Unlike most of its competitors, Amazon builds its hardware infrastructure
from commodity components. Commodity, in this case, refers to using equipment
from lesser-known manufacturers who charge less than their brand-name
competitors. For components for which commodity offerings aren't available,
Amazon (known as a ferocious negotiator) gets rock-bottom prices.
 |
| Amazon Web Services (AWS) Hardware |
On the hardware side of the AWS offering, Amazon's approach is clear: Buy
equipment as cheaply as possible. But wait, you may say, won't the commodity
approach result in a less reliable infrastructure? After all, the brand-name
hardware providers assert that one benefit of paying premium prices is that you
get higher-quality gear.
Well . . . yes and no. It may be true that premium-priced equipment
(traditionally called enterprise equipment because of the assumption that large
enterprises require more reliability and are willing to pay extra to obtain it)
is more reliable in an apples-to-apples comparison. That is, an enterprise-grade
server lasts longer and suffers fewer outages than its commodity-class
counterpart.
The issue, from Amazon's perspective, is how much more reliable the
enterprise gear is than the commodity version, and how much that improved
reliability is worth. In other words, it needs to know the cost-benefit ratio of
enterprise-versus-commodity.
Making this evaluation more challenging is a fundamental fact: At the scale
on which an Amazon operates (remember that it has nearly half a million servers
running in its AWS service), equipment — no matter who provides it — is breaking
all the time.
If you're a cloud provider with an infrastructure the size of Amazon's, you
have to assume, for every type of hardware you use, an endless round of crashed
disk drives, fried motherboards, packet-dropping network switches, and on and
on.
Therefore, even if you buy the highest-quality, most expensive gear
available, you'll still end up (if you're fortunate enough to grow into a very
large cloud computing provider like, say, Amazon) with an unreliable
infrastructure.
Put another way, at a very large scale, even highly reliable individual
components still result in an unreliable overall infrastructure because of the
failure of components, as rare as the failure of a specific piece of equipment
may be.
The scale at which Amazon operates affects other aspects of its hardware
infrastructure as well. Besides components such as servers, networks, and
storage, data centers also have power supplies, cooling, generators, and backup
batteries. Depending on the specific component, Amazon may have to use
custom-designed equipment to operate at the scale required.
Think of AWS hardware infrastructure this way: If you had to design and
operate data centers to deal with massive scale and in a way that aligns with a
corporate mandate to operate inexpensively, you'd probably end up with a
solution much like Amazon's. You'd use commodity computing equipment whenever
possible, jawbone prices down when you couldn't obtain commodity offerings, and
custom-design equipment to manage your unusually large-scale operation.
