People of ACM - Carla Ellis

May 17, 2016

Traditionally, many have believed that energy consumption in a mobile device is determined by its hardware, such as circuitry or system architecture. A tenet of the Milly Watt project was that the needs of applications should drive the hardware design process. What recent developments or findings give you confidence that applications can play a more central role in regulating energy consumption in mobile devices? Why do you think power consumption in mobile devices has not been a measure of performance in the eyes of engineers or computing professionals?

No matter how energy efficient each individual hardware component may be (and the relative efficiency among components may shift with future innovation), there are always tradeoffs in how they are used. Should we cache, compute, or communicate to achieve a result? One recent development that illustrates an application-appropriate systems-level solution for energy savings involves offloading computation to a nearby (one wireless hop away) cloudlet with cached state as an alternative to either on-device computation or remote execution on a centralized cloud service (as discussed recently in a GetMobile Retrospective, A Brief History of Cloud Offload: A Personal Journey from Odyssey through Cyber Foraging to Cloudlets, by Mahadev Satyanarayanan).

I don't agree that mobile computing engineers and developers can still ignore power consumption as a measure of performance. Successful mobile devices and apps must deliver at least acceptable battery lifetimes for user satisfaction. You rarely hear users complain of battery lifetimes that are too long!

I think one danger lies in the assumptions about what is "good enough" based on a designer's life experience. A developer, living in a city with pervasive high-bandwidth wireless access and habitual overnight charging, may make very different decisions about the needed lifetime of the battery and the stress placed on the networking system than a user might actually experience, say, vacationing on Washington State's Olympic Peninsula, where I live. I often encounter frustrated hikers in the Olympic National Park, trying to use GPS-based navigation or nature field guide apps that assume ubiquitous network connectivity (which doesn't exist there). Batteries can die very quickly searching for maps or data that are only available from the (unreachable) cloud rather than being locally stored on the device.

Looking ahead, how will the increasing use of mobile devices, cloud computing, sensor networks and the Internet of Everything impact power consumption considerations in these devices?

All of these disparate technologies and their interplay will present an amazingly rich space of power/energy-related opportunities and challenges, manifesting in a myriad of different ways. Whether the role of energy/power within an application is treated as a constraint (battery life of wireless devices, sensitivity to heat in embedded devices and sensors, or power costs of data centers), as an objective (automated control of the use, distribution, and harvesting of energy), or as a vulnerability (threat of supply disruptions), there is no doubt of the importance of understanding these issues. The devil is in the details.

You have advocated that power conservation in all computing devices (mobile and stationary) will be good for the environment. You have also argued that computing can help advance sustainability in other ways. Can you discuss the other ways in which computing can be a positive force for environmental sustainability?

Sustainability in my everyday life means reducing my personal carbon footprint. Most of my energy consumption comes from heating our home, heating water, and recharging our plug-in car. Computing is key to building smarter homes—understanding and controlling a house's energy use. Upon retiring and moving to Washington State, we built as energy-efficient a home as our site and the current building technology allowed, and it has become our lab. We instrumented the house with an energy monitoring system to understand how it was performing. The exciting thing was that we immediately recouped our investment in monitoring hardware by discovering and fixing inefficiencies in the configuration of both our hot water system and our ground-source heat pump that might not have been obvious without such data, since they were delivering heat—just not as efficiently as intended. More work is needed in computational analysis and visualization of such monitoring data to guide the typical homeowner or electric car driver.

The focus of your work with Systers, as well as CRA-W, has been to encourage and support the participation of women in computing. How has the participation of women in the field changed since you began your career? Looking forward, what are the most effective approaches to increasing the number of women entering the computing field?

Since I got actively involved in encouraging the participation of women in computing, the percentage of degrees earned by women, from Bachelor’s to PhDs, has gone up and down and still hovers around only 20% nationwide. The lack of growth in these numbers is disturbing. However, there are bright spots where considerable progress is being made.

I am especially proud of my involvement in CRA-W. CRA-W has experience with programs that have track records of over 20 years, and the organization has made a commitment to rigorous evaluation of our programs to learn about which interventions work and why they work. With CRA-W's focus on encouraging women to pursue computing research careers, it's exciting to be able to provide data that show, for example, that participants in our undergraduate research mentoring programs are significantly more likely to enroll in a PhD program than other equally qualified undergraduate students. The keys to success appear to be meaningful mentoring relationships, especially with female role models, and building/belonging to a community of women for mutual peer-level support. These efforts must often transcend institutional boundaries, since the number of local women may be too small. Our Grad Cohort Workshop is an example of a program offering such benefits.