|
 About Berkeley
|
|
 |
|
| Intel Research Berkeley Collaborating to Change the World |
|
At a small industrial lab in downtown Berkeley, California, researchers from Intel and UC Berkeley
are collaborating in a novel experiment that is designed to generate breakthrough results. Since the
lab was launched in November 2001, it's research has made a major impact on the world. The
secret? An open and collaborative research model pioneered by Intel.
TABLE OF CONTENTS |
|
RELATED DOCUMENTS & LINKS
Perspectives
Intel Research Berkeley Staff
|
|
 |
Carlos Guestrin - Senior Researcher |
UC Berkeley Admin, Faculty, Students
|
Joyce Freedman - Asst. Vice Chancellor |
|
Jitendra Malik - Chair, Dept. of EECS |
|
Michael Franklin - Professor, CS |
|
Scott Shenker - Professor, CS |
|
Ion Stoica - Asst. Prof, CS |
|
Ryan Huebsch - Ph.D. student, CS |
|
Sam Madden - Asst. Professor, CSEE, MIT (former UCB Ph.D. student.) |
Interview
|
A. Richard Newton - Dean of the College of Engineering, UC Berkeley |
|
|
|
On the 13th floor of the PowerBar building in downtown Berkeley, California, researchers are dreaming up new ways to deliver energy and meet performance challenges. But these scientists aren't focused on the latest protein bar or sports nutrition drink. They are designing a novel networking infrastructure that will enable computing and communications to reach remote regions of the developing world.
Welcome to Intel Research Berkeley-one of four labs in the Intel Research Network, an innovative approach to industry-university research pioneered by Intel to enhance and accelerate long-term research. Researchers flow freely between the lab and UC Berkeley, conducting joint research projects under a model designed to foster collaboration and generate breakthrough results. Since its opening in November 2001, the Berkeley lab has made an enormous impact.
"The Intel Research laboratory near the UC Berkeley campus epitomizes the kind of open collaborative research that makes a difference to us all, by tackling critical problems of society, through IT solutions. The interests and capabilities of Intel and Berkeley are brought together in the lab in a way that maximizes unfettered research and explores potential commercial applications of academic inquiry. The relationship is governed by an open collaborative research agreement that, since it was developed in 2001, has spawned additional open collaborative IT arrangements. Four years after its inception, the lab continues to produce tangible results in wireless networking that fulfill our public service mission and demonstrate the utility of open collaboration principles in action. The lab is a vital resource to our campus and we value our partnership with Intel tremendously."
Beth Burnside
Vice Chancellor
University of California, Berkeley
|
|
|
|
The open and collaborative research (OCR) model practiced at Intel’s university labs was designed to eliminate the conflicts over intellectual property (IP) rights that hinder or prevent many collaborations between companies and universities. “Our model provides a certain freedom that you can’t get with a more proprietary approach,” says Eric Brewer, Director of Intel Research Berkeley. “Research can take its natural course, since researchers don’t have to worry about IP. A more closed model, even if it’s well intentioned, slows down the research process. And it can produce a chilling effect: just worrying about potential violations of IP rights prevents work from getting done.”
Under the OCR model, Intel’s university labs operate on the principles of collaboration and non-exclusive IP rights. Intel owns and funds the labs, but much of their research is published and widely shared. Patents may be issued but are expected to be rare, because of the OCR model’s emphasis on collaboration, not competition. “The fact that Intel stresses sharing of results makes it very easy to collaborate with our lab,” says Brewer.
“Initially, I didn’t realize the extent to which Intel was willing to pursue open research. Because of its open and collaborative model, Intel was able to recruit very high-level academic researchers as lab directors. This has enabled the labs to capture a much larger mind share of academic departments than any of the other research groups that I’ve seen. That, to me, was the brilliant insight.”
Scott Shenker
Professor of Computer Science, UC Berkeley
Group Leader of Networking, International Computer Science Institute (ICSI)
|
The idea of foregoing IP rights is difficult for some companies to embrace, but the long-term benefits can be substantial, according to Kurt Brown, Co-Director of the Berkeley lab. “There seems to be a knee-jerk reaction by many companies that if we create new research, the IP should be held really tightly because we can make money off of it,” he says. “But Intel looks beyond the immediate commercial potential. We may not make money on the lab’s research directly, but the research might create a new class of computing that that could help us in the longer term.”
Intel’s open and
collaborative approach provides the company a window into the future. “Our
collaboration with UC Berkeley under this open research arrangement helps
us to see things on the horizon earlier than we might have if Intel had
built a traditional industrial research lab,” says Brown. “That is
extremely valuable.”
|
|
|
Both Intel and the university benefit from the proximity of Intel Research Berkeley to the UC Berkeley campus. “We have a large number of students flowing back and forth between the lab and the university,” says Brewer. “It’s this flow of students that’s the most critical form of sharing. The primary means of collaborating with the university is via students; they’re the mechanism that really makes it work.”
Working with researchers across the street from campus can be a richer experience than collaborating with colleagues across town or across the country. The interchange of ideas is more fluid; student researchers can easily stop by the lab to ask questions, discuss a difficult research problem, or work on a project. Lab researchers can meet with their university colleagues on campus or at a local café to talk about their work. These casual interactions, which require close physical proximity, make the collaboration much more interactive, relaxed, informal, and productive.
"It's always great to have a lab with first-rate researchers close by; it provides more opportunities for collaborations and for interesting work. We promote the fact that Intel Research Berkeley is close to campus when recruiting faculty and students. The lab is a big asset to us."
Ion Stoica
Assistant Professor of Computer Science, UC Berkeley
Faculty Affiliate Researcher, Intel Research Berkeley
|
|
|
|
For UC Berkeley researchers, there’s an additional benefit of collaborating with the Intel lab, according to Kurt Brown. “One challenge for academic departments and individual professors is that it’s very difficult to scale up a research project beyond a small group of graduate students,” he says. “The amount of grant writing required to pull in and coordinate multiple institutions is extremely challenging.” Intel Research Berkeley makes it easier for the university to scale its research by providing access to the resources of Intel.
”There are certain projects that need scale if they are to achieve their goals,” says Brewer. He points to the TIER project, which aims to bring technology to developing regions, as one example. “The work that we are doing at UC Berkeley and our lab on technology for developing regions is helped tremendously by the fact that Intel has an entire platform dedicated to emerging regions,” he says. “We are also working with product groups within Intel that care about developing regions, to learn from them about the needs of people in those regions and provide them with new technology ideas. That kind of access to Intel is critical, and it’s the key reason I was excited to accept the role of lab Director. Having the resources of Intel behind us enables us to go beyond ideas to really influence the ecosystem and products that are available in developing regions.”
"UC Berkeley obviously is a great research institution; it's a national treasure. We're not going to increase the quality of Berkeley's research or students, because they are already top of the line. But what we can do is help them in terms of scale, so together we can make a bigger impact on the world."
Kurt Brown
Co-director, Intel Research Berkeley
|
|
|
|
In exchange for providing resources to UC Berkeley, Intel gains access to some of the best and brightest faculty and students in the world. ”We’re working with talented young collaborators who might one day become our employees,” says Brewer. “By having a lab near campus and collaborating with the university, Intel significantly enhances its visibility and prestige with young researchers. They start to think of Intel not as some company off on the horizon but as those folks down the street who are doing some leading-edge research. That’s a huge factor in enhancing Intel’s reputation and ability to recruit top talent.”
"When I was a Ph.D. student at UC-Berkeley, I was able to work closely with several researchers at the Intel lab, publishing a number of papers and working on several software systems. This kind of relationship is advantageous for both Intel and UC-Berkeley, as it causes strong students to view Intel as a viable place for them to work after graduation, and gives those students access to resources and collaborators that they otherwise might not have."
Sam Madden
Assistant Professor, Electrical Engineering and Computer Science
Massachusetts Institute of Technology
|
The value to Intel goes beyond having access to top university researchers who may become employees one day. Having labs located near leading universities also helps Intel to keep up to date on the latest academic research developments. “When you have collaborations going on with four of the best universities in the world—and Intel has strategically located its labs close to four top schools—you will hear about every important new research concept,” says Brewer. “Either someone at the university will be working on new research that’s of interest to Intel, or some visiting professor will be giving a talk on it. So you have access to every big idea that comes along.”
That access to leading-edge academic research also helps to ensure that Intel’s university labs will not become insular. “One of the real dangers of industrial research labs is that, despite everybody’s best effort to the contrary, they become isolated and self-reinforcing,” says Scott Shenker, Professor of Computer Science at UC Berkeley and Group Leader of Networking at the International Computer Science Institute (ICSI). “The Intel labs, by being small and so closely connected to the university, have a constant influx of new ideas and new talent. That ‘s one aspect of the labs that I think is really crucial.”
|
|
|
In June 2005, after completing a two-year rotation as Director of Intel Research Berkeley, Joe Hellerstein returned to his tenured position as Professor of Computer Science at UC Berkeley and Eric Brewer took the reins of leadership. In addition to his academic credentials, Brewer brings industry experience to his new role. "What makes Brewer somewhat unique as the academic Director of the lab is that he co-founded and ran a very successful start-up company [Inktomi]," says Brown. "That gives him a certain base of experience that is not typical for academics. He has the skills needed to navigate a large organization like Intel and to scale up research projects. He also has developed good people management and employee career development skills along the way. That's a reasonably rare combination, and it makes for a very effective working relationship with him."
About Eric Brewer
Eric Brewer became the Director of Intel Research Berkeley in June 2005, taking a leave of absence from his position as Professor of Computer Science at UC Berkeley to assume leadership of the lab. Dr. Brewer's research focuses on all aspects of Internet-based systems, including technology, strategy, and government. As a researcher, he has led projects on scalable servers, search engines, network infrastructure, sensor networks, and security. His current focus is (high) technology for developing regions, with projects in India, Cambodia, Mexico, Sri Lanka and Bangladesh (so far), and including communications, health, education, and e-government.
In 1996, Dr. Brewer co-founded Inktomi Corporation with a UC Berkeley graduate student based on their research prototype, and helped lead it onto the Nasdaq 100 before it was bought by Yahoo! in March 2003. In 2000, he founded the Federal Search Foundation, a 501-3(c) organization focused on improving consumer access to government information. Working with President Clinton, Dr. Brewer helped to create FirstGov.gov, the official portal of the Federal government, which launched in September 2000.
Brewer received an MS and PhD in EECS from the Massachusetts Institute of Technology, and a BS in EECS from UC Berkeley. He was named by the World Economic Forum as a "Global Leader for Tomorrow," by The Industry Standard as the "most influential person on the architecture of the Internet", by InfoWorld as a top ten innovator, by Technology Review as one of the 100 most influential people for the 21st century (the "TR100"), and by Forbes as one of the magazine's 12 "e-mavericks," for which he appeared on the cover.
About Kurt Brown
Kurt Brown is Co-Director of Intel Research Berkeley. His research interests focus on information storage, retrieval and analysis, and how to make those tasks transparent and the technology easy to use, and self-managing. Prior to joining Intel, Brown founded and led 64K, a provider of database search middleware that was later sold to software maker Verity, Inc., and ELetter, a web-enabled direct mail service for small businesses. He is a 13-year veteran of IBM, where he worked on the MVS* and DB2* product lines, in addition to assignments at Yorktown and Almaden Research. Brown received his Ph.D. from the University of Wisconsin, for work on what is now commonly referred to as "autonomic" resource allocation in database systems.
"The Intel Research Berkeley lab, under the leadership of David Culler and now Joe Hellerstein, has provided a set of tremendously exciting and world-changing projects in PlanetLab and sensor networks. The intellectual depth of the ideas involved, their potential to change the world, and the level of resources Intel has made available for these projects has been extraordinary."
Scott Shenker
Professor of Computer Science, UC Berkeley
Group Leader of Networking, International Computer Science Institute (ICSI)
|
|
|
|
Under Intel's research model, when a university lab transitions to a new director, roughly two-thirds of the research agenda remains the same. Under Brewer's direction, most of the projects begun during Hellerstein's tenure continue to evolve and prosper. The new component of the research agenda reflects Brewer's strong interest in technology for the developing regions of the world.
Technology and Infrastructure for Emerging Regions (TIER)
The TIER project is addressing the challenges of bringing the Information technology revolution to the developing regions of the world, which have unique challenges, such as cost, deployment and power consumption, and require novel technologies. Intel and UC Berkeley are two of the key
collaborators in the project. Their focus is on developing new networking infrastructures for emerging markets. ...Read More
Delay Tolerant Networking
Today, most networking technology is designed to provide continuous connectivity and Internet access. In very remote areas and many portions of the developing world, this may be very expensive or simply impractical due to the lack of continuously available power and networking infrastructures. The goal of delay tolerant networking (DTN) is to provide reliable, secure, asynchronous communication for difficult to reach places that may only have intermittent connectivity. DTN is being designed for application in a wide range of environments, including rural villages in developing countries.
Through the Delay Tolerant Networking Research Group (DTNRG), a research group associated with the Internet Research Task Force (IRTF), Intel researchers and other government and industry researchers are working together on an alternative to the standard Internet TCP/IP communications protocol that is better suited to the communications challenges of environments where connectivity may be intermittent and where a choice of communication technologies may be available. The alternative uses hop-by-hop storage and retransmission to form a delay-tolerant, heterogeneous overlay network, and provides a messaging service interface that is conceptually similar to email but is application-independent and supported by specialized reliability, routing, and security capabilities. DTN can work in conjunction with TCP/IP in environments where connectivity is good, but can also operate in situations where TCP/IP is not available.
Implementation and wide deployment of DTN would represent a significant step forward in communications capability for remote areas and developing regions. DTN enables a range of applications to be used in environments with poor connectivity-from email and voicemail to offline search engine queries, electronic form filling, and "instant-enough messaging"-at a reduced cost.
Rural Connectivity Platform
There's a tremendous need for affordable, robust wireless networking in rural areas of the developing world. Existing infrastructure that is useful in high-density areas, where costs can be spread over many users (notably, cellular telephony), is not appropriate for sparsely populated rural regions
Intel researchers are addressing the challenge. Meshes of point-to-point long-distance wireless links between villages, augmented with a number of broadband satellite connections seem to be one of the best connectivity choices. A key component of this solution is a variation on IEEE 802.11 (Wi-Fi) technology. Low-cost Wi-Fi technology could form the backbone network in rural regions. While the 802.11 Media Access Control (MAC) protocol was not designed for long-distance communications, researchers believe that modifying the MAC layer should resolve the problem, without the need for hardware or driver modifications.
Researchers are currently assessing the performance of point-to-point 802.11 and 802.16 (WiMax) long-distance links in a wireless testbed in and around Berkeley, California. They also have deployed several test links in India and more recently, in Ghana, where they have demonstrated a bandwidth of 5-7 Mbps over an extremely long distance: 10.5 kilometers. Researchers will continue testing and modifying the technology in additional deployments in India in 2006.
Creating a platform to demonstrate the viability of this approach to long-distance communications in rural regions requires inputs from multiple disciplines. As a result, researchers at the Berkeley lab will collaborate with other researchers from Intel, UC Berkeley, the University of Washington, and UC Riverside as well as consultants from Thinkbank and ACME Laboratories on this phase of the research.
Networks as Databases
P2: Declarative Networking
Any widely distributed system must track its participating nodes and must be able to send messages among those nodes. This capability can be thought of as an overlay network, since it provides an application with customized networking functionality (naming, topology, and routing) that runs as a layer over traditional IP networking. Such overlays are in widespread use, including in commercial mail/directory servers, application-level multicast systems, and distributed hash tables (DHTs).
It is tricky to design, build, and deploy an overlay network for a particular application and environment. To ease the process, researchers from Intel Research Berkeley and UC Berkeley have collaborated to implement P2, a system that uses a high-level declarative language to express overlay networks in a highly compact and reusable form. P2 has been used to specify and execute working, detailed overlays in a tiny fraction of amount of code used in traditional implementations. P2 automatically compiles high-level specifications to a dataflow-oriented runtime system, which can itself be used by expert programmers to optimize overlay efficiency and performance.
P2 is part of a broader effort to revisit networking technology through the lens of distributed database query processing. The P2 research could provide simpler, safer specifications for network protocols and-within the same framework-the ability to query, monitor and control all aspects of the network's distributed state.
Enterprise Network Security
In a typical enterprise network, the traffic flowing between routers constitutes a massive data set. Intel researchers are applying sophisticated statistical models to extract patterns from such data, to identify "normal" traffic patterns so that anomalies, such as worms or denial-of-service attacks, can be quickly detected. The ultimate goal is to enable designers to create more secure, efficient and reliable enterprise networks. The research is being carried out in collaboration with UC Berkeley, Boston University and the University of Paris.
Unlike typical approaches, which focus on analyzing traffic data one link at a time, Intel researchers are examining all traffic flowing through the network simultaneously. Analyzing ensembles of nodes rather than individual machines enables researchers to more easily identify all sources of attacks or other network problems. Such knowledge is vital in developing effective network security applications.
The pilot is slated to begin in late July, and Buonadonna says he and BP are both optimistic about the prospect for success. "Some of the early data we have from the site survey is very promising, and I think BP is excited about the technology," he says.
In a related effort, researchers are attempting to model normal behavior for end users and hosts (e.g., laptops), in terms of communications patterns. The goal is to create models with enough flexibility to allow for development of customized security profiles for end users, to create more effective security applications. The next challenges are to learn how to identify and understand new forms of security attacks, and to relate communication patterns to application usage.
Digital Home
Micro-Mobility
Understanding how people interact with a growing number of portable electronic devices in homes could help in the design of devices and the architectural spaces that support them. Intel researchers are exploring these interactions, currently focusing on laptop computers in homes with wireless networks.
Researchers are creating a sensor-based visual record of the physical movement of people and devices in the home, to facilitate more accurate discussion during interviews. This approach has many advantages compared to traditional methods such as self-reporting, which tend to be inaccurate because people have difficulty recalling details of mundane activities.
In an initial study of ten family members in four households, researchers found that participants had a limited number of favored places in the home where they spent time, and wireless laptops were primarily used in these places. In addition, household members tended to interact with their laptops when they left the house.
These initial findings suggest that future homes might include an electronic foyer equipped with a built-in printer and electronic magazine rack that could hold the laptop and charge its batteries. Homes might also have a moveable wall to support the allocation of more space to whatever favored place is being used at the time.
In the future, researchers will focus on a broader sample of homes; explore alternative sensing methods, to track a wider range of portable devices; conduct more detailed quantitative analysis of the data (for example, to classify different types of movements); correlate people's movements and home computer usage with other data, such as weather or status of personal relationships; and continue to refine their methodology.
Urban Computing
Another research project is exploring how people who live in cities might want to use technology, how it could help them develop a sense of community or belonging, or play into their emotional experiences of urban living. The researchers seek to provoke discussion aimed at understanding the emerging space of computing within and across our public urban landscapes - Urban Computing. By gaining a better understanding of what matters most to people in the daily experience of city life, researchers hope to inspire useful new technologies for urban dwellers.
Part of the research involves urban probes. These are provocative interventions designed to engage people in direct discussions about their current and emerging public urban landscape-and in the process, reveal new opportunities for technology in urban spaces.
To gain a deeper understanding of the Urban Computing landscape and the people who inhabit it, researchers are developing relationships with professionals in urban planning, sociology, art, design and other domains. The Urban Computing team collaborated with the UC Berkeley Architecture Department to produce a novel installation at the San Francisco Museum of Modern Art Museum in October 2005, to explore the difference of statistically projected behaviors through an over-mapping with actual behavior. Using RFID tags and a range of antennas each corresponding to a particular behavioral zone, the individual and group behavior of participants at the event were tracked and projected live onto a tessellated screen.
Evolvable Internet Architecture
At the core of the Internet is the Internet Protocol (IP), which controls how routers transport data packets between arbitrary end hosts. IP is a powerful and critical piece of the overall Internet architecture.
As the Internet has grown, users' requirements of IP have evolved too. The research literature is replete with calls for extending IP to improve the security, robustness, and manageability of today's networks. Unfortunately, the troubled history of attempts to deploy new IP architectures such as IPv6 and IP Multicast has led to a deep pessimism about evolutionary architectural change. The prevailing wisdom is that ISPs have little incentive to deploy new architectures; since all ISPs have to act in concert, there is no competitive advantage to such advances, and the costs of universally deploying a new architecture are immense.
Intel researchers, in collaboration with UC Berkeley researchers, are tackling the problem of IP's inability to evolve. Their goal is to understand what makes an architecture evolvable-that is, capable of incremental change led by its incumbent providers within the existing market structure. This involves both technical and economic issues; ISPs must be able to offer new architectures and must have an incentive to do so.
The researchers propose a strategy they refer to as Universal Access whereby, if any ISP deploys a new IP technology, all Internet users can access the technology, regardless of whether their local ISP supports it. By enabling customer choice, Universal Access allows for competition among ISPs, fostering innovation.
To enable end users to access such new technologies, researchers are exploring the use of IP anycast technology to automatically re-route packets at the network level. The project team, in collaboration with Cornell University, is deploying an experimental Internet anycast test bed. In addition, they are working with the UC Berkeley School of Information Management and Systems (SIMS) to explore the economic implications of Universal Access in terms of inter-ISP settlements.
|
|
|
The close connection between Intel Research Berkeley and the university next door extends beyond research to education. "We have tried to create an environment such that the lab feels like it's an extension of the Berkeley campus," says Brewer. The lab regularly hosts seminars, at the lab or on campus, that are attended by Berkeley faculty and students. In addition, some of the lab's researchers teach courses at UC Berkeley.
"The presence of the Intel lab gives us access to a certain number of distinguished researchers who will occasionally offer courses on campus. I see this as a good thing, because it gives our students access to a greater variety of courses. These courses, which typically are at the advanced graduate level, provide students an introduction into an active area of research and an entree into a subfield."
Jitendra Malik
Chair, Department of Electrical Engineering and Computer Sciences
University of California, Berkeley
|
"Teaching at UC Berkeley complements the Intel lab's research mission, because the way graduate students move into research often is through advanced level courses," says Jitendra Malik, Chair of the Department of Electrical Engineering and Computer Sciences at UC Berkeley. "The availability of Intel researchers who can teach or co-teach such courses on campus makes it easier for them to get students involved in their projects."
Eric Brewer is pleased with how Intel Research Berkeley is evolving and excited about its prospects for the future. "The lab has become a crossroads for computer science research," he says. One indication of the lab's strong reputation for research is the fact that two leading postdoctoral students postponed accepting faculty positions to spend more time working at Intel Research Berkeley. Sam Madden spent six months at the lab before joining the faculty of MIT and Carlos Guestrin decided to work for a year at the lab before accepting an Assistant Professor position at Carnegie Mellon.
"The Intel lab is one of the best places in the world-if not the best-for research in sensor networks," says Guestrin. "When I was offered the opportunity to work there, I was thrilled. The lab has given me the opportunity to explore the interaction between sensor networks and machine learning. This unique learning opportunity, along with the possibility of interaction with Intel researchers and UC-Berkeley students and faculty, was irresistible and has been extremely productive."
The success of the lab’s fourth third open house in October 2004 provided more evidence that Intel Research Berkeley has become a computer science crossroads. A mix of visitors that included academic researchers and members of the Bay Area’s technical and venture capital communities mingled with researchers and viewed 20 demonstrations of research projects underway at the lab. “I think the most important thing about the open house is that it makes people realize how broad the research agenda is at Intel,” says Brewer. It shows that we are exploring not only classic technologies like networking and systems, but also innovative areas such as urban computing and technology for developing regions. That’s what makes Intel such a great place to do research; there are always some exciting projects underway.”
From the viewpoint of UC Berkeley, Intel’s experiment in establishing a new model of industry-university collaboration has thus far been a success. “Intel’s open and collaborative research model has set a new standard for university-industry research,” says Scott Shenker. “Being very open and public about the agreement between Intel Research Berkeley and UC Berkeley not only helped the relationship between Intel and the campus; it also gave the campus a template that it can show to other companies who want to collaborate, as a starting point for the discussion. I think that openness helps to preserve the integrity of university research.”
Under Brewer’s direction, the lab is expanding its relationship with the university next door, Kurt Brown notes. “We have great interactions with many faculty in the EECS department at Berkeley,” he says. “We have also been doing some work with the Architecture Department, and we’re hoping to branch out to other departments as well.”
Brewer’s goal is to continue building the relationship with UC Berkeley, and to sustain the lab’s record of success while evolving his research agenda for the developing world. “By the end of my rotation as Director of Intel Research Berkeley, I hope that Intel will have a strong focus on technology for developing regions, and maybe even have products in the pipeline that you could point to that came out of our lab’s research,” he says. “Our ultimate goal is to change the world with our research, and in my view, Intel is a big lever to help make that happen. If I can use that lever effectively while I’m the lab director, if our research can help to bring affordable computing and communications technology to the developing world, that would be incredibly rewarding.”
|
|
|
