Microsoft to support Azure with data center investments
Despite the economic downturn, Microsoft Corp. intends to ramp up the number of servers running in its data centers worldwide by 15 times over the next 5 years.
The growth, outlined in a presentation on Monday at its Professional Developers Conference, is designed to handle increased hosted computing demand from enterprise software running on its new Windows Azure platform, also announced Monday, as well as third-party services Microsoft hopes to attract.
Microsoft expects to boost the number of data centers it operates by three times, its power usage by 15 times, and the Internet traffic going out of its data centers by nine-fold, said Benjamin Ravani, general manager of Microsoft's Global Foundation Services, during a technical session.
Ravani said Microsoft operates "tens of thousands of servers" but would not disclose the exact number.
Microsoft had announced similar growth projections earlier this year. But Ravani's reiteration of those comments come a week after Redmond announced plans to tighten its fiscal belt, including cutting US$500 million in spending this fiscal year by slowing hiring and cutting travel and marketing expenses.
Despite its belated arrival to so-called cloud computing services, Microsoft appears to be sparing no dime on building out a back-end infrastructure that tops competitors such as Amazon.com Inc., Google Inc. and Salesforce.com Inc.
Microsoft has announced five data centers in the past 12 months, including in San Antonio, Tx., Chicago, Il. and Des Moines, Ia. Both its Chicago and Des Moines data centers will be massive, $500 million facilities that will have many of its servers pre-configured and installed in shipping containers.
Investing in a big way now, Microsoft has argued, will save money later.
Demand for some services is already huge. Microsoft's Windows Live Messenger Monday has more than 450 million unique users in the system, passes more than 8.3 billion messages and performs more than 1 billion Web authentications a day, said Ravani.
Besides touting the scale of Microsoft's cloud infrastructure, Ravani also touted the resiliency of its network, which he said was the result of several hard-earned lessons.
For instance, in November 2006, half of a million users of a Microsoft online service experienced authentication delays for several hours. The problem was an overload of the authentication system caused by a poorly-written internal batch job. As a result, Microsoft created a policy that all batch jobs need to be tested first, and increased security, so that users can't access key data services without authorization.
In a March 2007 case, 75% of users for Microsoft's online services were unable to log in for 5 hours. Microsoft discovered that a bug with a partner's service had also resulted in too many authentication requests. Ravani described it as a "denial of service attack on Windows Live ID."
Microsoft improved its monitoring to detect such service failures more quickly, developed fine-grained throttling at many different levels of its architecture to prevent out-of-control traffic, and implemented a policy says "your transactions get dropped if they exceed quota, so you don't eat other partners' lunch."