Testing Testing Testing

Architectural Framework for
Energy Optimization in Embedded Systems

Hasan Sözer, Arjan De Roo, Mehmet Akşit
Software Engineering Group,
University of Twente,
Enschede, The Netherlands
{sozerh, roo, aksit}@ewi.utwente.nl
Abstract account. For example, a component needs a certain amount

Energy consumption has become one of the important sys- of power under a given speed. In essence, this is an optimi-
tem properties that should be controlled by embedded con- zation problem known as multi-objective optimization
trol software. There is usually an inherent trade-off be- (MOO) [1]. There exists a vast literature on quality meas-
tween energy consumption and several system qualities. As urement/estimation, trade-off and optimization techniques.
such, optimization techniques should be adopted for mak- However, research efforts mainly focus on mathematical
ing the desired trade-off among quality attributes. Imple- problems in this domain rather than the application of the
mentations of these techniques are usually ad-hoc and sys- devised solutions in real software systems. The realization
tem-specific, leading to implicit design decisions distrib- of these solutions for embedded systems is usually done at
uted over several software components. We propose an the implementation phase of the software. The solution is
architectural framework for custom synthesis of control distributed over several components, and possibly imple-
software from reusable and programmable elements. The mented by different engineers. This ad-hoc approach results
goal is to facilitate systematic reuse of knowledge in the in several custom implementations and implicit design de-
optimization domain and explicit management of quality cisions. However, the design decisions regarding the qual-
trade-offs for energy optimization of embedded systems. ity trade-offs are critical for an embedded system and they
should be made explicit at the software architecture design
Keywords [2]. In addition, there is a well-established theory and a set
Energy consumption, embedded systems, optimization of solution techniques for MOO, which can be systemati-
cally reused. Therefore, we propose a flexible, architectural
INTRODUCTION solution in terms of a framework that comprises reusable
Software has become an integral part of today’s embedded solutions for energy optimization in embedded systems.
systems. For example, overall system operations and prop- The framework aims at facilitating i) systematic reuse of
erties in digital document printing systems are mainly con- knowledge in the optimization domain, ii) explicit man-
trolled by software. Energy consumption has recently be- agement of quality trade-offs with respect to the energy
come an important issue and as such, it has become as one consumption, and iii) custom synthesis of MOO solutions
of the important system properties that should be controlled from reusable and programmable artifacts. In the remainder
by the embedded software. However, there is usually an of the paper, we discuss our solution approach and related
inherent trade-off between energy consumption and several research directions.
system qualities. For instance, customers demand minimal
energy consumption for printing systems. However, they THE APPROACH
also demand high productivity, which requires running the Our approach is presented in Figure 1. The architectural
machine at a higher speed, and which in turn increases the framework consists of three main elements i) reusable arti-
energy consumption. Moreover, the print quality should facts that can be tuned and composed together with tool
also be kept at acceptable levels, which requires running support, ii) an architecture synthesis method which guides
the machine at a lower speed. Therefore, embedded control the software engineer for utilizing these artifacts, and iii) a
software must take multiple, conflicting objectives into set of analysis tools, which provide feedback with respect
account and choose the optimal values for control parame- to the quality attributes of the design instantiated for a par-
ters (decision variables), while keeping energy consump- ticular application. The analysis process takes a set of pro-
tion below the allowed limits. Examples of control parame- vided quality requirements and constraints into account. As
ters are the power given to a component and the speed of long as these requirements and constraints are not met, the
the system. In addition, the software should employ adap- software engineer makes iterations to improve the design.
tive control to retain optimal system qualities under varying Once the design is complete, (partial) code generation is
circumstances such as changing context/environment, us- possible since the design elements are based on predefined
age profile and evolution of system properties. The con- reusable artifacts.
straints for control parameters should also be taken into
Figure 1. The Overall Approach
As the basic requirements, the framework should provide which is implicit in the current approach, is the integration
the following: of the developed control software with the system.
• Support for the formal specification of quality re-
quirements and constraints including the ones re- ACKNOWLEDGMENTS
lated to energy consumption. The ideas presented in this paper are based on our work
• Architectural styles and a uniform notation that that has been carried out as a part of the OCTOPUS project
[3]. This project is managed by the Embedded Systems
enables the instantiation and specification of soft-
Institute and it is partially supported by the Netherlands
ware architectures.
Ministry of Economic Affairs.
• As a part of the notation, support for expressing
energy and quality related properties/semantics REFERENCES
used for automatic generation of formal models as [1] K. Deb, Multi-Objective Optimization using Evolu-
inputs for simulation and analysis tools. tionary Algorithms, Wiley, UK, 2001.
• Extensible artifacts enabling the incorporation of [2] A. de Roo, H. Sozer and M. Aksit, “An architectural
additional optimization and trade-off techniques, style for optimizing system qualities in adaptive em-
when needed. bedded systems using Multi-Objective Optimization”,
In Proceedings of the 8th Working IEEE/IFIP Confer-
FURTHER RESEARCH DIRECTIONS ence on Software Architecture, pages 349-352, Cam-
Each of the basic requirements, as indicated above, can be bridge, UK, 2009.
extended with sub-questions. For instance, the notation for [3] Octopus project, Embedded Systems Institute, [online]
specifying the architecture can support multiple views. http://www.esi.nl/projects/octopus, Eindhoven, The
Some of these views would require special annotations Netherlands, 2010.
and/or elements for energy consumption. Similarly, formal
models and analysis techniques needed to be extended for
the estimation of energy consumption. Another issue,

Testing Testing Testing

Uploaded by

Copyright:

Available Formats

Testing Testing Testing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Testing Testing Testing

Uploaded by

Copyright:

Available Formats

Architectural Framework for

Energy Optimization in Embedded Systems

Abstract account. For example, a component needs a certain amount

You might also like