Computers are increasingly specialized both in hardware and in software. One might even say they are becoming differentiated in a manner analogous to biological cells. Routers are specialized for bandwidth. Data base servers are specialized for I/O, caching, and query. Web servers are specialized for small transactions and throughput. High power parallel processing engines such as IBM\92s 'Watson' are specialized for massively parallel Hadoop operations. Portable devices such as PDAs, cell phones, and MP3 players are specialized for user-friendly interfaces, low power and long battery life. Game boxes are specialized for rapid graphics calculations. And the many embedded devices such as those in cars are further specialized for reliability, durability, precise real-time control and the like.
Specialization in computing is possible because the various roles computers play in modern life have become more specialized. The role of a router is nothing like the role of an iPhone. Specialized roles allow manufacturers to produce software or hardware that is best suited to the specific roles. The computing industry continues to evolve to provide more options to support more specialized needs at different costs. For example, the specialized needs of a game box have driven the development of very fast graphics processors. Market forces have brought forth CPU chips that vary in cost from less than a dollar to several hundred dollars depending upon speed, function, power usage and the like. Different types of memory, disk, display, and communications are also available at different prices.
The growing multicellularity of computing also facilitates
specialization. A specialized computer can rely on others via
network connections for services it does not provide itself. No
longer does one computer have to do everything. That's the point
Oriented Architectures (SOA).
Specialization in computing is necessary for three reasons. First, many of the specialized requirements are incompatible. A PDA or cell-phone, for example, must run on minimum battery power whereas a computation engine must expend power with abandon in order to maximize FLOPS. Second, excess generality, especially in software, imposes the excess costs of a larger software engineering team. Software that supports a larger than necessary set of functions also lengthens time-to-market, is almost inevitably more buggy, and thus requires a larger customer support staff and risks customer dissatisfaction. Finally, the more function a system supports, the larger the \93surface area\94 exposed to attack by viruses, worms, spyware, sniffers, and other malware. It is no accident that most "malware" enters our systems by first infecting a general-purpose end-user PC.
At first blush, there appeared until recently to be one glaring
exception to increasing specialization in computing: Windows PCs
attempted to support all possible function. Windows was the pinnacle of single-cell computing,
not the basis of multicellular computing. The cost in increased
complexity nearly crippled Windows Vista which was shunned
by most corporate IT departments. Yet Windows still boasts a
rich ecology of third-party applications that rely on the
Windows marketplace as well as the Windows API. A host of
Windows programmers have, at great cost in time and effort,
learned how to use the Win32 API set. While protecting their
livelihood, these programmers helped to perpetuate the Windows
software ecology and discourage new competitors. So the
transition from Windows to more specialized personal computers
has been slower than it otherwise might have been. It is
happening nonetheless. More specialized devices such as iPads,
iPhones, Android phones, and other smartphones continue to make
Windows desktops and laptops less critical to business. Outside
of business, especially among the young and hip, Windows boxes
seem hopelessly antique. And the "App" ecologies for iPhones and
Android smartphones has drawn the lion's share of new
developers. That's where the money is now.
Microsoft's ambitions have not been the only factor opposing specialization. Biological systems are self configuring, self protecting, and self optimizing in ways that computing systems are not. IBM\92s efforts to create "autonomic" computing systems were an attempt to redress this deficiency. But creating truly self configuring systems is a difficult task. So we still must configure, provision, and maintain our systems by the mostly manual efforts of IT professionals (or gifted amateurs). The more specialization in computing, the more specialization will be required in hard-pressed IT staffs. Until computers are primarily self configuring, corporate staffing inertia will continue to work against specialization.
Economic forces and user needs also tended to counter specialization. A few years ago there was a flurry of interest in what then was called \93Thin Clients.\94 That meant, in effect, a replacement for the ever-present Wintel PC that would be specialized to support the needs, and only the needs, of the knowledge worker. Unfortunately, it turned out to be much harder than expected to decide what needs were common to such workers. More recently we saw the rise of the Netbook, a very inexpensive PC-like laptop somewhat specialized for people who primarily wish to browse the web. That market was expanding until the iPad and iPhone (and its imitators) captivated the public. Now the most popular innovations in mass-market computing emerge first in the iPad/iPhone market rather than the Wintel ecology. Microsoft struggles to keep up.
Now that Microsoft's death grip on mass-market software has been broken,
innovation is blossoming. One result is that specialization is
accelerating in most, if not all, areas of computing, especially
in small low-power personal computing devices and
special-purpose sensor/effector devices. Worldwide shipments of
new cellphones, PDAs and smart-phones already far outstrips
those of Wintel boxes. A one-size-fits-all operating system
simply cannot mutate fast enough to keep up.
Server computing is specializing too. Within the corporate IT
world, the needs of large-data analytics are beginning to come
to the fore. And the huge Web server farms such as Google rely
on stripped-down cheap boxes running specialized operating
systems often based on Linux. It is a new world in which one
size does not fit all.