Until now the target for energy savings by computers has been the
hardware. But hardware is only part of the story, perhaps even the
smaller part, as new a EU project is showing. Even with energy-efficient
hardware, much of the potential for energy savings is wasted by the
drain that inefficient software places on systems.
The
ENTRA project, which runs
until September 2015, has come up with an experimental software
prototype based on programming semantics that tells programmers how much
energy will be consumed as a result of the code they are writing. The
ENTRA tool runs alongside the program and, through code analysis and
energy modelling, shows how much the code is going to cost the computer
in terms of energy use, and what the impact of design decisions on
energy use will be.
‘Compare it to the fuel efficiency of a car,’ said ENTRA coordinator
Professor John Gallagher, of
Roskilde University
in Denmark. ‘We think of that being the property of the car. You buy a
car and it is supposed to consume 8 litres per 100 km or whatever, but
of course that depends on the way you drive it. The same is true for
computing. You think of energy being associated with the hardware, but
it seems that potentially more energy is savable at the software level
with the same hardware, rather than putting a lot of effort into
developing more energy-efficient hardware. That’s where the ENTRA
project comes in.’
ENERGY USE VISIBLE TO PROGRAMMER
The ENTRA project focuses on energy efficiency as a design goal and
aims to bring energy efficient products by IT companies - quicker to
market. The ENTRA tool shows energy use in terms of Watts (power
consumption) or in absolute energy requirements (the energy needed to
finish the task) depending on the speed of the processor (GHz). Rather
than having to wait until the program is installed and running on a
machine and then measuring its energy (often too late to do anything
about it), the programmer gets an early picture from the energy
predictions in the code.
The energy consumption of computers has become a major concern, both
environmentally in terms of carbon footprint, and practically through
the need for greater battery life. The huge growth in cloud computing
over the past decade has meant data centres have become very large
consumers of energy. Power is needed not just for cooling, but also to
process and transport data. The ENTRA approach can help both with the
programming and with scheduling of tasks to minimize energy.
BIG SAVINGS ACHIEVABLE FOR SMARTPHONE BATTERIES
As the demand for computing power increases exponentially, software
applications will need to be designed in an energy-efficient way to
allow smaller batteries to be charged less frequently.
With an eye on meeting these challenges, the ENTRA prototype is
being tested on three typical energy-hungry components: real-time audio
processing, robot and motor control, and real-time networking. The
software defining them is being run on hardware at
XMOS,
a semiconductor company in Bristol, the United Kingdom, which produces
for the automotive, computing and games industries. Researchers in the
project are seeing in these case studies that energy savings could be at
least 20-50 percent, e.g. by compacting programs, and optimizing timing
of operations to permit lower processor speeds.
XMOS plans to add ENTRA components for energy optimization into its
tool-chain. The ENTRA results are also being exposed to internationally
leading companies such as ARM and ST Microelectronics, and world leading
academics in energy-efficient computing at the EACO (Energy Aware
COmputing) workshops.
Scaling up ENTRA’s results to system level is feasible, said
Professor Gallagher. “The whole approach is independent of any
particular hardware or application domain in the sense that it is based
purely on the programming language semantics and a general energy model.
So we could apply the same techniques to high performance computers,”
he pointed out.
The ENTRA project, consisting of 4 partners in Denmark, the UK and
Spain, began in October 2012, ends on 30 September 2015, and receives
FP7 funding of 2.1 million euros.
Link to project's website