Q&A: Scaling Up

The University of California, Berkeley’s Eric Brewer has covered a lot of ground in his 20-year career. He was among the earliest to recognize the need for large-scale Web services, building scalable servers with clusters of commodity nodes and laying the foundation for contemporary cloud computing. He co-founded Inktomi, a search engine startup that peaked at $241 per share and $300 million in annual revenue in 2000 before collapsing as clients like Exodus filed for bankruptcy. (It was sold to Yahoo! in 2003.)

He has also been deeply involved with Information and Communication Technologies for Development, spearheading projects to bring telemedicine to Indian villages and develop long-distance Wi-Fi networks in rural areas. In May, he began yet another chapter with a two-year assignment at Google, where he is working on developing the company’s next-generation infrastructure.

For the past 10 years, you’ve been involved with a number of computing projects that benefit developing countries. Tell us about your recent work in that domain.

One of the things we’re working on is building a low-cost GSM base station that’s appropriate for rural villages. Rural connectivity is expensive. Base stations take a lot of power, so you need a big diesel generator. Then you need to bring diesel to the generator, which means you need a road—often, you’ve got to build it—and you need trucks to bring the diesel to the generator. On top of that, if you’re building a road, you probably want to be in a relatively flat area, and you need a very tall tower to get coverage.

The base station we’re building takes only roughly 50 watts, which means it can be run on solar or wind power and can be located up on a hill, in a place that has good visibility to villages.

Does it leverage your previous work on low-cost, long-distance Wi-Fi?

Our previous work in solar solutions and long-distance Wi-Fi are both very relevant as they greatly reduce the cost of the power system and backhaul solution; we use long-distance Wi-Fi instead of microwave links to backhaul the traffic into an urban area that has relatively low-cost bandwidth.

What inspired you to tackle the project? In the past, I understand you were less convinced that cellular connectivity is the best solution for rural areas since Wi-Fi is cheaper to implement.

The strong urban success of cellular means that many rural folks have phones, even if their village does not have coverage. Some use these phones when in coverage, others use them as FM radios, and still others have them mostly as a status symbol. Nonetheless, the demand for rural cellular is very clear, and the phones are often already there and waiting.

Much of your work is done through Technology and Infrastructure for Emerging Regions, or TIER, a research group you founded at the University of California, Berkeley. How did TIER get started?

The biggest influences on the founding of TIER came out of Inktomi. First, I was traveling quite a bit, and I’d been invited to the World Economic Forum, where I had the privilege of meeting a wide variety of very sharp people from developing nations. Many of these folks were articulate about the problems in their country, and almost all the time my reaction was that technology had a role to play in solving them. And another factor was that Inktomi had done so well that I was, at least on paper, extremely wealthy, and starting to think more seriously about addressing some of these problems. Of course, I don’t have that money anymore, so I decided to focus on solutions via research by creating a community within computer science that could address these great challenges.

What’s your process for finding new projects?

I tend to prefer infrastructure problems. I like to have at least half of my students working on core infrastructure, things like connectivity and power in particular. I also tend to look in health care, where it’s relatively easy to get good problem definitions and good deployment coverage.

Is it difficult to get funding?

Funding the equipment isn’t difficult, because the equipment isn’t very expensive. It’s harder to fund the research. Research funding was doing better before the 2008 financial crisis, but I am optimistic about relationships with foundations and nonprofits, which are more into these things than they used to be. We’re also starting to see interest from nontraditional parts of the government, like the U.S. Department of State.

What about your wireless hypothesis, which posits that it’s more useful to provide communications and computing capabilities to developing nations than more traditional infrastructure?

In terms of the percentages, I think there are signs that the wireless hypothesis is coming true—that countries that have cellular infrastructure are getting things like roads, too. It’s hard to know what it means, but it’s certainly correlated, and we’ll probably know in another 10 years or so.

What’s your take on the larger ICTD community?

I’m very happy with our progress. There’s the ACM SIGDEV [ACM’s Symposium on Computing for Development, Univ. of London, Dec. 17–18, 2011) that’s coming, and ACM India has interest in this space. There are also several conferences that cover different disciplines, as well as workshops from many different fields—for AI, for networking and systems, for HCI. This is the right model, because we’re trying to solve problems that require many disciplines. So people in this space have both a community position and a position in which they specialize and teach and do traditional work in their discipline.

How can scientists balance those two roles?

It varies by discipline. I would say that HCI is the easiest because they have a long history of looking at their users as part of the focus of the domain. For other fields that are perhaps more narrowly defined, I would say these projects tend to be about 20% new technology and 80% other stuff. You need a nugget and a strong insight for your domain, and then you have to do all this other work like understanding the problem, maybe even discovering the problem, and trying something in the field, because that’s the litmus test. It’s unpredictable how things are going to work.

---

"Computer science needs to be a part of almost every discipline now, and it’s not clear to me that computer scientists have stepped up to that role yet."

---

Earlier this year, you began a two-year stint at Google. How did that happen?

I was doing some consulting with Google, and in particular with Google. org, which is its philanthropic arm. Some of the senior Google management asked me about my opinion on some things, and I guess they wanted my opinion on a lot more things, which led to an offer I couldn’t refuse.

What are you working on?

I’m looking quite broadly at ways that Google can improve its own infrastructure to make it more innovative in the long term. There’s great hardware coming, and things like flash storage that really change some of the equations. Looking long term, I would also like to see how we can provide the cloud to another billion or two billion people. There’s bandwidth coming into Africa at unprecedented levels, particularly to east Africa, because of the undersea cables. We still have to figure out how to get it inland, and we need to figure out how to build and operate mobile phone and cloud-based services in developing countries. Right now there are very few data centers in Africa. You end up having to go to Europe and the U.S., and that’s a long way to go for every object on a Web page.

You’ve been involved with a good range of environments throughout your career, from startups to academia to traditional industry.

I like both academia and industry for different reasons. There are certain things where academia is a better place to have an impact, where things are a little longer term or where there’s not a clear market yet. There are other places where I prefer industry, especially when you want to get something from an idea stage to affecting a billion people. So I will continue to cross that line back and forth.

What are some of the things you think the field still needs to work on?

One thing that needs more thought is how to make computer science a good player in multi-disciplinary research. Computer science needs to be a part of almost every discipline now, and it’s not clear to me that computer scientists have stepped up to that role yet. It’s not easy to do. Tenure cases are still based on a single discipline. Funding at NSF is single discipline. As I mentioned earlier, a lot of projects end up being 20% technology and 80% other stuff—so you need reviewers that respect that other stuff and understand what’s hard or valuable about it.

I imagine you’ve learned a lot about interdisciplinary research through your ICTD projects.

I’ve learned that it’s hard. It’s harder for faculty than for grad students. Here at Berkeley, we’ve been able to train a generation of graduate students that really know both social science and computer science. It’s much easier to learn in grad school, when you have the time. I’ve been learning as I go—learning from my students, from colleagues, sometimes sitting in on classes. But the future of multi-disciplinary research will be through students who have been trained in multiple disciplines.

Back to Top