The Counting House

B. P. Hayes · American Scientist · 1994

Lan ethos, an indigenous style. At the Fermi Nati nal Accelerator Laboratory the style is utilitarian chic. Fermilab delights in the ingenious and elegant use of the found object, the industrial artifact converted to some higher purpose, hi the 1970s, while a rival European laboratory quarried Carrara marble for shielding in a large detector, Fermilab filled a similar need by scrounging ar mor plates from two decommissioned battleships. The laboratory's auditorium is built out of con crete castings left over from the construction of a beam-line tunnel. These design choices were not a product of mere frugality; there is also an aesthet ic principle behind them, a pride in turning to good account materials that others might have discarded or covered up. This same aesthetic sense can be seen at work in decisions that have shaped the laboratory's computing program. A major consumer of com puting resources at Fermilab is a task called event reconstruction, which was traditionally done by mainframe-class computers, such as the larger members of the Digital Equipment Corporation VAX series. As the volume of data increased dur ing the 1980s, the computing requirements out grew those machines. One might have expected them to be replaced by bigger and costlier super computers, such as those made by Cray Research, or by massively parallel machines, such as the CM-2 (with 65,536 processors) made by Thinking Machines, Inc. The Fermilab Computing Division took a different path: They created a farm of off-the-shelf workstations and developed the soft ware needed to make hundreds of them cooper ate. The individual computers are not very pow erful, but they are cheap, and so a budget that would barely pay for air-conditioning a super computer can buy quite a number of them. I recently spent a few days visiting Fermilab and talking with Thomas Nash, who was then the head of the Computing Division (he has since been named Associate Director for Scientific Technology and Laboratory Information), and with others who do various kinds of computing

Read the paper · More papers on PaperTik