ENERGY AWARENESS NEEDS A RETHINKING

IN SOFTWARE DEVELOPMENT

Hagen H¨opfner

and Christian Bunse

Mobile Media Group, Bauhaus-University of Weimar, Bauhausstr. 11, 99423 Weimar, Germany

Software Systems, University of Applied Sciences of Stralsund, Stralsund, Germany

Keywords:

Software engineering, Data processing, Energy awareness.

Abstract:

Energy efﬁciency and -awareness are buzzwords in various areas of information and communication technol-

ogy and are at the core backbone of GreenIT. Computing centers aim for reducing energy consumption in order

to save money and carbon dioxide emissions. Furthermore, GreenIT labels are perfect selling points for com-

puter equipment. Especially battey-powered and mobile devices must consider software’s energy consumption

in order to prolong their uptime, while keeping the desired or agreed quality of service (QoS). Even if energy

awareness regarding hardware has been researched intensively for a couple of years, the analysis of the impact

of software on energy consumption is rather novel. We claim that the development of energy-efﬁcient and

-aware software systems require a careful re-examination of the many paradigms in software development.

1 MOTIVATION

The history of software engineering shows that in-

novation is often stipulated by antedecent improve-

ments in the hardware sector. In the “good old

days” developers had to understand registers, CPU-

opcodes, or interrupt handling in order to develop

professional software systems. They had to optimize

source- and assembly code in order to improve per-

formance and quality. Interestingly, this is still true in

some domains (e.g., automotive) where resources are

scarce. In the following years, software became more

and more complex and development paradigms like

object-orientation, or model-driven and component-

oriented software engineering allowed for distinct and

clearly separated views regarding hardware and soft-

ware kiwi-optimization. kiwi is an acronym for “kill

it with iron” and describes the approach to solve per-

formance problems with additional resources or hard-

ware. Unfortunately, kiwi does not solve every opti-

mization problem regarding the non-functional prop-

erties of a software system. Reliability, safety or se-

curity may need substantial adaptations of the system

and not just new, additional resources.

A a rule of thumbs it that more complex software

requires more energy. However, functional require-

ments demand a certain complexity. So, making soft-

ware less complex is not a proper approach for reduc-

ing energy consumption. Developers and researchers

must rethink many paradigms in software develop-

ment if they want to reduce software dependent en-

ergy consumption. In our research we addressed var-

ious hypotheses on how to develop energy-efﬁcient

software. This position paper will brieﬂy introduce

our ideas and support our opinion that energy aware-

ness requires a rethinking in software development.

2 ALGORITHMS

For the last couple of years, we taught our students

that various implementation alternatives for certain

problems exist and that algorithms that need less

CPU-cycles are usually the better choice. Complex-

ity theory, which aims on describing the boundaries

of the number of required CPU-cycles or operation

steps as a function over the number of input data,

formed the basis for the performance prediction of

an implementation alternative. If we look, e.g., on

different sorting algorithms, we prefer Mergesort to

Insertionsort because the complexity of Mergesort is

O(n· log(n)) while the complexity of Insertionsort is

O(n

). On the one hand, both algorithms can be used

for sorting an input list with n elements and Merge-

sort needs less comparison steps than Insertionsort.

On the other hand, Mergesort is a recursive algorithm

that reduces the number of comparisons through a di-

294

Höpfner H. and Bunse C..

ENERGY AWARENESS NEEDS A RETHINKING IN SOFTWARE DEVELOPMENT.

DOI: 10.5220/0003578002940297

In Proceedings of the 6th International Conference on Software and Database Technologies (ICSOFT-2011), pages 294-297

ISBN: 978-989-8425-77-5

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

vide and conquer strategy. Hence, it requires more

memory than Insertionsort. In previous works we an-

alyzed the energy consumption of different sorting

(Bunse et al., 2009a) and join algorithms (H¨opfner

and Bunse, 2010a) and showed that there is no sta-

tistical correlation between energy consumption and

computational complexity Even if Insertionsort takes

more time to ﬁnish its task, it consumes less energy.

Further research (cf. (Bunse et al., 2009b)) has

shown that energy consumption can be reduced by

switching between different algorithms based on their

energy efﬁcient operational range and on the input

data size. However, if it comes to energy-aware soft-

ware, algorithm designers and programmers must for-

mally understand the energy consumption of an algo-

rithm: The simple assumption that faster algorithms

require less energy as they ﬁnish faster is wrong!

Thus, we have to ﬁnd a complexity measure that, sim-

ilar to computation complexity, allows us to specify

boundaries for an algorithm’s energy consumption.

We have to accept, that especially for battery-powered

devices, processing a limited amount of data, slower

but energy-efﬁcientalgorithms are preferable to faster

but energy intensive alternatives.

3 RESOURCE SUBSTITUTION

Almost all computers, smartphones, laptops, embed-

ded systems, etc. use a Von-Neumann-Architecture

(von Neumann, 1993): Data and program code are

stored in main memory and a CPU accesses them

over a bus system. Ubiquitous availability of net-

work accesses enables information exchange and task

distribution among devices with different built-in re-

sources. They form clouds that are either used to store

data or to distribute workload. Thus, various hard-

ware resources with different energy consumptions

are utilized for data processing and storage. However,

certain resources can be substituted by other ones (cf.

(H¨opfner and Bunse, 2007)). Several projects ana-

lyzed such resource substitution strategies with regard

to energy consumption. The authors of (Marwedel,

2007) state that CPU usage needs comparatively less

energy than memory storage. The authors of (Vei-

jalainen et al., 2004) state that compressing a ﬁle by

more then 10%, transmitting and decompressing it re-

quires less energy than transmitting it uncompressed.

According to the authors of (Kansal and Zhao, 2008),

ﬁle compression also reduces the energy consumption

of operations on hard disks.

Substituting resources only reﬂected/reﬂects the

capacity of the involved resources. Compression

was/is used to save storage memory or communica-

tion afford. View materialization in database man-

agement systems reduces the number of executions

of CPU-intensive view deﬁnition queries by stor-

ing/materializing the queries results. Task distribu-

tion like the well known SETI@home helped/helps

to realize computations that require additional CPU-

cycles. However, only a few researchers tried to re-

think the idea of resource substitution while focussing

on energy awareness. In (H¨opfner and Bunse, 2010b)

we started to summarize these works and started to

deﬁne a classiﬁcation scheme for energy related as-

pects of a resource substitution strategy.

4 INFORMATION QUALITY

The previousely discussed issues focussed on energy

aspects while maintaining the quality of processed

information. However, certain application scenar-

ios (e.g. mobile information systems (Veijalainen

and Gross, 2002)) allow for trading information qual-

ity off energy consumptions (i.e., graceful degrada-

tion). A similar idea has been introduced in terms

of lossy compression for reducing the size of multi-

media ﬁles: One trades the quality of an image, a

movie, or an audio ﬁle off its ﬁle size. Especially

in networked environments, this approach is com-

mon as users accept lower quality to a certain extent.

With regard to energy consumption, one has to adapt

this idea. Of course, transmitting smaller ﬁles needs

less energy than transmitting larger ones, but this is-

sue was already addressed. In addition, energy-aware

lossy compression algorithms might help to ﬁnd a bal-

ance between information quality (quality of a video

stream, etc.) and energy consumption.

The authors of (H¨uls, 2002) found out that the en-

ergy consumption of an MPEG-algorithm is largely

affected by the used data types. By replacing

double

float

, the algorithm’s energy consumption was

reduced to 34%. Of course, using less precise num-

bers leads to a drastic reduction of the resulting qual-

ity of the MPEG-ﬁle. Our own experiments on sort-

ing algorithms (H¨opfner and Bunse, 2010a) show that

although data sizes were simply doubled (2-Byte in-

teger was replaced by 4-Byte ﬂoating point numbers),

the algorithm requires signiﬁcantly more energy. In-

terestingly, the differences are not simply correlated

with a factor of two. Beneath additional memory re-

quirements, one reason for this growth might also be

the software emulated realization of the ﬂoating point

unit of the processor used for this experiment.

The conclusion for this section is twofold. On

the one hand, energy efﬁciency strongly depends on

the correct choice of data types. This issue is al-

ENERGY AWARENESS NEEDS A RETHINKING IN SOFTWARE DEVELOPMENT

295

ready reﬂected per default in software engineering

and database design, as proper data types should be

chosen anyway. On the other hand, compromises on

the information quality allow to use less complex data

types and therefore to reduce energy consumption.

However, we are not aware of any systematic research

on the user acceptance in such a scenario.

5 SOFTWARE ENGINEERING

Energy awareness and energy optimization strategies

have to be integrated into the processes and methods

of software development. This requires means for

specifying energy related properties as well as pre-

diction and optimization approaches.

A literature review has shown that quantitative de-

scriptions of system dependability are generally done

via combinations of attributes: Reliability, availabil-

ity, safety, integrity, conﬁdentiality and maintainabil-

ity. In modern component based software develop-

ment (CBSD), these attributes are validated by mod-

eling the system components using binary fault states

and by combining the fault states to an overall sys-

tem state using Boolean logical function. The fault

states themselves are usually modeled as probabilis-

tic faults due to component wear out or other proba-

bilistic sources. In practice, a variety of techniques,

languages and tools are used for modeling both hard-

ware and software. Examples are SysML, Modelica,

Matlab, FMEA/FMECA, RBD, FTA, Markov chains,

and Petri-nets. However, these techniques do not ad-

dress the challenges of the GreenIT-era, nor do they

always follow sound engineering principles.

Speciﬁcation and modeling of software compo-

nents is a much less mature discipline. At code level,

there are three well-known component models: Mi-

crosofts .NET, Suns Enterprise Java Beans and the

OMG’s CORBA Component Model, each with its

own interface deﬁnition language (IDL). However,

IDL speciﬁcations are purely syntactic and do not

contain a semantic description of a components func-

tionality. At “model” level, the UML2.0 component

model offers a rich speciﬁcation of components in

terms of various UML diagrams, but much of the se-

mantic information is missing. The semantically rich-

est speciﬁcation languages are currently associated

with web services and service-oriented architectures.

The Web Service Description Language provides syn-

tactic information like the component IDLs. How-

ever, a new suite of semantic service speciﬁcation

languages such a WSDLS, OWLS and METEORS

have recently been developed to address this prob-

lem. None of the aforementioned technologies pro-

vides a development process, and although there are

several methods, that aim to support speciﬁc aspects

of CBSD (such as component identiﬁcation), no one

is integrated in mainstream development processes.

For many years, non-functional requirements

were written textually as part of the overall system

requirements documents. However, with the increas-

ing use of hardware and software components and the

commoditization of computing power via grids, the

speciﬁcation of the QoS requirements has gained in-

creasing attention. At the hardware level, QoS has

long been speciﬁed in terms of bandwidth and fault

probability. At the software level, several languages

have recently been introduced for the speciﬁc pur-

pose of modeling and describing QoS requirements.

Moreover, the web service technology also includes

mechanisms speciﬁcally for deﬁning QoS require-

ments and policies such as WSPolicy. To date, how-

ever, there are no mainstream methodologies focusing

on the creation and use of energy-related QoS speci-

ﬁcations. Interestingly, research on specifying the re-

sources of software systems (Weilkiens, 2008; Pare-

dis et al., 2010), (Graf et al., 2006; Thomas et al.,

2008) have their focus on of platform models and per-

formance. Energy consumption is widely neglected.

To overcome these problems and to also address

the problem of optimization, another pillar of en-

gineering energy-aware software systems, we devel-

oped MARMOT (Bunse et al., 2007), a method that

facilitates reuse in embedded software system de-

velopment. MARMOT is a component-based and

model-driven development framework. Composition

is a key activity in component-based development

with MARMOT. Within MARMOT, the energy con-

sumption of a software system can either be statically

or dynamically, and one can optimize it at develop-

ment or at runtime (Bunse and H¨opfner, 2008).

A ﬁnal aspect that is currently beyond the scope of

our research is the prediction of the effects of changes

or more complex actions such as software composi-

tion. Regarding quality properties such as reliabil-

ity or maintainability, there exist already tools (cf.

(Becker et al., 2007; Happe et al., 2010a; Happe et al.,

2010b)). Regarding energy ﬁrst projects (Kounev,

2011; Rathfelder et al., 2010) have been started but

did not provide any results yet. In addition to predic-

tion, engineering also requires support for optimiza-

tion. Optimization can either take place at develop-

ment or at runtime. At development time energy con-

sumption can be improved by adapting system mod-

els, implementations, and infrastructures (Siegmund

et al., 2010). At runtime a system might adapt itself

(Grassi et al., 2009) according to external needs and

requirements, such as energy consumption.

ICSOFT 2011 - 6th International Conference on Software and Data Technologies

296

6 SUMMARY AND OUTLOOK

We claimed that developers and researchers have to

rethink many paradigms in software development if

they want to reduce software-dependent energy con-

sumption. We illustrated that, so far, almost all well

accepted-approaches starting from complexity theory

issues, over resource substitution issues and informa-

tion quality compromises, to the software engineer-

ing process do not properly support energy aware-

ness. We presented some of our ﬁndings as well as

research conducted by other researchers. However,

all presented results need more generalization in or-

der to form the basis and a profound framework for

energy aware software development. The authors of

this paper welcome all interested researchers to start

the discussion about these issues.

REFERENCES

Becker, S., Koziolek, H., and Reussner, R. (2007). Model-

Based performance prediction with the palladio com-

ponent model. In WOSP’07, pages 54–65, New York,

NY, USA. ACM.

Bunse, C., Groß, H.-G., and Peper, C. (2007). Applying

a Model-based Approach for Embedded System De-

velopment. In EUROMICRO’07, pages 121–128, Los

Alamitos, CA, USA. IEEE.

Bunse, C. and H¨opfner, H. (2008). Resource substitu-

tion with components — optimizing energy consump-

tion. In ICSOFT’08, volume SE/GSDCA/MUSE,

pages 28–35, Set´ubal, Portugal. INSTICC.

Bunse, C., H¨opfner, H., Mansour, E., and Roychoudhury, S.

(2009a). Exploring the Energy Consumption of Data

Sorting Algorithms in Embedded and Mobile Envi-

ronments. In MDM’09, pages 600–607, Los Alamitos,

CA, USA. IEEE.

Bunse, C., H¨opfner, H., Roychoudhury, S., and Mansour, E.

(2009b). Choosing the “best” Sorting Algorithm for

Optimal Energy Consumption. In ICSOFT’09, vol-

ume 2, pages 199–206. INSTICC.

Graf, S., G´erard, S., Haugen, Ø., Ober, I., and Selic, B.

(2006). Modeling and Analysis of Real-Time and Em-

bedded Systems. In Satellite Events at MoDELS’05,

volume 3844 of LNCS, pages 58–66, Berlin / Heidel-

berg. Springer.

Grassi, V., Mirandola, R., and Randazzo, E. (2009). Model-

Driven Assessment of QoS-Aware Self-Adaptation. In

Software Engineering for Self-Adaptive Systems, vol-

ume 5525 of LNCS, pages 201–222, Berlin / Heidel-

berg. Springer.

Happe, J., Becker, S., Rathfelder, C., Friedrich, H., and

Reussner, R. H. (2010a). Parametric Performance

Completions for Model-Driven Performance Predic-

tion. Performance Evaluation, 67(8):694–716.

Happe, J., Groenda, H., Hauck, M., and Reussner, R. H.

(2010b). A Prediction Model for Software Perfor-

mance in Symmetric Multiprocessing Environments.

In QEST’07, pages 59–68, Los Alamitos, CA, USA.

IEEE.

H¨opfner, H. and Bunse, C. (2007). Ressource Substitution

for the Realization of Mobile Information Systems.

In ICSOFT’07, volume SE, pages 283–289, Set´ubal,

Portugal. INSTICC.

H¨opfner, H. and Bunse, C. (2010a). Energy Aware Data

Management on AVR Micro Controller Based Sys-

tems. ACM SIGSOFT SE Notes, 35(3).

H¨opfner, H. and Bunse, C. (2010b). Towards an energy-

consumption based complexity classiﬁcation for re-

source substitution strategies. In Proc. GVDB’10, vol-

ume 581 of CEUR Workshop Proc. CEUR-WS.org.

H¨uls, T. (2002). Optimizing the energy consumption of an

MPEG application. Master’s thesis, TU Dortmund,

Fakult¨at f¨ur Informatik, Dortmund, Germany.

Kansal, A. and Zhao, F. (2008). Fine-grained energy pro-

ﬁling for power-aware application design. SIGMET-

RICS Performance Evaluation Review, 36(2):26–31.

Kounev, S. (2011). Self-Aware Software and Systems Engi-

neering: A Vision and Research Roadmap. In SE’11,

Nachwuchswissenschaftler-Symposium.

Marwedel, P. (2007). Embedded System Design. Springer.

Neumann, J. V. (1993). First Draft of a Report on the

EDVAC. IEEE Annals of the History of Computing,

15(4):27–75.

Paredis, C. J., Bernard, Y., Burkhart, R. M., de Koning, H.-

P., Friedenthal, S., Fritzson, P., Rouquette, N. F., and

Schamai, W. (2010). An Overview of the SysML-

Modelica Transformation Speciﬁcation. In INCOSE

Int. Symposium 2010.

Rathfelder, C., Klatt, B., Brosch, F., and Kounev, S. (2010).

Performance Modeling for Quality of Service Predic-

tion in Service-Oriented Systems. In Handbook of

Research on Non-Functional Properties for Service-

Oriented Systems: Future Directions. IGI Global.

Siegmund, N., Kuhlemann, M., Pukall, M., and Apel, S.

(2010). Optimizing non-functional properties of soft-

ware product lines by means of refactorings. In Va-

MoS’10, volume 37 of ICB-Research Report, pages

115–122. Uni Duisburg-Essen.

Thomas, F., G´erard, S., Delatour, J., and Terrier, F. (2008).

Software Real-Time Resource Modeling. In Embed-

ded Systems Speciﬁcation and Design Languages, vol-

ume 10 of LNEE, pages 169–182. Springer, Berlin /

Heidelberg.

Veijalainen, J. and Gross, T. (2002). Mobile Wireless Inter-

faces: In Search for the Limits. In Developing an In-

frastructure for Mobile and Wireless Systems, volume

2538 of LNCS, pages 153–163, Berlin / Heidelberg.

Springer.

Veijalainen, J., Ojanen, E., Haq, M. A., Vahteala, V.-P., and

Matsumoto, M. (2004). Energy Consumption Trade-

offs for Compressed Wireless Data at a Mobile Termi-

nal. IEICE ToC, E87-B(5):1123–1130.

Weilkiens, T. (2008). Systems Engineering mit SysML/UML

— Modellierung, Analyse, Design. dPunkt.Verlag,

Heidelberg, 2nd edition.

ENERGY AWARENESS NEEDS A RETHINKING IN SOFTWARE DEVELOPMENT

297