Green Computing is becoming critically important. However, the energy efficiency problem of computing is a multidimensional one and tough to tackle as a whole. Different Information Communication Technology (ICT) sectors and specific application constraints can yield a demand for energy-efficient computing for various reasons. For example, the increase in battery-dependent mobile systems, the large energy-related operational cost of big data centers, and the negative impact on the environment due to the growing Carbon Dioxide (CO2) emissions of ICT are all urging the scientific community for more energy efficient computing.
The long period of achieving further performance gains through Moore’s law came to an end. Increasing the silicon density and at the same time scaling up the operating frequency is no longer an option due to overheating and quantum effects. Quantum effects will make silicon transistors unreliable. Overheating affects the reliability of computation, decreases the life of electronic components, and makes computing more expensive as advanced cooling technologies are needed. Although the shift to multi-core solutions and three-dimensional stacked silicon allowed for further performance gains, the overheating problem remains a significant limiting factor.
The replacement of the old, bulky desktops and laptops with smaller, more energy efficient devices such as smartphones and tablets, contributes to the reduction of the ICT energy consumption. However, new energy efficiency challenges emerged. Mobile computing sparked the rapid growth of the cloud online services that serve billions of mobile devices. Almost every mobile device application involves some interaction with the cloud. This includes video streaming, news feeding, social networks usage, software updates, etc.
Looking at the cloud
Greenpeace released a report in 2014 on the contribution of cloud computing to climate change. Figure 1, represents the 2020 forecast for the carbon footprint of the two main cloud infrastructure elements; data centers and telecoms.
A similar figure, figure 2, represents the 2020 energy consumption forecast for the same cloud components. According to these figures, the cloud carbon footprint will at least double and the cloud energy consumption will at least triple, from 2007 to 2020.
The research concluded that the energy consumed by the ICT sector is much larger than previously estimated, on the scale of around 70%. Another study conducted in 2013 concluded that more than 10% of the world’s energy consumption is spent on computing and that ICT’s consumes 50% more than the aviation sector. Therefore, ICT’s carbon footprint is becoming a threat to the environment. This can no longer be ignored.
Is smaller more energy efficient?
The dramatic shift towards mobile computing comes with another great energy-related challenge. Mobile computing depends on limited portable energy sources; mainly batteries. Figure 3.a shows the trend of the performance improvement for the iPhone devices released between 2011 and 2016. The figures are based on the GeekBench-4 multi-core benchmark. Figure 3.b shows the standby time trend, retrieved from here, for the same range of devices found in figure 3.a. While the performance of the devices has been growing significantly, the standby time growth has not been able to make substantial gains over the same period of time. Standby time captures the battery life while the device is probably in its lowest power consumption mode. Therefore, when a device is under normal usage, the time a battery can keep the device alive is dramatically reduced compared to the standby time. This demonstrates the disproportional technological advancements between the mobile computing performance and the battery technology. Chemistry constraints limit the energy densities of electrochemical batteries. Experts in the area, have been struggling for a breakthrough for decades.
To mitigate the above problem, the energy efficiency of mobile devices has been mainly improved by hardware innovation. It is a remarkable achievement that the battery life of a mobile device, such as the iPhone, is kept at the same levels over a six years period, while significantly improving the device performance. However, there is still a lot to be done to expand the operational time of a mobile device; the time before recharging is needed. Any increase in the devices operational time will significantly improve the overall user experience. Therefore, energy efficiency is one of the most important selling points of a portable electronic device, so much so that it can determine the market success or failure of the device.
And what about the IoT?
The challenge is even bigger when it comes to energy-critical deeply embedded applications. In this case, an energy-related failure will not just cause an inconvenience to the user. Instead, it can have severe consequences. For example, a health-monitoring device that fails to deliver crucial medical data due to inadequate energy budget can cause the loss of human life.
After pervasive computing transformed into today’s IoT, the number of energy-critical deeply embedded applications has increased dramatically. The majority of the computation performed by such applications depends on limited or unreliable sources of energy, such as energy harvesting. Striving for energy efficiency became the primary goal for electronic system engineers.
Is energy-efficient hardware the only weapon we have?
Traditionally, for any power/energy challenge in the Information ICT, hardware innovation has been the safe heaven to achieve energy savings. Similarly, hardware innovation is currently the prominent response to tackle the energy challenge that IoT faces. The hardware community has been introducing new ultra-low-energy embedded devices and customizing existing technologies to create more energy efficient versions, such as the Bluetooth Low Energy, which are well suited for IoT energy-critical applications. But is that all we can do to tackle this challenge? Steve Furber, the principal designer of the ARM microprocessor, in one of his article in 2010 stated:
If you want an ultimate low-power system, then you have to worry about energy usage at every level in the system design, and you have to get it right from top to bottom, because any level at which you get it wrong is going to lose you perhaps an order of magnitude in terms of power efficiency.
Software has the ultimate control over hardware. Inefficient software can drive energy-efficient hardware to waste the system’s energy budget. Custom software that is well suited for a particular platform is always more resource efficient than more generic software versions which perform the same set of tasks. Therefore, focusing solely on optimizing the energy consumption of hardware is only part of the story. Then, in the same article, Steve Furber continues with:
Programmers will not be able to afford to be ignorant about the energy cost of the programs they write ... You need tools that give you feedback and tell you how good your decisions are. Currently the tools don’t give you that kind of feedback.
These tools are now needed more than ever to overcome the ICT energy challenge. However, there is still too little done to expose how different coding styles and algorithms affect the energy consumption on hardware. Therefore, very few programmers at present have much of an idea of how much energy their programs consume, and which parts of a program use the most energy.
A new shift in the history of computing is needed once more. One that will look at energy-efficient software to tackle the increasing energy consumption of the ICT. This shift might be harder than the shift from sequential to parallel computing, but it is of the same importance.
This article is derived from several of my published articles. If you consider citing this article in a research paper, please consider citing my related publications:
A dissertation submitted to the University of Bristol in accordance with the requirements of the degree of Doctor of Philosophy Philosophy in the Faculty of Engineering. Department of Computer Science, University of Bristol, March 2017. [ pdf | bib ]
Kyriakos Georgiou, Samuel Xavier-de-Souza and Kerstin Eder.
IEEE Embedded Systems Letters, vol. PP, no. 99, pp. 1-1, 2017. [ bib | DOI | http ]