People of ACM - Ramón Cáceres

September 5, 2024

Your recent work at Google involved developing and maintaining privacy infrastructure for more than a billion users. Will you tell us a little more about what this role involved? And what are the challenges of building and operating such large systems?

I was a software engineer on Zanzibar, Google’s global authorization system. Zanzibar constitutes critical privacy infrastructure at planetary scale. It controls access to trillions of data objects managed by products used by more than a billion people every day, for example Calendar, Drive, Maps, Photos, and YouTube. It guarantees external consistency while serving millions of authorization requests per second with 95th-percentile latency under 10 milliseconds and availability above 99.999%.

In my seven years on the Zanzibar team, I improved the system’s scalability, reliability, and security as its workload and use cases grew. I also served on the on-call rotation that responded to production issues 24x7. Finally, I led the publication of an experience paper that has inspired a number of authorization systems outside Google.

Along with co-authors Mahadev Satyanarayanan (Carnegie Mellon University), Paramvir Bahl (Microsoft Research), and Nigel Davies (Lancaster University) you received the 2022 ACM SIGMOBILE Test of Time Award for co-authoring the 2009 paper “The Case for VM-Based Cloudlets in Mobile Computing.” How did this paper change the state of the art in cloud computing at the time?

In our paper we identified shortcomings of the established cloud computing model when serving interactive mobile applications that handle large volumes of data generated by end devices. An example would be performing vision tasks on a video stream from a mobile camera to help users with cognitive disabilities interpret what surrounds them. For such applications, sending the data to a faraway cloud computing center consumes more network resources and incurs higher latency than processing the data close to the devices that produce the data and to the users who consume the results. At the same time, mobile devices often do not have enough computing or battery power to do the work by themselves.

To address these shortcomings we introduced Cloudlets, computing resources deployed near the edge of the network to support resource-intensive and latency-sensitive applications. The award citation reads: “What is known as edge computing today can be traced back to this Cloudlets paper.” Edge computing has found broad use in the Internet of Things and other technology areas. It remains an active research topic and is a growing industry segment with a current market size estimated in the tens of billions of dollars. Our paper has garnered more than 4,800 citations.

What emerging technology or technologies will be especially impactful in shaping mobile computing in the next five to ten years?

The union of mobile computing, artificial intelligence, and privacy preservation is becoming increasingly important. It has recently become possible to do significant machine learning on mobile devices themselves, for example for speech and facial recognition. Another emerging technology is federated learning, where instead of sending raw training data to a central location, each device performs a portion of the learning task on the data it collects locally. This federated architecture has strong privacy-preserving properties, and I believe we’ll see a lot more of it in coming years.

You’ve served on the board of the Computing Research Association’s Committee on Widening Participation (CRA-WP). Could you describe that work?

Since I was in graduate school in the 1980s, I’ve been involved in efforts to increase the participation of women and others from groups that are underrepresented in computing. It’s well documented that having broader representation in the field leads to technology that serves more people and serves them better.

With CRA-WP, I’ve co-chaired a series of mentoring workshops that bring together roughly 150 graduate students from underrepresented groups with more than 30 senior researchers from academia and industry. Taking place over several days of talks, panels, and one-on-one sessions, the workshops have proven successful in strengthening in the students the feeling that they belong in computer science and have the tools to succeed in their careers.

Improving diversity at all levels of the profession is an ongoing struggle. We’ve made some progress but there’s a lot more work to do.

To learn more, listen to the recent ACM ByteCast featuring Ramón Cáceres, where he discusses the development of his career and his research interests.