Cloudonomics: The Business Value of Cloud Computing, + Website

CHAPTER 19

Location, Location, Location

Every realtor knows that the three most important characteristics of any property are, in priority order: location, location, and location. This is important not only in residences, but logistics centers, malls, factories, coffee shops, and of course, data centers. Tim Harford, author of The Undercover Economist, pointed out:

Starbucks’ most significant advantage is its location on the desire line of thousands of commuters. There are a few sweet spots for coffee bars—by station exits or busy street corners. Starbucks and its rivals have snapped them up. If Starbucks really did have the hypnotic hold over its customers that critics complain about, it would hardly need to spend so much effort getting people to trip over its cafés. The nice margin that Starbucks makes on their cappuccinos is due neither to the quality of the coffee nor to the staff: it’s location, location, location.1

In other words, when time is of the essence, we can either do what New York Stock Exchange Euronext did, and bring the users closer to the processing, or do what Starbucks and Web sites do, and bring the Java closer to the end user. However, since users are globally dispersed, service nodes must also be globally dispersed. But how many nodes is enough: 5? 10? 20? 20,000?

In this chapter, we explore how increasing the number of service nodes reduces latency, but with diminishing returns.