An O/S for the data center?

I can’t take credit for this idea – it goes to Rob Gingell, Cassatt’s CTO, previously a VP and Sun Fellow. (Rob's blog)

But think of this: Software operating systems broker resources between applications, manage storage, I/O hardware and drivers, peripherals, etc., and even set priorities between them. They even (especially in Unix) isolate failed applications, if not, on occasion, re-start them. Also, mainly with Unix/Linux, OSs essentially isolate the underlying hardware from the applications – the user may never know (or care) if the platform is x86, RISC, etc.

So why not carry this analogy to the datacenter (where “the network is the computer” anyway)?

The problem set in the data center is similar: Operators spend most of their time balancing application resources between machines and across the network, installing drivers, adding patches and new software, re-starting failed servers, adding storage, doing backups, etc. In fact, in talking to customers, I’ve heard that almost 90% of time is spent “keeping the lights on” rather than doing value-add work. So, then why isn't there a technology that takes these manual tasks and and handles them as-would an O/S?

In full disclosure, this concept is essentially what Cassatt is trying to do: automate day-to-day operations so to have the data center operate as if it had an O/S. In this manner, administrators’ tasks are up-leveled to monitoring/maintaining overall capacity of compute power, memory & storage, deploying (but not managing) applications, and aligning the “O/S” operation with their business policies.

Then, maybe, operators will have more time to do fun stuff.