints.

In this paper we describe how in the EU funded

ECO

Clouds project the assessment of

environmental impact of cloud applications is

performed, and by virtue of this assessment we

evaluate the potential benefit of an adaptive

approach that we have developed and implemented

in the project. The focus being on the IT components

of federated clouds and efficiently running

applications in the cloud, considering also the

dynamically varying energy mix in the different

sites, we show how such an approach could bring

more than an order of magnitude of improvement in

the environmental impact of cloud applications.

In addition to the discussion about the potential

of adaptive and context-aware approaches, the goal

of the paper is to put forward a discussion about the

possible uses of indicators, not limited to techniques

for automatically reducing absolute values of energy

consumption and environmental impact.

In this position paper, first, in Section 2, we

briefly introduce ECO

Clouds project and discuss

the assessment of environmental impact of

applications as carried out in ECO

Clouds and

possible approaches for evaluating the assessment.

This is followed by a brief description of adaptation

techniques used in the ECO

Clouds project for

reducing the environmental impact.

In Section 3 we evaluate the potential of using

adaptation techniques on experimental data while

discussing possible options. Finally, in Section 4 we

discuss possible approaches for assessing the

environmental impact of applications and their

potential for improvement.

The paper ends with summary remarks and

acknowledgements.

2 ENERGY EFFICIENCY AND

ENVIRONMENTAL IMPACT IN

ECO

CLOUDS

The ECO

Clouds project (http://eco2clouds.eu)

investigates strategies and develops mechanisms for

environmentally aware cloud sourcing with the aim

to reduce energy consumption and CO

emissions of

cloud applications as well as of the underlying

infrastructure.

In ECO

Clouds, assessment of environmental

impact of cloud computing is carried out by defining

a set of layered metrics. The evaluation of the

metrics is performed within an adaptive and context-

aware application deployment and management

approach discussed in the subsequent sections.

2.1 Metrics and Assessment

In ECO

Clouds, three levels for evaluation of

environmental impact are considered: physical,

virtualization, and application level. While it is not

the goal of the paper to provide details on all metrics

concerning each of the three levels (described in

detail in ECO

Clouds D3.1, 2013), we focus here

only on the evaluation of energy and CO

emissions

related parameters. In ECO

Clouds assessment and

improvements concerning these parameters relies on

the management of IT component of the cloud

infrastructure, and in particular on the computing

components.

In ECO

Clouds energy is evaluated at physical

level of cloud infrastructure, based on measurements

from PDU (Power Distribution Units), for each site

of the federated cloud. From these measures, energy

consumption is evaluated for each server or host of

each cloud site. From measurements at host level,

considering the running virtual machines and their

AssessmentoftheEnvironmentalImpactofApplicationsinFederatedClouds

257

CPU usage, the energy consumption is mapped and

evaluated for each virtual machine.

A third level introduced in ECO

Clouds concerns

the actual use of VM by applications. We assume

that one or more applications (or service) can be

executed on a VM, and, based on their execution

parameters, such as start and termination times and

CPU usage for each service, power level and energy

consumption are evaluated for each running service.

ECO

Clouds utilises a federated cloud

infrastructure composed of three geographically

distributed cloud sites located in Germany, France,

and UK (details of the characteristics of the sites are

provided in the ECO

Clouds web site), where the

energy mix (i.e. proportion of coal, nuclear, hydro

etc.) is known for each site. Such energy mix can

vary based on the time of the day and of the year.

Using the energy mix information alongside energy

consumption measure the CO

e (Carbon dioxide

equivalent) emissions can be evaluated, for all three

levels, i.e., physical, VM, and applications as a

measure for environmental impact of the underlying

cloud infrastructure and applications. In this way, it

is possible to give to users a rather precise

evaluation of the environmental impact of their

cloud applications based on above-mentioned data.

2.2 Adaptation Actions

Based on the defined metrics, their evaluation can be

the basis for establishing an adaptive control system

for reducing the environmental impact of cloud

computing.

As mentioned above, several approaches have

been proposed in the existing literature for

improving energy efficiency in the cloud (for

example, as discussed in (Beloglazov et al., 2011)

and (Lindberg et al., 2012).

In ECO

Clouds, the problem of energy

consumption and CO

emissions is addressed at the

software level, assuming that optimization at the

server and operating system level is not under

control of the cloud platform or the users.

In this respect, actions that can be applied within

the context of ECO

Clouds are at two levels i.e. VM

level and application level. This limitation is

primarily based on the use of BonFIRE platform

(BonFIRE, 2013) and its functionalities that are

made available for usage and experimentation in the

ECO

Clouds project.

Further actions could be envisioned in other

platforms; however, we briefly mention here the

principal two types of actions under consideration,

as a basis for the following discussion:

- VM level: here the main goal is scheduling of

VMs on available cloud resources with a focus

on consolidation, to be able to save energy by

possibly switching off unused machines, and

freeing less used computing resources for other

uses.

- Application level: actions inspired by dynamic

service composition approaches, VM

reconfiguration and usage management of long

running applications (such as the ones described

in (Cappiello et al., 2013)), starting, freezing,

reactivating, and terminating one or multiple

VMs on which the application is running based

on the applications’ internal and execution

context information.

3 POTENTIAL OF ADAPTATION

A first approach for a general evaluation of the

environmental impact is to evaluate the total energy

consumption and consequent CO

emissions over a

given period of time.

In ECO

Clouds, the Zabbix monitoring solution

used in BonFIRE has been extended with the

defined eco-metrics, since each cloud site in the

BonFIRE federated cloud already provides a Zabbix

server which uses Zabbix client agents to capture the

cloud resource utilization data at host and VM

levels. These Zabbix agents can be used to retrieve

energy consumption and CO

related data for two of

these layers: physical and virtual resources

(Tenschert and Gienger, 2013).

Application level metrics are evaluated within

the application deployment environment. The

general architecture is shown in Fig 1. Three cloud

sites are considered as testbeds in ECO

Clouds.

An example of display of power metrics is

shown in Figure. 2. In addition to the graphical

interface and exports in spreadsheets, a REST

interface for interacting with the application level

control environment has been developed.

In this section we base our discussion on an

analysis from the experimental data, starting from

the assumption that in the beginning there is no VM

management.

For an evaluation of the potential of considering

the execution context of applications for their

dynamic management, we consider for a rough

evaluation an average CO

e conversion factor for

each given site. A further improvement could be

gained considering also daily and seasonal variations

in the energy mix, but we do not consider this in this

paper to simplify the presentation, with the goal of

SMARTGREENS2014-3rdInternationalConferenceonSmartGridsandGreenITSystems

258

giving an approximate estimation of the possible

gains. However, the energy composition is actually

available as a monitored parameter within Zabbix,

together with the other eco-metrics defined in

ECO

Clouds project.

We assume to run applications initially on 10

separate physical servers, running with an average

power of 133 W. Due to the different energy mix in

the three countries, as reported in (ECO

Clouds,

D3.1, 2013) we can assume an average value of 85

gCO2e/kWh for France, 706 gCO2e/kWh for

Germany, and 567.17 gCO2e/kWh for UK.

Figure 1: ECO

Clouds architecture.

Figure 2: Monitored eco-metric (courtesy of INRIA).

Due to the application of VM consolidation

techniques, based on the target goals of DCMM

(GreenGrid, 2011), we can assume that the load of

the servers can raise from 20% to 60%, owing to

activating and deactivating the VMs as needs arise.

However, when running VMs in a BonFIRE

cloud, we have also to assume that an experiment

(i.e. the allocation of VMs to a user according to his

requests) is immediately started, and in addition,

since a duration for the experiment has to be

specified in advance, it is likely that the user will

have to ask for more resources and for a longer time

than needed because in case the application is not

completed at the end of the requested period, the

experiment will be stopped (and in this case will

have to be started again).

From application level it is difficult to assess the

potential reduction of environmental impact that can

be gained from the ability to manage VMs, thus

there are opportunities for user and the application to

better control and manage their allocated resources.

However, this depends on the initial configuration of

the experiment, as well as from the type of the

applications. For instance, in our experimentation

first simulations show a potential impact from 10 to

50%.

Figure 3: Adaptivity analysis vs initial configuration.

In Fig. 3 we report a first evaluation of the

potential reduction, with respect to the given initial

situation. In the figure it is clear that in cases such as

the one considered in the cloud federation, running

in three different countries, the impact of the

location is preponderant, and, combined with

dynamic resource management, the potential gain

could be on the order of two orders of magnitude.

However, some additional considerations are

necessary and will be discussed in the next section.

4 DISCUSSION ON EVALUATING

THE ENVIRONMENTAL

IMPACT OF CLOUD

APPLICATIONS

From the reported example, it becomes clear that the

simple definition of eco-metrics is not sufficient for

an assessment.

First of all, a coarse evaluation based on total

energy and CO

e emissions can be misleading in

some cases. In fact, it might lead to allocating the

load only on a few of the available servers, and only

in a single country when the energy mix is very

different in the federated cloud.

In addition, it has to be considered that reducing

power does not always imply reducing energy

per

year max

CO2e min

CO2e

max

energy

reduction

max

CO2e

reduction

initial

configuration 11650800 8225464.8 990318 ‐ 0.88

server

consolidation 3883600 2741821.6 330106 0.67 0.96

application

1941800 1370910.8 165053 0.83 0.98

AssessmentoftheEnvironmentalImpactofApplicationsinFederatedClouds

259

consumption (see, for instance the discussion in

(Chen et al., 2011)). In fact, applications might take

longer to execute, so the reduction in energy

consumption is not necessarily proportional to the

reduction of power, and in some situations might

even result in an increase of power consumption.

Alternative approaches can also be considered,

based on the available metrics, to take into

consideration other factors. One aspect is that

considering only energy and CO

e could result in

violations of other parameters, for instance Quality

of Service. To avoid this, assessment of

environmental impact should be based on the

evaluation of many factors. As in service-oriented

systems, this approach can result either in

considering a set of thresholds to be satisfied, or in

defining a utility function for a global assessment of

the system. In this case, other factors could be

considered, such as the importance given by the

users to some quality parameters or for a given

energy mix (e.g., excluding coal or nuclear sources).

Since the evaluation of importance could vary based

on the user preferences, or even based on the

specific nature of each single application, the user

should be enabled to establish his own preferences

for making an evaluation.

Other factors, even if not relevant at the moment

within the BonFIRE platform as it is currently

provided, could be linked in the future to variable

costs of the provided service under different

conditions such as the use of specific infrastructure

at specific time, under specific QoS parameters and

under specific energy consumption or CO

footprint

related constraints.

These considerations would result in different

evaluations when considering the user views and

cloud provider view, which would give a different

emphasis on different parameters. A comparison in

such cases might be difficult, as well as an

assessment of the final result.

A completely different approach could be based

on management considerations, such as avoiding

critical events (as proposed, for green IS, in (Reiter

et al., 2013)). Critical events in cloud computing,

when focusing on energy and environmental impact,

could be the need to avoid reaching power limits in a

data centre, which usually results in limited service

or in additional costs for energy provisioning, or

using on-demand resources in one of the sites of the

federated cloud when other resources are available

instead in other sites.

Finally, giving the users information about the

energy consumption and related monitored

information could raise awareness of ecological

implications of their computing requirements, which

in turn could encourage modification in user

behaviour. For instance, if a user needs to report

e emissions for his applications, and he sees a

potential for reduction while running the application

in a given site or in a given time of the day, he might

schedule to run his applications, if not time critical,

taking also these factors into consideration. In other

cases, the user might realise that his use of allocated

VMs is not optimal (in terms of energy consumption

or CO

e emissions) in a specific deployment

configuration and that different configurations might

reduce wastage of available computing resources

e.g., by consolidating certain VMs in a specific host,

exploiting better multicore architectures, or

interleaving I/O intensive applications with

computing intensive ones. In this respect, awareness

of energy consumption or environmental impact of

computation can encourage users to change their

behaviour and consider different factors in

application deployment and execution decision

making model.

5 CONCLUDING REMARKS

In the paper we have presented the ECO

Clouds

approach to monitoring energy consumption and

emissions as eco-metrics in a federated cloud

environment. We have discussed the potential for

exploiting the use of eco-metrics in different

directions, ranging from the evaluation of

environmental impact of cloud computing to

encouraging the changes in behaviour of users.

ECO

Clouds will also develop sophisticated

mechanisms for automatic optimization of

application deployment configurations based on

context-aware service provisioning in the cloud.

Optimization will be carried out both at the

scheduler level and at the application level during

the execution lifetime of applications.

Experimentation in the project will be aimed at

testing different types of application case studies,

ranging from computing intensive to more

interactive and cloud specific long running

applications, to analyse how different factors could

have an impact on approaches for assessment of

environmental impact and create new opportunities

for optimization of energy consumption and CO

footprint or for a more informed user behaviour that

can lead to further developments and innovation in

this area.

SMARTGREENS2014-3rdInternationalConferenceonSmartGridsandGreenITSystems

260

ACKNOWLEDGEMENTS

This work has been partially supported by the

ECO

Clouds project (http://eco2clouds.eu) and has

been partly funded by the European Commission's

IST activity of the 7th Framework Program under

contract number ICT- 318048. This work expresses

the opinions of the authors and not necessarily those

of the European Commission. The European

Commission is not liable for any use that may be

made of the information contained in this work. The

authors thank all the participants in the project for

common work and discussions.

REFERENCES

Beloglazov, A., Buyya, R., Choon Lee, Y., Zomaya, A.Y.,

2011. A Taxonomy and Survey of Energy-Efficient

Data Centers and Cloud Computing Systems.

Advances in Computers 82: 47-111

BonFIRE, 2013. http://www.bonfire-project.eu/ link

checked on Dec. 29, 2013

Bucchiarone, A., Cappiello, C. Di Nitto, E., Barbara

Pernici, Sandonini, A., 2012: A Variable Context

Model for Adaptable Service-Based Applications.

IJARAS 3(3): 35-53

Cappiello, C., Datre, S., Fugini, M. G., Gienger, M.,

Melia’, P., Pernici, B. 2013. Monitoring and Assessing

Energy Consumption and CO

Emissions in Cloud-

based Systems, IEEE International Conference on

Systems, Man and Cybernetics (SMC)

Chen, D., Henis, E., Kat, R. I., Sotnikov, D., Cappiello, C.,

Ferreira, A. M., Pernici, B., Vitali, M., Jiang, T., Liu,

J., Kipp, A., 2011. Usage centric green performance

indicators, Proceedings of GreenMetrics 2011

(SIGMETRICS workshop).

ECO

Clouds, 2013. D3.1: Layered Sets of Metrics and

Application Profile, ECO

Clouds project deliverable,

http://eco2clouds.eu

Greenpeace, 2012. How clean is your cloud?,

http://www.greenpeace.org/international/en/publicatio

ns/Campaign-reports/Climate-Reports/How-Clean-is-

Your-Cloud/ link checked on Dec. 29, 2013

Khosravi, A. Garg, S. K., Buyya, R., 2013. Energy and

Carbon-Efficient Placement of Virtual Machines in

Distributed Cloud Data Centers. Euro-Par 2013, 317-

328

Kolodziej, J., Khan, S. U., Wang, L., Zomaya, A. Y.,

2012. Energy efficient genetic-based schedulers in

computational Grids. Concurrency and Computation:

Practice and Experience

Kipp, A., Jiang, T., Fugini, M. G., Salomie, I., 2012.

Layered Green Performance Indicators, Future

Generation Comp. Syst. 28(2), 478-489

Lindberg. P. et al., 2012. Comparison and analysis of eight

scheduling heuristics for the optimization of energy

consumption and makespan in large-scale distributed

systems. In Journal of Supercomputing, Vol. 59, Issue

1, 323-360.

Papazoglou, M. P., Pohl, K., Parkin, M., Metzger, A.

(Eds.), 2010. Service Research Challenges and

Solutions for the Future Internet - S-Cube - Towards

Engineering, Managing and Adapting Service-Based

Systems. Springer, Lecture Notes in Computer Science

S-Cube Team, http://www.s-cube-network.eu/, link

checked on Dec. 29, 2013

Reiter, M., Fettke, P., Loos, P., 2013. Towards a

Reference Model for Ecological IT Service

Management, International Conference on

Information Systems, Milan

Singh, H. Reuters, T. (eds), 2011. Data Center Maturity

Model, http://www.thegreengrid.org/en/Global/Conte

nt/white-papers/DataCenterMaturityModel Dec. 29,

2013

Tenschert, A., Gienger, M., 2013. Cloud Federation

Monitoring for an Improved Eco-Efficiency,

eChallenges 2013, Dublin, Ireland

Wajid, U., Marín, C. A, Mehandjiev, N., 2013. Optimizing

Service Ecosystems in the Cloud, in The Future

Internet, Lecture Notes in Computer Science, Volume

7858, Springer.

AssessmentoftheEnvironmentalImpactofApplicationsinFederatedClouds

261