More about Microsoft’s Data Centers
November 23, 2008
Business Week has tilled an old field with its “Microsoft to Google: Get Off My Cloud” by Peter Burrows. You must read here Mr. Burrows’ description of Microsoft’s initiative to build out its data centers. The multi year effort under the firm hand of the national basketball association’s (if I understood Mr. Burrow’s undefined acronym) former technology wizard, Debra Chrapaty. Mr. Burrows picks up the story in 2003 when Microsoft started its initiative to build “20 supersize data centers that can cost as much as $1 billion apiece,” Mr. Burrows informed me that:
Rather than spend hundreds of hours opening server boxes, and connecting them with cables and loading them with software, Microsoft can roll in a container in just a couple of days. The hope is to run the facility with half as many people as at its previous sites. Even better, it’s easier to monitor and whisk away heat generated in these confined modules than cooling an entire building. One source says Microsoft hopes the design will help cut by one-third the power bills that typically take up some 40% of a site’s operating cost.
Mr. Burrows talks to a Googler who points out that Google’s been doggedly building infrastructure for a decade. The only issue I had is that Mr. Burrows did not seek out one of Google’s infrastructure experts, settlling for a marketing type in my opinion, not a Google research wizard like Jeff Dean of Simon Tong. I guess one Googler is as good as another for Business Week’s fact checkers. The Google source quoted in Mr. Burrows’ article understates what Google has in place. From Google’s point of view understatement is part of the game plan. For the reader, understatement makes Microsoft’s investments and chances look quite amazing.
The story ends with a contrast between Google’s and Microsoft’s approach to secrecy. Microsoft talks about what it is doing. Microsoft talks to Business Week, hence the story. Microsoft talks to Dell, which I am not sure makes Hewlett Packard too happy. Microsoft talks to developers, users, and even addled geese like me.
In my opinion, talk is cheap, a lot cheapter than buckling down and addressing the issues that Google has confronted and resolved to its satisfaction; for example:
- Google has addressed the need for speed in Web type applications. In fact, the GOOG can sort a petabyte of data in a few minutes, and that’s not easy to do even on Microsoft’s whizzy new super computers.
- Google has jumped over the database problems with its dataspace initiative, Chubby, and some innovations in tokenization.
- Google has long been a driver in commoditization and packaging of low cost computing horsepower and lights out data center operation.
I cover more of Google’s innovations from the period 1998 to 2006 in my Google studies. For our purposes in this opinion piece, look at the time span. Google’s been working for a decade, and now Microsoft is starting according to this article later in the game. As a result, Microsoft has to spend money to close the gap. Unfortunately, the closing of the gap does not include confronting the performance issues that will persist in Microsoft’s new data centers. These issues are related to the Microsoft architecture and server software itself.
One quick example. Two load balance, Microsoft has a clever design. The only hitch is that additional servers are needed to prevent bottlenecks at the points originally designed to reduce bottlenecks. The result is a need for additional servers. Does this information mean anything to Business Week? Nope.
It means something to Google, and it will mean something to Microsoft’s financial officers when the costs for resolving some fundamental architectural problems keep Microsoft’s data center investments in the category “hard to control.” If you have other information, please, use the comments section to put me on the true path to understanding. I have the Microsoft technical papers in front of me, and these are indeed quite interesting with regards to core engineering issues in the Microsoft data center initiative. Hardware, gentle reader, can’t solve some performance issues. That’s what makes Google a formidable competitor: software engineering, not commodity hardware.
Stephen Arnold, November 23, 2008
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
