Perspectives in Science (2016) 8, 611—613

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.elsevier.com/pisc

VLSI systems energy management from a

software perspective — A literature survey夽
Prasada Kumari K.S. ∗, Sampath Kumar, Shivaji Sinha

Dept of Electronics Engineering, JSS Academy of Technical Education, C-20/1, Sector 62, Noida, UP 201301,
India

Received 20 February 2016; accepted 11 June 2016

Available online 4 July 2016

KEYWORDS Summary The increasing demand for ultra-low power electronic systems has motivated
research in device technology and hardware design techniques. Experimental studies have
Software energy;
proved that the hardware innovations for power reduction are fully exploited only with the
Embedded system;
proper design of upper layer software. Also, the software power and energy modelling and anal-
Green energy;
ysis — the first step towards energy reduction is complex due to the inter and intra dependencies
Modelling;
of processors, operating systems, application software, programming languages and compilers.
Power awareness
The subject is too vast; the paper aims to give a consolidated view to researchers in arriving at
solutions to power optimization problems from a software perspective. The review emphasizes
the fact that software design and implementation is to be viewed from system energy con-
servation angle rather than as an isolated process. After covering a global view of end to end
software based power reduction techniques for micro sensor nodes to High Performance Com-
puting systems, specific design aspects related to battery powered Embedded computing for
mobile and portable systems are addressed in detail. The findings are consolidated into 2 major
categories — those related to research directions and those related to existing industry prac-
tices. The emerging concept of Green Software with specific focus on mainframe computing is
also discussed in brief. Empirical results on power saving are included wherever available. The
paper concludes that only with the close co-ordination between hardware architect, software
architect and system architect low energy systems can be realized.
© 2016 Published by Elsevier GmbH. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Introduction
夽 This article belongs to the special issue on Engineering and Mate-
Traditionally, power reduction meant hardware power
rial Sciences. reduction in VLSI systems in portable and mobile bat-
∗ Corresponding author.
tery operated systems. With faster processors, cheaper
E-mail addresses: [email protected]
memories and vast choice of programming languages and
(P. Kumari K.S.), [email protected]
(S. Kumar), [email protected] (S. Sinha). development support, today’s software intensive systems

http://dx.doi.org/10.1016/j.pisc.2016.06.035
2213-0209/© 2016 Published by Elsevier GmbH. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).
612 P. Kumari K.S. et al.

Software Energy Management

Source code

Analysis Front End

Low level
High level Middle level O.S services
Source code Machine
Scheduling, I/O, User Intermediate code
(In High level Instructions
Interface, Memory
language (Compiler/
Management, Process &File Back End
/Assembly Assembler
management Routines/files
level Synthesis
in various languages and generated)
language)
forms
Machine code

(a) System Elements (b) compiler structure

Figure 1 Functional overview of system blocks related to software.

need a thorough analysis of the software dependent energy VLSI based Embedded, portable and mobile systems where
reduction schemes in overall system power management. the major concern is battery Ampere hour rating. For the
Although energy consumption is dictated by hardware, soft- sake of completeness a section on Green software for larger
ware dictates the way the energy is consumed. Initiative computing installations is also included.
to reduce power should first identify the sources of power,
techniques to model and means to measure. Literature
covers software energy management from different per- Research findings
spectives with respect to software layers, application areas
and system architectural complexities in isolation. But there Processor instructions abstract processor architecture and
is lot of interplay between software units and a consolidated research on software energy centred on instruction sets orig-
view is essential in realizing energy efficient systems. An inated in the 1990s by embedded system developers with
insight into the role of software in the overall power sav- upper layers treated as black boxes (Tiwari et al., 1994).
ing techniques would help designing systems optimized for Further, source code optimization for compiler construc-
power. Fig. 1 shows (a) the functional overview of the soft- tion along with compilation techniques evolved. Compilers
ware blocks related to Software Energy Management and (b) which reduce bus switching activity through instruction
compiler model relevant in the following discussions. re-ordering/scheduling, reduced memory operations and
The survey is based on articles published by jour- memory hierarchy utilization improved energy efficiency by
nals/books by academicians and white papers published by almost 14%. Choice of algorithm, data type and data struc-
industry giants like Intel, IBM, Microsoft, etc. till December ture for coding in addition to optimization techniques like
2015. ‘Related terminologies’ section gives an overview of loop unrolling, software pipelining and recursion free soft-
the related terminologies. ‘Literature survey’ section deals ware designs were found highly effective (Simunic et al.,
with survey of literature, observations and discussions based 2000; Mehta et al., 1997). Parallelism at run time and com-
on the survey. ‘Conclusions and future research’ section cov- pile time combined with power aware scheduling techniques
ers conclusions and future research areas. aiming reduced switching activities between instructions
was also proved to be effective from energy point of view.
Related terminologies The strong interaction of the middle layer operating sys-
tems with the application software, hardware, peripherals
Though terms Energy (Joules) and Power (Watts) are used and users necessitated detailed energy models for each
interchangeably in literature, energy is relevant in situa- of these components (Mera and Santiago, 2010; Belleudy
tions with battery operated systems where the system has et al., 2012). Power optimization of O.S in terms of O.S ser-
to run for longer duration and at the same time energy vices thus emerged in due course with strong influence of
is a constraint. In cases where energy (power*execution application software features and machine level features.
time) constraints are not experienced, instantaneous power Experiments proved that that Processor idle state manage-
consumption is the prime issue. The design objective of ment saves energy from 11% to 25% depending on application
lowpower/energy system is to minimize power whereas sys- (Kambadur and Kim, 2014).
tem power awareness built into systems enables the system Power efficient design of high level application software
to minimize energy consumption by adapting to changes in at source code level had also been extensively studied.
operating conditions. Both concepts are relevant for man- Algorithm design eliminating complex dynamic memory
aging power. allocations, reduced code size, data type choice (fixed,
floating, integer, double, etc.) is proved to be effective
Literature survey (Kambadur and Kim, 2014). With multi processors and multi
cores dominating the computing systems, parallel program-
Discussions are broadly classified into those related to aca- ming techniques enhance energy efficiency (Kambadur and
demics and R&D and those related to Industry. Prime focus is Kim, 2014). The increased demand for mobile systems and
VLSI systems energy management from a software perspective 613

internet of things furthered focused research in com- context, underlying hardware and the overall system func-
bined effects of O.S and application software elements tionalities. In a nut shell, the co-ordinated concentrated
(Liu et al., 2015; Crowcroft and Vallina-Rodriguez, 2013). efforts of system architect, hardware architect and soft-
Much Research in Wireless Sensor Networks and Body Area ware architect alone will bring out energy efficient systems.
Networks demanding energy aware low power software The tight interplay between energy-oriented hardware and
development solutions is also contributing to the ultra-low energy-aware software calls for innovative structural, func-
energy sector. tional and mathematical models for analysis and design.
Innovations in the above areas are highly supported by Model based Software engineering practices supported by
embedded systems hardware — software co — design tools, appropriate tools will definitely accelerate development of
reconfigurable computing tools, power monitors, simulators modern complex VLSI systems operating under severe energy
and profilers from chip vendors and academic researchers. constraints.

The industry track Conflict of interest

Industry has adopted research results and established poli- None declared.
cies and best practices in bringing out products with low
energy consumption. The pioneers in this field are Intel,
Microsoft, Apple, Nokia, H.P and Fujitsu. They have pub- References
lished a series of white papers on the topic (Steigerwald
and Agarwal, 2011). Intel categorized power reduction tech- Belleudy, B., Ouni, C., Senn, E., 2012. 15th Euromicro Conference on
niques emphasizing active and idle application behaviour Energy Characterization and Classification of Embedded Operat-
along with platform level context awareness and hard- ing System Services , pp. 684—691.
Bener, A.B., Morisio, M., Miranskyy, A., 2014. Green software. IEEE
ware acceleration. Microsoft Best Practices for energy
Software 31 (May—June (3)), 36—39.
efficiency addresses power management under 2 major Crowcroft, N., Vallina-Rodriguez, J., 2013. Energy management
categories — operating system directed and application techniques in modern mobile handsets. IEEE Commun. Surv.
software/services related. Literature on efforts by Apple, Tutor. 15 (1), 179—198.
Nokia, H.P, and O.S vendors like Android are also available. Dickov, B., Pericas, M., Carpenter, P., Navarro, N., Ayguade, E.,
2014. Software-managed power reduction in infiniband links.
HPC and the green software engineering concept ICPP, 311—320.
Hurson, A., 2013. Green and Sustainable Computing — Part 2. Aca-
demic Press, USA, pp. 199—237.
For large computing systems which are highly software Kambadur, M., Kim, M.A., 2014. An experimental survey of energy
oriented, data oriented and I/O oriented, techniques and management across the stack. ACM SIGPLAN Notices 4 49
guidelines for green software development is important (October (10)), 329—344.
(Hurson, 2013; Bener et al., 2014). Energy reduction of up Liu, Y., Xu, C., Cheung, S.-C., 2015. Diagnosing energy efficiency and
to 30% with negligible increase in execution time related performance for mobile internetware applications. IEEE Soft-
to HPC interconnection network is already reported (Dickov ware 32 (January—February (1)), 67—75.
et al., 2014). On large-scale HPC systems where multiple Mehta, H., Owens, R.M., Irwin, M.J., Chen, R., Ghosh, D., 1997.
applications are running concurrently, scheduling proved to Techniques for low energy software. In: 1997 International Sym-
posium on Low Power Electronics and Design, Proceedings,
be an important aspect from energy point of view. Algorithm
18—20 August, pp. 72—75.
choice and Programming techniques like LU factorization, Mera, D.E., Santiago, N.G., 2010. Low power software techniques
vectorisation, etc. for scientific computing reduces power for embedded systems running real time operating systems. In:
(Bener et al., 2014; Dickov et al., 2014). As a rule of thumb, MWSCAS, pp. 1061—1064.
since each line of code is ultimately translated to machine Simunic, T., Benini, L., De Micheli, G., Hans, M., 2000. Source code
cycles, reducing code size saves power. Finally, power anal- optimization and profiling of energy consumption in embedded
ysis and measurement tools (RAPL from Intel and tools systems. In: The 13th International Symposium on System Syn-
from Fugitsu, etc.) help power reduction based on empirical thesis, 2000, Proceedings, pp. 193—198.
results. Steigerwald, B., Agarwal, A., 2011. Software& Services Group,
INTEL Corporation Developing Green Software.
Tiwari, V., Malik, S., Wolfe, A., 1994. Power analysis of embed-
Conclusions and future research ded software: a first step towards software power minimization.
IEEE Trans. Very Large Scale Integr. Syst. 2 (December (4)),
From the above discussions, it is evident that energy/power 437—445.
management has to be analyzed with reference to the