Hardware Reference
In-Depth Information
is often to build a data center with hundreds, or even thousands, of servers. The in-
coming requests are then sprayed among the servers to allow them to be processed
in parallel. For example, Google has data centers all over the world to service
search requests, the largest one, in The Dalles, Oregon, is a facility that is as large
as two (U.S.) football fields. The location was chosen because data centers require
vast amounts of electric power and The Dalles is the site ofa2GWhydroelectric
dam on the Columbia River that can provide it. Altogether, Google is thought to
have more than 1,000,000 servers in its data centers.
The computer business is a dynamic one, with things changing all the time. In
the 1960s, computing was dominated by giant mainframe computers (see below)
costing tens of millions of dollars to which users connected using small remote ter-
minals. This was a very centralized model. Then in the 1980s personal computers
arrived on the scene, millions of people bought one, and computing was decent-
ralized.
With the advent of data centers, we are starting to relive the past in the form of
cloud computing
, which is mainframe computing V2.0. The idea here is that
everyone will have one or more simple devices, including PCs, notebooks, tablets,
and smartphones that are essentially user interfaces to the cloud (i.e, the data cen-
ter) where all the user's photos, videos, music, and other data are stored. In this
model, the data are accessible from different devices anywhere and at any time
without the user having to keep track of where they are. Here, the data center full
of servers has replaced the single large centralized computer, but the paradigm has
reverted back to the old one: the users have simple terminals and data and comput-
ing power is centralized somewhere else.
Who knows how long this model will be popular? It could easily happen in 10
years that so many people have stored so many songs, photos, and videos in the
cloud that the (wireless) infrastructure for communicating with it has become com-
pletely bogged down. This could lead to a new revolution: personal computers,
where people store their own data on their own machines locally, thus bypassing
the traffic jam over the air.
The take-home message here is that the model of computing popular in a given
era depends a lot on the technology, economics, and applications available at the
time and can change when these factors change.
Now we come to the mainframes: room-sized computers that hark back to the
1960s. These machines are the direct descendants of IBM 360 mainframes ac-
quired decades ago. For the most part, they are not much faster than powerful ser-
vers, but they always have more I/O capacity and are often equipped with vast disk
farms, often holding thousands of gigabytes of data. While expensive, they are
often kept running due to the immense investment in software, data, operating pro-
cedures, and personnel that they represent. Many companies find it cheaper to just
