Supercomputers,
A History Of
By: Andrew
Danilovic
Modern
electronic computers are extremely pervasive in today’s society. They are used
by both large corporations and private individuals for a variety of purposes. A
somewhat smaller sub-branch of the modern computer is the supercomputer.
The
supercomputer is a computer system that provides extremely fast performance as
compared to the average home computer. Research labs and large corporations are
the main users of such computing systems and employ them to solve a variety of
problems. The reason supercomputers are so helpful in solving very large
problems is because they can analyze mathematical models of the real world,
such as differential equations, and give results much faster than would
otherwise be available. Some of the main areas supercomputers are useful are
physics, biology, materials science, medicine, climate modeling and
intelligence data (Graham, et al.).
A
discussion of the history of supercomputing must begin with a diatribe of the
early corporate mergers and business acquisitions that resulted in the modern
computer industry of the latter part of the 20th century. A few
major historical events, namely WWII and the Cold War, created a ready marketplace,
i.e. the U.S. government, which allowed these early corporations to gain the
necessary funding for future research in supercomputing.
During
World War II, the navy had a codebreaking facility located on Nebraska Ave. in
Washington D.C. The codebreakers there were able to break the Enigma code
German U-boats used to communicate with bases in Germany. Needless to say, this
was very valuable information to the United States. After the war ended,
though, a few of the top engineers at the Nebraska Avenue facility, such as
William Norris and Howard Engstrom, wanted to pursue the commercial
applications of the codebreaking machines. The U.S. Navy could not order the
men to remain at the Nebraska Ave. facility, but they knew that such work would
continue to be valuable because of the looming threat posed by the Soviet
Union. So, top Navy officials decided to help Norris and Engstrom start their
business by drumming up support in the private sector. This would allow the
government to later call upon their skills for important, national security
research. The Navy even went so far as to have Chester W. Nimitz meet with John
Parker, the owner of the Northwester Aeronautical Corporation. Parker decided
to help fund the new venture, which ended up being named Engineering Research Associates,
or ERA for short. Before this, it was actually very difficult for Norris and
Engstrom to find outside investors to fund their project, because many of the
industry leaders they talked to did not believe there was a future in
electronic machines.
One
of the main attributes of ERA, which we will see also played an important role
for Control Data and other supercomputing corporations, was the lack of
conformity and cynicism that often characterized much of the corporate culture.
There was instead a sense of informality that allowed the engineers there to
work creatively (Murray, 26).
In
late 1951, Parker decided to sell his stake in ERA to Remington Rand, now Unisys, a
corporation mostly known for selling typewriters. Parker’s decision to sell was
based on the need to help the young ERA eliminate its debt and acquire the
funding it needed to pursue more advanced research. This sale occurred after an
earlier Remington Rand acquisition of the Eckert-Mauchly Corporation, best
known for its development of the ENIAC,
or Electronic Numerical Integrator and Computer. The ENIAC was first developed
to calculate artillery firing tables for the U.S. Army. Though both groups were
now under the umbrella of Remington Rand, it soon became clear that different
ideologies were present in each. The ERA engineers could be characterized as
being more practical while the Eckert-Mauchly group were more of the
theoretical type. An example of this would be the ENIAC and the UNIVAC 1
computers, both products of the Eckert-Mauchly group, which were plagued by
downtime. One reason for this is though the engineers in the Eckert-Mauchly
group had many brilliant ideas, they could rarely implement them in a timely
fashion (Murray, 37). This resulted in lower reliability because, many times,
they would try and incorporate new theories into nearly finished designs. What
Norris, the future chair of the Control Data Corporation, would take away from
of these years of working under Remington Rand and alongside the Eckert-Mauchly
group, was the importance of balancing cutting edge research and practical, reliable
design (Murray, 40).
In
July, 1957, Norris and Frank Mullaney, one of the engineers who worked with
Norris, left then Sperry-Rand to form Control
Data Corporation. CDC would be one of the major innovators in supercomputer
design. One of the reasons for Norris and Mullaney’s split was because of the
lack of control the former ERA engineers had over they own designs and plans.
They also felt that after being acquired by Remington Rand, their working
environment changed to that of a stifling corporate culture.
One
of the key engineers working at CDC was Seymour Cray, who first joined the
company in 1958. Seymour Cray is considered by many to be the “father of
supercomputing” (Cray Inc., Cray History). He was
able to design the 1103’s control system almost entirely from scratch. The 1103
was one of the first computers he worked on while at CDC. In the late 1950s and
1960s, Cray was able to fashion high-powered, reliable supercomputers using low
quality, 37 cent transistors, a device used amplify electronic signals.
The
Cold War, like WWII, also played a very important role in the development of
supercomputers and the corporations that build them. Because of the threat of
Soviet aggression, the United States Government decided to supply much of the
funding for the development of Seymour Cray’s supercomputers. The need for such
high-powered computing was readily observable by scientists and researchers
working on the problem of nuclear weapon design in the early 1950s at the
Livermore Branch of the University of California Radiation Laboratory (Murray,
60). Because experiments involving nuclear weapons were extremely difficult and
dangerous to pull off, the scientists working on the problem needed extremely
fast computers to process theoretical data and give meaningful results which
lowered the need for such experiments.
The
need for high powered computing at two government labs in the mid 1950’s, one
in Los Alamos National Laboratory in New
Mexico and the other at the Lawrence Livermore
National Laboratory in California, created a great opportunity for many of
the computer manufacturers at the time. Because of the talent of Seymour Cray
and the other engineers working at CDC, and because of the creative and relaxed
environment there, the company was able to capitalize on the market demand for
supercomputers. Their first big success came with the CDC model 1604. The 1604
was the fastest computer available at the time of its production in 1960
(Murray, 70). CDC next produced the 6600 model in 1966, which was their first
comprehensive, full-fledged supercomputer. One of the reasons behind the
success of the 6600 was its ability to increase the speed at which computations
were done while doing so cost-effectively.
Here,
the performance versus the year of introduction is given for a number of
computer systems. The performance is measures in MOPS, or millions of
operations per second. We can note here that the CDC models 6600 and 7600
performed better than some of the systems produced by IBM, which was one of the largest computer companies for
most of the 20th century. In
1972, because of disagreements with management, Seymour Cray left CDC to
form Cray Research, which would also become a major player in the
supercomputer industry. CDC and Cray research dominated the supercomputer
market from the early 1960s, and were actually able to dominate the global
supercomputer industry up until the 1980s (Graham, et al.). The Cray 1,
developed by Cray Research and first shipped to Los Alamos National
Laboratory in 1976, was one of the first supercomputers to support what is
called vector architecture. The term architecture, as applied to computer
systems, means operational structure. This type of computer construction
quickly became the industry-accepted format for supercomputers at the time. The
next big shift in the computing industry came when Japanese semiconducting
companies began their push to the technological forefront of innovation.
There was a strong backing by the Japanese government to fund research and
development projects in the early 1980s. Two of these projects were called
the Fifth Generation Computer System project and the SuperSpeed project.
Referenced from Getting Up to
Speed, The Future of Supercomputing, figure 3.1
This
competition from Japanese computer manufacturers caused the U.S. government
concern and compelled DARPA, the
Defense Advanced Research Projects Agency, to create the SCI, or the Strategic
Computing Initiative, in the mid 1980s. To the left, we can see just how
big of a market share was held by CDC, Cray Research and three of the top
Japanese supercomputer manufacturers NEC, Hitachi and Fujitsu. This concern
over strong competition from Japan came about because the U.S. government
viewed supercomputing as an extremely important factor in the superiority
of U.S. weapons systems. A direct impact of the DARPA initiative was the
manufacture of non-vector supercomputers. The reason for this was the idea,
held by DARPA and the United States government, that U.S. companies could
not overtake the Japanese in computing power and speed by merely
introducing faster components. This resulted in a push to create a new
supercomputer architecture (Graham, et al.). The new supercomputers that
were designed after the introduction of the DARPA initiative made use of a
technique called parallel-processing. This type of architecture involves a
large number of processors connected together to create a single computing
system. While possible orders of magnitude increase in speed is possible
with this type of architecture, one drawback is the difficulty in writing
programs and operating systems to make efficient use of the many processors
included in the system. The
first of such systems were built using custom designed and built hardware.
In the early 1990s, however, a shift to the use of COTS, or commercial off
the shelf, components began. One of the more well known designs of a
commodity based computer system is the Beowulf
computer system, sometimes referred to as a cluster. This system was first
developed in late 1993 by Donald Becker and Thomas Sterling while working
at NASA. The reason Beowulf systems are
so popular is because of the reduced cost of building them. This is again
because they are designed using mostly, if not only, COTS components. A
useful resource for anyone studying the supercomputer industry is the Top500 list, which was a project
started in 1993 by Hans Meuer of the University of Mannheim, Germany, in
association with researches from the U.S. The purpose of this list is to
rank the 500 most powerful computer systems in the world. The most recent
update to the list in November 2007 shows the IBM BlueGene/L System as the
most powerful computing system in the world. Below, we can see a list of
supercomputers from the Top500 project by country of maker. The
importance of supercomputing to scientific discovery is evident in its
ability to produce meaningful results in a variety of applications. Over
the years supercomputing has become faster and more reliable, though there
continue to be ways to improve the performance and efficiency of the
supercomputer. Because of this, supercomputing will continue to be a main
research area in the field of computing.
Referenced from The Top500
Project: Looking Back Over 15 Years of Supercomputing Experience,
figure 1
Referenced from Getting Up to
Speed, The Future of Supercomputing, figure 3.7, and was compiled using
data from the Top500 list
Here we can see the different types
of operating systems currently being used on supercomputers. Referenced
from The Top500
Project: Looking Back Over 15 Years of Supercomputing Experience,
figure 14 References LLNL.gov.
Barney, Bliase. “Introduction to Parallel Computing.” September 9,
2007. March 13, 2008 http://www.bedfordstmartins.com/online/cite5.html#1 Beowulf.org.
Merkey, Phil. “Beowulf History.” 2007. March 20 2008 http://www.beowulf.org/overview/history.html Meuer, Hans Werner.
“The TOP500 Project: Looking Back Over 15 Years Of
Supercomputing Experience.” Janurary 20, 2008. March 12, 2008<http://www.top500.org/files/TOP500_Looking_back_HWM.pdf> This is an article from TOP500.org. Works
Cited Graham, Susan L. et
al. ED. “Getting Up to Speed, The Future of Supercomputing.” 2005. National Academic Press. March 9, 2008 <http://books.nap.edu/html/up_to_speed/notice.html>.
Murray, Charles J. The
Supermen: The Story of Seymour Cray and the Technical
Wizards Behind the Supercomputer. New York: John Wiley &
Sons, Inc., 1997. Cray Inc. Homepage.
“Cray History.” http://www.cray.com/about_cray/history.html This list was
prepared using the general outline of: Author's name (last
name first). Document title. Date of Internet publication. Date of access <URL>. This
webpage was created by: ·
Andrew Danilovic ·
CSE 301 – History of
Computing ·
Spring 2008