G. Christopher Anderson’s generally excellent article on upgrades of the NSF supercomputer centers (The Scientist, Aug. 7, 1989, page 1) contains some errors regarding the history of NSF centers in general, and of the San Diego Supercomputer Center (SDSC) in particular, that I would like, for the record, to correct.
Anderson says we “chose the DOE-standard CTSS operating system. . . in 1985. Unfortunately, most of the other centers chose UNICOS.” Both the SDSC and the University of Illinois supercomputer centers began operations in 1985 with the sole interactive Cray operating system then available, CTSS. The only other NSF center to operate a Cray supercomputer, the Pittsburgh Supercomputing Center, began in late 1986 using COS—the original Cray operating system, which was a batch system and had no resemblance to Unix.
Illinois made the switch to UNICOS when they acquired a CRAY-2 in 1988. It was not until 1989 that UNICOS was substituted for CTSS on their CRAY X-MP/48. PSC switched from COS to UNICOS with the arrival of the Y-MP at about the same time. Similarly, we will bring our Y-MP up at the end of this year under UNICOS. Anderson describes the switch as “a long and time-consuming process that requires rewriting much of the center’s software,” implying that we are making a switch no one has made, when in fact all three centers have had to change operating systems to arrive at a standard operating system.
Our conversion should take much less time than Illinois’ switch, and itshould be less painful for the SDSC and its users than it was for Illinois and its users. We are benefiting both from their experience and from the continued development of the UNICOS operating system. Thus, in discussing our move to UNICOS, Anderson is not quite fair to either NSF or the SDSC. He imputes to NSF a newfound sensitivity to standards and implies that the SDSC has somehow lagged.
The adoption of standards for supercomputing, as for any other kind of computing, requires careful calculation of both benefits and costs. When a standard is adopted, communication among machines is eased, software is more readily portable from machine to machine, and the difficulties individuals suffer (from operating in more than one environment simultaneously) are correspondingly reduced. But a long view demands recognition that standards can only temporarily freeze the natural liquidity of innovation; a standard that was once an engine of progress can become a brake on future development.
What I would like to emphasize most particularly is the overall spectacular success of the NSF centers program, which has resulted in everything from new tire designs to greater understanding of the origins of the universe, to say nothing. of the appearance of thousands of new publications in the scientific literature. These are the important advances in which the calculations and conclusions depend com- pletely on the availability of supercomputers to the U.S. academic and industrial research community. (Anderson quite properly pointed out the importance of scientific visualization to these developments. While giving Illinois its due as a pioneer of visualization, The Scientist published, in the very same column and without credit, a picture of a neuron that was rendered at the SDSC.)
But that is a quibble. The main point, which cannot be over-emphasized, is that the NSF centers are that rarest boon, a government program that has given the taxpayer full value.
San Diego Supercomputer Center
San Diego, Calif