34 Part I • Information Technology
The demand for mainframe computers goes up and
down over time, varying with the price–performance ratios
of mainframes versus servers, with the introduction cycle
of new machines, and with the health of the economy.
Some commentators keep suggesting that the mainframe
will disappear, but that doesn’t appear to be happening. In
recent years, IBM and other vendors have introduced new
technology, added Linux options to proprietary operating
systems, and slashed prices drastically. The addition of
Linux capability has been particularly important in the
twenty-first century resurgence of the mainframe, with
many companies finding out that it is more economical to
run multiple virtual servers on a single mainframe than to
run a large number of Intel- or AMD-based servers. In one
example, an IBM customer redeployed 800 servers onto 26
of the 64 processors of a single z10 mainframe, leading to
a “tremendous saving in software licensing costs, power,
cooling, and space” (Adhikari, 2008). The role of the
mainframe will continue to evolve as we move further into
the twenty-first century, with more emphasis on its roles as
keeper of the corporate data warehouse, server in sophisti-
cated client/server applications, consolidator of smaller
servers, powerful Web server, and controller of worldwide
corporate networks.
Supercomputers
Supercomputersare the true “number-crunchers,” with
MFLOPS ratings in excess of 250,000 and price tags from
$1 million to $100 million or more. The high-end supercom-
puters are specifically designed to handle numerically inten-
sive problems, most of which are generated by research
scientists, such as chemists, physicists, and meteorologists.
Thus, most of the high-end supercomputers are located in
government research laboratories or on major university
campuses (even in the latter case, most of the machines are
largely supported by grants from the National Science
Foundation or other government agencies). Midrange super-
computers, however, have found a variety of uses in large
business firms, most frequently for research and develop-
ment efforts, Web serving on a massive scale, data mining,
and consolidating a number of smaller servers.
Until the mid-1990s, the acknowledged leader in the
high-end supercomputer arena was Cray Inc. However, IBM
mounted a concerted effort in supercomputers in the 1990s
and 2000s, and IBM now clearly holds the top spot in terms
of numbers of large supercomputers—but Cray Inc. has
reclaimed the overall top spot (see the box “World’s Fastest
Supercomputer: Jaguar”). In the November 2009 listing of
the world’s top 500 supercomputers, the top 100 machines
were distributed as follows: IBM 33, Cray Inc.14, Silicon
Graphics International (SGI) 12, Hewlett-Packard and Sun
Microsystems 8 each, and 10 other vendors with 4 machines
or less, plus 3 other machines that were built by a combina-
tion of vendors (two of these including IBM) (Top 500,
2009). These large computers use one or more of three high-
performance computer architectures, all of which involve a
large number of processors: parallel vector processing, mas-
sively parallel processing,^5 and symmetric multiprocessing.
As an example of a machine which uses both massively par-
allel processing and symmetric multiprocessing, the IBM
Blue Gene/P computer named JUGENE, located at the
German research center Forschungszentrum Juelich, incor-
porates 73,728 compute nodes, each of which contains 4
symmetric multiprocessors, for a total of 294,912 proces-
sors—which makes JUGENE the fastest computer
in Europe. This configuration has 144 terabytes of main
memory and operates at speeds up to 1 petaflop (quadrillion
floating point operations per second) (IBM, 2009a).
In addition to the vendors mentioned earlier, a trio of
Japanese firms—NEC, Hitachi, and Fujitsu—are also
important vendors of midrange supercomputers. An inter-
esting development in the supercomputer arena occurred in
1996 when Silicon Graphics, Inc., acquired Cray Research,
thus becoming (for a time) the world’s leading high-
performance computing company. Cray Research contin-
ued to operate as a separate unit, focusing on large-scale
supercomputers. Then in 2000, Tera Computer Company
purchased Cray Research from Silicon Graphics, Inc., with
the combined company renamed Cray Inc. In 2009, when
SGI was about to file for bankruptcy (for a second time), it
was purchased by Rackable Systems, which promptly
renamed itself Silicon Graphics International (to be able to
keep the SGI label). In the supercomputer arena as in other
areas, sometimes it is hard to keep up with the players!Key Types of Software
We have completed our overview of computer hardware,
ending with a discussion of the types of computers in use
today—from PCs and smartphones through supercomput-
ers. Now we turn to software, the programs that control
the operations of computer systems. We began our consid-
eration of software when we introduced machine lan-
guage programming in “The Stored-Program Concept”
section earlier in this chapter. All computers use the
stored-program concept; on all computers, a machine
language program is loaded in memory and executed by(^5) Aparallel processoris a multiprocessorconfiguration (multiple proces-
sors installed as part of the same computer system) designed to give a sepa-
rate piece of the same program to each of the processors so that work can
proceed in parallel on the separate pieces. A massively parallel processor
has a very large number of parallel processors, often over 1,000.