% fortune -ae paul murphy

The meaning of Microsoft Server 2008

Server 2008 looks like a bit of an odd duck - it doesn't meet the promises made for "Longhorn" and "Blackcombe", it's probably neither more reliable nor more efficient than its 64 bit 2003/XP based predecessors, and unlike Vista (with which it shares some code) the kernel changes amount to rather more than just another point release in the NT schedule.

What I think happened with it is that Microsoft needed a new release for marketing reasons, but changed the goals for that release somewhere around two years ago. Originally the goals would have been in line with naive expectation: basically to advance the technology while protecting backwards compatibility and patching a few outstanding performance, reliability, and security problems. During the change, however, I think they dropped the technology goals in order to implement an extremely defensive strategy: making the server product both easier to bundle and harder to unbundle.

If you believe, as I do, that the motivation is implicit in the outcome -i.e. that Server 2008 is more or less what Microsoft wanted it to be - you can see the evidence for this in the operational changes they've announced.

On 2008 Server, for example, the kernel is still a typical x86 object furball with multiple hand-offs in almost all processes, but it's also become much easier for Microsoft to treat chunks of it as optionally loadable. As a result, Microsoft will now be able to do things like:

  1. offering different APIs in the European product than the American one;

  2. offering Linux run time compatibility in the Chinese market while withholding it elsewhere;

  3. extending their version of direct IO to pass through a cryptology processor in some markets but not in others;

  4. providing bundled (and non unbundable) OS/Application combinations - for example, dedicated database servers assembled from unduplicated OS and SQL-Server code components could easily have a 15-20 percent performance advantage over the unbundled versions; and,

  5. replacing the entire server OS file management stack with a kernel set giving thousands of server blades shared access to the same global file system - something that's apparently aimed at competing with google by selling cloud computing for Windows desktop applications.

None of this advances technology: on the contrary a lot of the underlying design choices recapitulate changes made to Unix in the eighties and nineties - meaning, among other things, that a lot of the stuff MCSEs are going to be learning about next year, stuff like text format local configuration files, looks kind of cute and nostalgic to people like me.

And that, I think, reflects Microsoft's other strategic concern: that MacOS X and Linux bracket Microsoft's market and are both reaching out to the people in the middle - the managers and MCSE decision makers who now consider Apple too consumer oriented and Linux too technical.

As defensive moves go this is dangerous for Microsoft -and potentially life style changing for its supporters. The reason for that is simply that this is a classical slippery slope situation: every time Microsoft "legitimizes" some Unix idea from a few years ago, it creates more pressure to go further because it creates more opportunities for someone else to sell Linux or Solaris or BSD as the more complete, more up to date, and significantly less manpower intensive way of doing the same thing better.

Surprisingly it could also turn a long term Microsoft strength, the lack of interest in technical issues among bosses, into a major negative: because the superficiality that leaves most of them unable to differentiate an OS kernel from popcorn could also lead them to conclude that new stuff like PowerShell is so much like the 80s and 90s Unix stuff they saw in school that Microsoft's general claim to data center inevitability should be questioned - and, as I said yesterday, Hillary Clinton's continuing decline in the face of Barack Obama's self-immolation on every policy issue in sight illustrates when happens when people who really don't like you start to doubt other people's commitment to you.

So, bottom line, I think the 2008 server bundles will prove commercially successful for Microsoft in the short term, but represent an extremely high risk strategy forced on the company by regulatory activities around the world and the growing acceptance of both the MacOS X and Linux alternatives.


Paul Murphy wrote and published The Unix Guide to Defenestration. Murphy is a 25-year veteran of the I.T. consulting industry, specializing in Unix and Unix-related management issues.