Laboratory personnel filled the Materials Science Laboratory Auditorium Thursday to hear former Computing (C) Division Leader Roger Lazarus talk about the history of computing at the Lab. Photos by LeRoy N. Sanchez, Public Affairs
Supercomputers of today may be thousands of times faster than the early machines at Los Alamos, but the ingenuity of computer scientists and engineers remains the key ingredient, Roger Lazarus told a full house Thursday at the Laboratory's Materials Science Laboratory Auditorium. Lazarus, who founded Los Alamos Scientific Laboratories' Computing (C) Division in 1968, recounted how Manhattan Project calculations evolved into supercomputing through a series of carefully chosen anecdotes from the Laboratory's early days in his talk, "35 Years in 45 Minutes: 1951-1985."
The history talk kicked off a new series of seminars presented by the Computer and Computational Sciences (CCS) Division whose goal is to provide a forum for public discussion of key issues in computing and information technology. Schedules for future seminars are available at the seminar Web Site at computingandscience.lanl.gov online.
Lazarus came to the Laboratory in 1951 and began working on single-stage yield calculations that could be modeled in one dimension, a job he stuck with in one form or another for 17 years.
In those days, "the weapons work was really driving the financial aspects of the Laboratory," Lazarus said. He ran early codes on primitive computers from IBM and worked in the Theoretical (T) Division with Nick Metropolis and others on the design of MANIAC, the first stored memory computer. When then-director Norris Bradbury wanted someone to establish a new division focused on computing, he called on Lazarus to pull together about 100 weapons code and machine specialists.
Computing technology doesn't progress in precise linear fashion, Lazarus said. In some respects the early hand calculators and Card Programmable Calculators, with their tables and plug boards acting like primitive software, were more sophisticated than the speedier, larger, and more automated machines developed by John von Neumann.
By comparison to the hand calculators, the CPC computers were blazingly fast — about five operations per second. The scientists who developed the early weapon codes laboriously punched out the color-coded cards, then "shuffled" them after the "deck" ran through the machine.
The next major computer was the Standards Eastern Automatic Computer, or SEAC, developed at the National Bureau of Standards. It sported 512 words of reliable memory, 512 words of unreliable memory written onto electronic tubes, and used paper tape instead of punch cards.
Although Lazarus remembered just one totally error-free run on the SEAC computer, the machine did permit more sophisticated computing. The basic structures of codes haven't changed all that much, Lazarus said, recalling that he and his colleagues were able to run hydrodynamic codes on a mesh with the SEAC
Los Alamos users also trained the cantankerous machine to print out a tell-tale warning in case of major errors: "CALL A CAB."
"When that happened, we'd go home and leave it to the engineers," Lazarus said.
MANIAC was "pretty reliable," and was quickly upgraded to a 10,000-word drum memory. The big machine used many of the same codes as the first Card Programmable Calculators computers, but scientists had to build in floating point scaling. When the machine acted up, the problem most likely was caused by an intermittent loose connection, so the scientists kept a large rubber mallet handy to the frame of the MANIAC until the circuits behaved.
Most visitors to MANIAC were impressed by a glass box that prominently displayed the legend, "In case of emergency, break glass." Behind the glass, Lazarus said, was an abacus.
The first IBM 701 computers of 1953 featured punch cards with 2,000-word memories and frequent hardware errors that forced users to develop energy checks and restart dumps, still integral to successful large-scale computing. Installation of those machines led to another first, the hiring of the first professional computer operator.
During this period, Lazarus and Dave Woods developed the venerable HENRE code, an acronym for Hydrodynamics, Energy, Neutron, Radiation, Etcetera. It was the last component that led to the HENRE's longevity, as researchers continually built on the code for 20 years.
Lazarus and Mark Wells also developed the first major model of a nuclear reactor meltdown. He told of taking several months to write another, special-purpose code for a one-time weapon test that provided a single mathematical result.
"Certainly, you didn't need any visualization output for that," he joked.
In 1956, Lazarus said, IBM proposed joint development of an entirely new type of supercomputer that would be 100 times faster than the IBM 704 and cost more than $2 million. Following a formal procurement process, the Laboratory and IBM began work on Stretch, a revolutionary machine whose basic technology would set the standard in computing for more than a decade.
The three-year moratorium on nuclear testing was a fruitful period for advances in machine and code development because weapon designers and computer scientists had more time to collaborate on improvements with IBM researchers.
Stretch was the first solid-state computer. Its magnetic disk changed the way information was stored. Input/output functions ran on a separate machine, greatly speeding up operations. And Stretch featured single-error correction and double-error detection.
When Stretch was delivered in 1961, it proved to be about 25 times faster than the machines Los Alamos was using at the time, and the cost was closer to $4 million, but the computer still revolutionized weapons work at the Laboratory.
Once the machine was working, IBM's advertising juggernaut began to tout its advanced design. In addition to the real achievements, the company began to brag that Stretch was the first computer capable of unattended operation.
"We thought that was hysterically funny," Lazarus recalled. "We pointed out that they had not yet developed anything for catching the fan-fold paper."
One of the biggest leaps forward in computing that came with the Stretch machine was a "look-ahead" function that began fetches of future instructions while decoding data in the current instruction set. Among other achievements, IBM's Werner Buchholz of the Stretch team coined the word byte, Lazarus said.
About 1963, a team from Control Data Corp. visited Los Alamos and praised the designs of the legendary Seymour Cray. For once, Lazarus said, the reality matched the hype. He called Cray a "true genius."
The Laboratory began buying Cray machines, starting with several CDC 6600s in 1966 and progressing through a long relationship with Cray and the company he started in 1972, Cray Research.
"We essentially went with him," when he started Cray Research, Lazarus said.
Early Cray designs featured 65 kilobytes of memory and the first versions of Reduced Instruction Set Computing, or RISC. The key to success of the CDC machines and later Cray computers, he said, were the discrete packaging of components and clever cooling technologies.
Lazarus stepped down from management and returned to working as a computer user in 1973 in part because of the transition from batch computing to timesharing, a technique he said he knew nothing about.
Similarly, the job of Los Alamos managers changed drastically during the 1970s, Lazarus said, from "management by scientists who facilitated the work," to a more formal management structure whose priorities and decisions are strongly influenced by the federal government.
He traced that change to "Black Monday" in 1970, when the Nixon administration ordered a 10-percent cutback in all federal agencies, and the Laboratory suffered the first layoffs in its history.
That period marked another major change in computing, Lazarus said. In the early days, code writers, computer engineers and later computer operators were seen as a cheap way to maximize the utilization of very expensive machines. Today, computers are viewed as a means to the end of maximizing the utilization of expensive labor.
Asked how computing has changed, Lazarus recalled how success for code writers used to depend on getting weapons designers to simplify their designs for ease of calculations. He added that he knows how to use the Google search engine, but doesn't know the details of how his PC works.
"People are much more distant from their computers, but the computers are much more reliable," he concluded.