Architecture – MIT Builds Its Dream House

It’s a few days before the big March move and Jack Costanza squats and touches the inside of a large black optical switch, a “collapse point” that will act more or less like a router. “See this?” he says, holding up his finger. There it was. A little white smudge of plaster dust. On the tip of the $100,000 switch, it would, according to Costanza, seriously compromise the equipment’s performance. “This tells me I have a lot more shouting to do,” he says.

Costanza is being facetious. He doesn’t shout; he has a quiet, bemused air that says, in effect, “Problem? What problem?”

But problems abound. MIT’s computer science labs are moving into a new building?the hypermodern, $285 million, 719,000-square-foot Ray and Maria Stata Center designed by architect Frank Gehry. And for the close to 1,000 teachers, students, researchers and support staff working in dozens of groups representing fields ranging from robotics to biological computing, the new building is drastically different from the only labs most of them have ever known. And hardly any of the solutions, relationships and customs of the past can be taken with them. Everything has to be made new, and all these new systems have to mesh.

As director of IT for the newly formed Computer Science and Artificial Intelligence Lab (CSAIL), Costanza owns a major piece of the move. Networks are mission-critical in any organization, but considering the nature of the CSAIL labs (ubiquitous computing, advanced network architectures, parallelized operating systems and more), given a choice between having the walls up or the network up, everybody here would pick the latter. After all, if a chilly wind blows, you can always wear a sweater. But if Costanza can’t deliver a functional network on moving day, there’s just no point in moving. And he’s not going to be able to deliver if construction grit keeps drifting into his switches.

Of course, networks move all the time. But in this case the stakes are unusually high. Engineering has passed through two major intellectual renovations over the course of the past 150 years, and MIT has led both of them. Part of that leadership was defining the architecture that supported each vision. The building Costanza is moving his network into reflects the Institute’s belief that a third engineering renovation is sweeping the technosphere. If MIT is right, a lot of us might be moving into buildings much like the Stata Center in the years to come. If not, if the building doesn’t work, if, instead, it gets in people’s way…

Well, the truth is, nobody dares even think about that.

Old Paradigms, Old Buildings

For the first 75 years of MIT’s history, from 1865 to 1940, engineering meant building and maintaining big pieces of equipment?locomotives, turbines, refineries?for large industrial companies. MIT’s job was to train a sort of superapprentice, a person with an industrywide mastery of current designs and a sound understanding of the ideas behind them. Innovation was not a focus, although, naturally, good ideas were always welcome when they appeared.

After World War II, a second model for engineering emerged, one less interested in the day-to-day issues that governed industry and more concerned with mounting long-term assaults on big problems of general importance, like fusion or artificial intelligence. (The prime example of the form was of course the Manhattan Project, which birthed the atomic bomb.) This kind of activity was known as engineering science. It chased breakthroughs, and typically it was conducted by teams headed by a master of the subject. Researchers often were supported by grants from governmental agencies like the Office of Naval Research or the Department of Defense’s Advanced Research Project Agency (DARPA).

Both models for engineering were expressed in the unique architecture of the school’s buildings. The “superapprentice” era favored 30-foot bays connected by long, straight corridors wide enough for two carts piled with heavy equipment to pass. (From this requirement we got MIT’s famous “Infinite Corridor,” an 825-foot-long passageway that skewers five buildings on a single shaft of space.)

Engineering scientists, however, working long-term on big problems, needed little more than access to a computer and a workbench cluttered with circuit boards. Consequently, they did their work in small, scattered spaces that from the outside looked very much like office buildings in industrial parks.

In the ’60s, MIT needed space for the growing number of researchers interested in exploring the new digital disciplines. Since computer scientists didn’t need access to the Institute’s corridor network (their devices were not heavy and didn’t need to move), MIT decided to house them in a commercial office building at the northeastern edge of the campus. This structure came to be known as NE43 (from its MIT mail address). It was in NE43 that the research and innovations associated with MIT computer science?artificial intelligence, public-key encryption, time-sharing?were made. This was where Marvin Minsky worked on AI, and David Clark supervised the development of many of the early Internet protocols. In fact, for a while the machines sitting in NE43 handled nascent Internet services (known as ARPAnet) for the eastern United States.

But as brilliant as its history may be, NE43 itself is a dim, narrow, characterless place. If you piled eight shoeboxes on top of each other you would have a fair model of the building. Ordinarily, this wouldn’t have been a problem?the space was functional, especially as you could hack it up at will?but during the ’80s and ’90s, the culture turned computers and networks into the technology equivalent of movie stars. Any prospective donors or patrons looking for a tour of MIT’s hot spots ended up in NE43. Housing computer science under such conditions began to seem like sticking the corporate offices of a Fortune 500 company in a double-wide.

A Star Building for a Star Science

MIT’s administration wanted to make a sweeping declaration of its support for the computational sciences, and in 1997, Analog Devices cofounder Ray Stata and his wife Maria lit the fuse with a gift of $25 million. They were rewarded with the promise that a new building would bear their names. A year later, as if compensating the labs for all those years they spent stuck in dingy old NE43, MIT hired the best (or at least most famous) architect in the world, Frank Gehry, to build its dream house.

Gehry toured NE43, listening to the researchers who, as much as anyone, were his clients. According to a representative at MIT, Gehry didn’t ask the researchers what they wanted; he asked them about their routines?what they did during the day, who they went to when they had a problem, where they headed when they wanted to think. He used that information to figure out what they needed (as opposed to what they wanted).

The Stata Center’s design indicates that Gehry came away from those interviews convinced that something big and new was going on in engineering.

The building is split vertically into alternating yellow brick and aluminum-colored metal sections. The brick pieces seem sliced out of the plain, rectangular, office building-like engineering science labs that were thrown up all over the country in the ’50s and ’60s. The metal sections bulge out between the brick walls?futuristic, full of energy, forward leaning. It is as though engineering science was a chrysalis and now a new suite of energies is bursting out.

Gehry’s design describes an epic metamorphosis in the way MIT understands engineering and in what it expects from its engineers.

The New Engineering Paradigm

Many of CSAIL’s projects are not targeting inventions on the engineering science model, not looking for breakthroughs, but seeking instead continuous, rapid improvements along a specific dimension. Project Oxygen, for example, which is billed on its website (oxygen.lcs.mit.edu) as an effort to bring “abundant computation and communication, as pervasive and free as air, naturally into people’s lives,” will never culminate in a single grand device, a locomotive of computation or a nuclear bomb of communication. It will, hopefully, make better and better versions of what we already have.

The power of device networking is a function of the number of devices, of sensors and actuators, connected to the network: a billion devices now, a trillion later and so on. In area after area, the projects at CSAIL today are about exploiting resource growth as it passes from the giga- to the tera- to the petabyte stage.

Engineering in the age of continuous improvement is not beholden to industrial clients (they’re nice to have, but they come and go) as it was in the era of heavy industry, when the only consumers of the engineers’ heavy metal were industrial giants. And subject matter experts are not as critical as they were in the time of science engineering because the nature of the constraints on a project are ever changing. Today’s expert becomes tomorrow’s neophyte, as one year’s improvements come out of efficiency, the next out of extending operating life, the third out of better ease-of-use. Riding this engineering tiger demands the ability to manage huge flows of information, leap from issue to issue, organize new collaborations quickly, and plan to the curve?”we’re going to have to be here in 2006 and here in 2008.”

The Stata Center appears to have been constructed expressly to support this new engineering model.

Inside The Structure

“It’s a people building,” says Chris Terman, associate chair of CSAIL, who asserts that the Stata Center’s layout promotes conversation and interaction among people. It’s designed for ease of movement. It promotes collaboration. It welcomes change.

The floors are raised so that cables can be pulled anywhere. Many of the walls are moveable, and many of those are half-height, like Jersey barriers, so that people can see each other. “The walls here are like furniture,” says Garrett Wollman, a technician who works for Costanza.

The headwaters of intellectual activity in the Center will be found in small groups of desks clustered together in open spaces. While the desks can be moved anywhere in the space, it’s easiest to visualize them pushed together, in tight, almost intimate collaborations. Each work space has enough dedicated resources (kitchens, conference rooms) to allow a team to keep to itself when it chooses. Take one step outside those spaces, and you are swept immediately into broader streams of traffic. Spiral staircases punch up and down to mix vertically (no need to go find an elevator); corridors, internal streets, flow horizontally through the work spaces like a network of streams connecting pools.

The architecture seems to support three levels of interaction: first among team members, then among the members of adjoining teams, and finally among all the teams, in a town center on the first floor. The center invites people to mix in every possible way. It has a cafŽ, a pub, a fitness center, a dance studio, an amphitheater and a direct connection to the campus swimming pool. In the internal courtyard that’s surrounded by the building on all sides, there are cascading terraces, raised gardens and a hill with actual trees. Stata’s is a landscape for interaction.

Inside the Infrastructure

Costanza walks into an empty room. “This is the holodeck,” he says. “Who knows how much computer power this is going to need.” He’s not joking. This room is intended to support large-scale experiments in vision and imaging, and there’s a computer center right next door ready to go.

Back in the day, networks were planned according to need. In designing the Stata Center’s network, Costanza has assumed eight years of steady increases in demand. (The primary switch runs at a scorching 1.6 terabits per second; the dataports scale up to 1Gbps.)

The logic of continuous improvement engineering also affects network services. Historically, MIT labs have tended to maintain their own services?for example, their own e-mail and Web servers, backup systems and virus checkers. This made sense when research targets were long-term and teams were stable. But when flexibility is a requirement of research, one cannot assume continuity. Nontechnical people and people on short-term assignments (who might or might not be physically present) will be rotating in and out. As a result, networks have to support both a complete range of services?all the tools, all the operating systems?and provide the ability to opt out of any one of them should a researcher wish to design his or her own (for instance, the people working on vision research might want mailers optimized for very large files).

“Selection has to be ˆ la carte,” Costanza says. “You need a centralized and a decentralized services structure, running in parallel and equally accessible to everyone.”

Costanza’s Brave New World

All this novelty plays havoc with planning. Costanza walks into a room with a large U-shaped desk. “This is an assistant’s desk,” he says. He looks at the wall. “It has no phone.” He walks around to the other side of the desk. “It has no dataport.” He looks at the floor. “It has no power.”

Every day when Costanza comes to work, he puts on a hard hat and tours the building. People with problems line up like refugees looking for food. Their questions go well beyond network issues. One worker asks if Costanza knows where the nearest working rest room is. He does. A team wants to confer about an upcoming fire inspection. Costanza has personally put a match to the insulation being installed. He has built makeshift clean zones around his $100,000 switches with yellow tape (like the police use at crime scenes). The tape keeps getting moved by mysterious forces.

The Stata Center not only reflects a new type of engineering, a new way of working and thinking, it also illuminates the problems we will have adapting to the new kind of world it creates. NE43 was a box, easy to figure out. There was nothing but flat walls and right angles, and everything stayed put. If a service was delivered to a desk, that desk would not turn up 10 feet away the following week.

The Stata Center, however, offers so many choices that thinking through them could take forever. And perhaps one shouldn’t. Perhaps analyzing all the choices inherent in any given problem is no longer a worthwhile investment of time and energy. New decision-making skills may be needed. And if the rhythms of continuous improvement dominate the future of engineering, as seems likely, these are skills that many of us will have to learn.