Beruflich Dokumente
Kultur Dokumente
COMPUTING IN
THE CLOUDS
Powerful services and applications are being
integrated and packaged on the Web in what the
industry now calls “cloud computing”
one is right. There are many shapes But data centers aren’t new, either.
in the clouds. In the dot-com boom of the mid-
’90s, many a startup invested venture
Cloud Shapes: The Data Center capital into traditional enterprise
It is not news that today’s major solutions like Sun SPARC servers. Is
Internet companies have built massive this the cloud?
data centers to power their online Yes and no. Google is often credit-
businesses. Decades ago, computing ed with innovating search on the
power was concentrated in main- Web and, more recently, advertising.
frames tucked away behind the scenes But to many, Google’s architecture
because there was no alternative— behind the scenes has spawned just
only a hulking room-sized box that as significant a revolution.
could contain any significant amount The data centers of the early dot-
of computational power. The idea com era were, in some respects,
that this power could be distributed direct descendents of Watson’s main-
rather than centralized seemed like frames. Physically smaller, perhaps,
such folly that in 1943, IBM but Sun servers and their ilk contin-
Chairman Thomas Watson said ued to represent an exclusive kind of
famously (or infamously) that “I computing—concentrated power
think there’s a world market for designed, and priced, for exclusive
maybe five computers.” customers—namely, enterprise.
The era of the personal computer But Google turned the data
that has flourished since the 1970s center model on its head. Rather
directly contradicted Watson’s predic- than power a network with a
B
ing,” a thin, wide network can recov- esides cheaper power,
er gracefully from the most common data centers are making
ailments, such as connection and heavy use of virtualiza-
hardware failures, because there are tion to squeeze the most
so many more drones available to out of the watts they’re
take on the work. consuming. With major
But a cloud can consume a lot of vendors like VMWare
power to run. Aside from the power and Citrix, which recently acquired
needed to drive thousands, or hun- the Xen virtualization platform,
dreds of thousands, of processors increasingly targeting the data center,
and peripherals—hard drives, cooling virtualization allows a single server
fans—all these whirring machines to run multiple operating instances
generate lots of heat. It is estimated simultaneously. By sandboxing each
that 50 percent of energy costs in OS inside artificial boundaries, not
running a large data center are only can each instance run indepen-
derived from cooling needs alone. dently of the others, but CPU idle
Worse still for the cloud, the world time is minimized.
is immersed in a global energy Just what distinguishes a “cloud”
crunch, as both demand and specula- from “a bunch of machines” can be a
tion has driven up pricing for most little fuzzy. But the next evolution
conventional energies toward record that may illuminate the fog is the so-
levels. called “data center OS”—or, in the
Reducing the operating costs of a spirit of the buzz, the CloudOS. In
Google-inspired data center cloud the fall of this year, VMWare and
involves both physical and virtual Cisco announced a joint venture to
solutions. Physically, data centers are develop such a “fabric” (to use their
like plants arcing toward the sun- word in spite of the mixed
light, migrating toward cheap energy. metaphor).
Google is building a major data cen- In a data center like that employed
I
requires an abstraction layer of soft- deally, if tasks are broken into
ware—often, highly specialized, cus- their smallest constituent jobs,
tom software—that intelligently and each job could be complet-
divvies up jobs. But a more efficient ed simultaneously using avail-
and resource-friendly solution would able processing resources
be a single operating system, which somewhere in the cloud, you
intrinsically utilizes the resources of could achieve an optimally effi-
many machines. cient architecture: the most optimistic
Essentially, an operating system is definition of distributed computing.
designed to manage resources—hard In reality, some jobs are dependent
drive space, memory, and so on. A on the results of other jobs.
true data center, or cloud, OS will Furthermore, designing algorithms to
treat every processing unit available most effectively divide jobs and dis-
as just another resource, relying on tribute them throughout the cloud in
networking channels to replicate the real time is complex, to understate
kinds of intra-server channels that the case.
now coordinate events within a single But distributed computing, like the
physical machine. Under the com- data center itself, is not inherently
mand of a single “omniscient” oper- new to the era of the cloud. If we
ating system, the cloud becomes a consider the entire Internet a cloud,
more cohesive entity. one need only look at popular pro-
jects like SETI@home and
Cloud Shapes: Distributed Computing Folding@home to see public exam-
Whether “the cloud” represents a ples of distributed computing at
data center at a single physical loca- work. In these projects, individuals
tion or dozens, hundreds, or thou- run software on their PC which con-
sands of data centers spread around nects them to a server that divides
the world, its speed and efficiency is large jobs among small clients to
limited by how intelligently it dele- crunch numbers toward a particular
gates responsibility. goal—in these examples, searching
Completing any general computing for alien life among radio waves and
task—say, retrieving the results of a computing protein folding simula-
search with relevant contextual tions to aid medical research. You
advertisements—requires a long could even consider botnets—soft-
series of smaller jobs to be complet- ware that maliciously infects unwit-
ed. Database queries, parsing of ting client machines to send out
results, construction of result sets, parcels of spam—a form of nefarious
and formatting of result pages, to distributed computing.
name the most common. Even these The open source project Hadoop
tasks can be further broken down provides a general-purpose frame-
A
considerable concern been the rage for several years.
among major enterprise Fueled by technological evolutions
is that today’s computer like AJAX, which allows browser-
science students lack based code to behave more like a
access to the distributed local application, and people’s desire
computing environ- for mobility and data ubiquity, we
ments that make up the increasingly use the Web for applica-
cloud. Companies like Google will tion functionality. E-mail was both
need new hires experienced in writing the first “killer app” for the Internet,
code designed to run in a cloud of and later, on the Web. For many
perhaps thousands of processors. But today, Web-based e-mail is the only
educational institutions don’t have kind they use.
F
or cloud computing to those without an access device from a
move front and center, the single brand. Some fear that the
networks that tie every- cloud could encourage the growth of
thing together need to be walled-gardens, a potential step back
extremely robust. After compared to the relatively open
all, the cloud represents a Internet of today.
significant inversion from
personal computing in the 1970s and Partly Cloudy?
1980s, when machines stood alone There are those who scoff that
and derived all of their utility from “cloud computing” is just the latest
their own capabilities. Under the branding of some old, familiar com-
cloud, client machines become nearly puting models—which is partly true.
useless on their own. Are our net- It’s a buzzword almost designed to be
works ready to handle the load? vague, but cloud computing is more
Many would say no, especially than just a lot of fog.
those living in the large U.S. market, The cloud concept draws on many
where broadband quality and pene- existing technologies and architec-
tration fares poorly relative to many tures. Centralizing computing power
smaller industrialized nations. In con- is not new and, if anything, is a
trast to local computing power, which return to the computer’s roots. Nor is
continues to boom year after year, utility computing new, or distributed
broadband advances here have been computing, or even software as a
slow and may create at least a short- service.
term bottleneck holding back a major But the cloud is new in that it inte-
shift toward mainstream cloud com- grates all of the above computing
puting. models. This integration requires
Network bandwidth aside, the computing’s center of power to shift
cloud raises concerns among privacy from the processing unit to the net- PERMISSION TO MAKE DIGITAL OR HARD
advocates. Most significantly, the work. Inside the cloud, processors COPIES OF ALL OR PART OF THIS WORK FOR
Significant amounts of data which work that holds the cloud together, ARE NOT MADE OR DISTRIBUTED FOR PROFIT