5. Steve Lohr, “For Impatient Web Users, an Eye Blink Is Just Too Long to Wait,” New York Times (2012), http://
nyti.ms/1esukXm
6. Greg Linden, “Make Data Useful,” Amazon.com and Findory, http://bit.ly/1k7ypZw (PowerPoint file down‐
load).
nect to a single ecommerce platform. An outage now is the equivalent of barring cus‐
tomers from entering all physical stores.
Because of increasing competition and the maturity of offerings, customers are in‐
creasingly fickle about performance. They expect response times to be instant. The New
York Times recently said: “The old two-second guideline has long been surpassed on
the racetrack of Web expectations.”
5
Going on further to say: “Two hundred fifty mil‐
liseconds, either slower or faster, is close to the magic number now for competitive
advantage on the Web.” Amazon.com saw a 1% increase in revenue for every 100 mil‐
liseconds of response-time improvement.
6
In today’s world, milliseconds matter.
Availability and performance are becoming increasingly difficult to offer as traffic has
become more prone to rapid spikes due to an increasing reliance on promotions and
marketing-driven events. We’ll discuss this more later, but it’s not uncommon to see
spikes in traffic that are one or two orders of magnitude above steady state. Social media–
based marketing can lead to campaigns going viral. From an IT administrator’s stand‐
point, the traffic can come so quickly that it looks like a distributed denial-of-service
attack, when in reality it’s likely to be a few million kids hitting refresh on their pages
in anticipation of the release of the latest hot basketball shoe.
While ecommerce has been maturing over the past two decades, the prevailing deploy‐
ment architecture looks largely as it did in the beginning—mostly static environments
fronted by web servers deployed out of a single data center. Many simply guess at what
their peaks will be and then multiply that number by five for safety. Hardware is statically
deployed and idle for most of the year. It’s been done this way for four reasons:
• IT administrators fear losing their jobs because of outages. It’s simply less risky to
throw hardware at problems.
• For a while, ecommerce deployments were small enough that the hardware cost
was negligible.
• There hasn’t been a good alternative to the static approach—cloud in its present
form didn’t exist until very recently, and it’s matured only recently.
• The old models of hosting had more accountability. If there was an outage, you
could always escalate to your vendor.
The current approach to ecommerce deployment architecture is not scalable. The rise
in traffic has ballooned environments from dozens to hundreds or even thousands of
servers. Given today’s extremely competitive business climate, it’s not feasible to have
xvi | Introduction
我的内容管理
展开
