At LISA 2007, John Strassner did an excellent job of presenting how he and his team were building an autonomic system to manage a really complex infrastructure.
What he observed was that while the business bought best of breed widgets (eg: applications, servers, operating systems, switches and cables) what the business wanted was a tightly integrated system that delivered a specific business service. The challenge is that the best of breed widgets do not readily snap together to build that specific business service.
The challenge he faced is
How to build a service such that the cost in administration was not prohibitive.
John then described his autonomic management system. The fundamental notion, was that if you can make more of the sub-components intelligent and able to, on their own, deal with many of the problems that currently require human intervention, you could in principle simplify the overall administration.
In effect, he seemed to be arguing that if we have more reliable components, where reliability is not defined in terms of up-time but defined in terms of how well the device is delivering to the business service, then we can actually build these tightly integrated systems out of these best-of-breed widgets.
His approach was to try and build an infrastructure that wrapped the widgets in a large amount of software that provides the autonomous management capability. He very specifically admitted that asking the vendors to do what he needed to do was unrealistic.
The meta-point he seemed to be implying was that the vendors could never build the kind of autonomous systems that he required because their goal was to build the best widget, not build the kind of business system he was trying to build.
So what does this have to do with NetApp? Although, I wouldn't go so far as to argue that our management software is "autonomic", a lot of what John wants is what we are trying to build. Namely, a system that allows an administrator to encode what the business policy should be such that software is responsible for detecting when the business policy is being violated and if possible automatically correct things.
The other part that is relevant to NetApp is John's comment on not relying on the vendor to do everything.
I really don't think that any infrastructure vendor can solve the entire problem. As a data management provider, our job is to make it easier for folks building the complex distributed systems to build them using our components. And to the extent that we provide the hooks and tools that are necessary to build those systems we, as data management providers, will be successful.
