the 10 Berkeley projects on which I participated.

The multidisciplinary nature of the project means students gain hands-on knowledge about other areas by working closely with students and faculty in other fields. The experience they gain building the common prototype helps them develop taste in research topics, which in turn helps them pick interesting research topics for their dissertations and later in the rest of their careers.

Group projects create communities where students have others with whom to interact. In particular, the more senior students can mentor the junior ones. Being a Ph.D. student can be a very lonely experience, especially when it comes time to write a dissertation; being part of a larger group can allay those feelings of isolation.

We recently started celebrating the 10-year anniversary of the end of projects. The high participation level at these reunions indicates that these personal ties in such communities remain 10 years later. The accompanying photo shows the Network of Workstations (NOW) group reunion held last year.^a

Research retreats. Key to the success of these projects, and to the development of Berkeley systems graduate students, has been twice-a-year, three-day retreats where students on the project present their results to one- or two-dozen guests from industry or non-academic labs. These are intensive events, lasting from early breakfast to late-night discussions, although we do take off one afternoon to have some fun. Retreats act as project milestones, with the specter of presenting to outside visitors motivating students to meet the milestones. We close the retreats with an outsider feedback session that offers advice on any aspect of the research. It’s surprisingly rare in academia to get frank feedback about research, but who can’t benefit from constructive criticism?

Retreats give graduate students two chances per year to give a serious talk

Patterson’s research projects.

Years

1977–1981

1980–1984

1983–1986

1985–1989

title

X-Tree: A Tree-Structured Multiprocessor

RISC: Reduced Instruction Set Computer

SOAR: Smalltalk On A RISC

SPUR: Symbolic Processing Using RISCs

RAID: Redundant Array of Inexpensive Disks

NOW: Network of Workstations

IRAM: Intelligent RAM

ROC: Recovery Oriented Computing

RAD Lab: Reliable Adaptive Distributed Computing Lab

Par Lab: Parallel Computing Lab

Professors

3

3

2

6

students

12

17

12

21

1988–1992

3

16

1993–1998

1997–2002

2001–2005

2005–2010

4

3

2

7

25

12

11

30

2007–2012

8

40

a Additional photos are included with the version of this Viewpoint available at the Communications Web site, cacm.acm.org. The online version has names and group photos for RAID and SPUR reunions and for the most recent Par Lab and RAD Lab retreats.

and receive advice from experienced researchers outside academia with different experience and perspectives from the faculty on the project. Students are energized when external people care about their work and find it important. When we advisers say something is good, many students will assume we are just acting as cheerleaders or just trying to get them to work harder. I believe interaction with thoughtful colleagues from industry and non-academic labs is vital to acquiring research taste in computer systems by learning to identify critical problems and impactful solutions. Retreats also introduce students to a network of colleagues that may prove useful later in their careers.

Such projects and retreats might be difficult at some places. Building collaborations with local universities and industry can produce many of the same benefits. The key is to get everyone to stay the full time and have people outside your group provide candid feedback. For example, there is an annual Boston Area Architecture workshop involving Brown, Harvard, UMass, Northeast, RPI, and local industry so that their students can cut their teeth in front of a friendly audience and get feedback from outsiders.

We have been doing retreats for 25 years. To my surprise, three years ago we discovered another technique that is becoming just as important to the success of projects and graduate

students.

Open collaborative laboratory. We were increasingly seeing people optimize their schedules to avoid disruptions by working from home when they didn’t have classes or meetings, since computers and networks were just as fast at home as in the office. The negative global impact of such a local optimization can be thought of as corollary of Metcalf’s Law: if the value of a network is proportional to the square of the number of connected users, even a small group leaving a network can significantly decrease its value. This drop in value can in turn cause others to leave, with the negative feedback loop continuing until the network nearly collapses.

In 2006, we experimented by creating a physical office area with contiguous open space for everyone in the project, including the faculty. We hoped that easy access to faculty would draw students to campus and that the open space would inspire innovation by increasing the chances of spontaneous discussions. ¹

The open space makes it very convenient to quickly grab a group of interested people on a moment’s notice for a discussion rather than trying to wander around the building or exchange a volley of email messages to schedule a meeting. We have also been surprised to see new students in this space quickly act like senior graduate students. Apparently, easy access to faculty plus