Dear Dr. Cloud,
There’s has always been a lot of talk about policy-based management of IT. When I was in my last job, we thought about doing it for virtualization. Before that, autonomic computing elevated policy-based management to trekkie-like heights of sentience. And all the while, precious little of it has actually been implemented in any datacenter – and we’re all still alive and well (though distinctly lacking in android peers.) And now it’s back.. even in the BMC announcements. What’s the point?
--Poo-Pooing Policy in Portugal
Dear Portugal (since Poo Poo is just not right),
It’s true. Systems management vendors for years have described a utopian world in which our IT environments are self-healing, self-diagnosing and self-soothing. Then, Sci Fi writers writer about HAL and Data and Talkie Toaster*. I’ve often wondered which was the cause of the other.
Nonetheless, let’s look at reality. Back in the good old days when servers were things you could kick, connections were brightly colored and glee was gotten through self-reporting broken system fans, the idea of autonomic computing was somewhat less than compelling. Sure, you could ostensibly flip on a new chip if the old one burned out, or engage a back-up if the fan was dying. But, almost all the use cases for self-healing policy-based behavior were fundamentally physically tied.
Now, a server lives in a cluster – somewhere in that cluster – or datacenter – floating virtually over the hardware. In fact, a good multi-tier cloud service can float across many bits of hardware. The good news is, you can’t stub your toe kicking it. Still better is the ability to make a bunch of different changes to the service without setting foot in a datacenter. You can move it, grow it, shrink it, turn it off, back it up, scan it, and on and on. And, you can look at its needs in the context of all the other cloud services with which it shares space, and prioritize between them. You can make better initial placement decisions – and better ongoing management choices.
But, that’s a lot of data to collect. You’re basically describing an ongoing optimization algorithm with many many data points – dozens or hundreds of services to consider. If you aren’t actually a sentient android, that’s a great deal of math for a single person to do on an ongoing basis. And, even if you have a fairly stable cadre of services and resources, how fast can you actually respond to a change in the environment? Humans tend to be .. slower than machines.
Hence the need for rules that automatically govern the actions taken by the systems. And a policy-engine is simply the big giant brain that makes sense of those rules, prioritizing some and dismissing others in order to execute an order and Make It So. Certainly, some situations will still call for human intervention, but most of the routine drudgery of ensuring things are healthy and happy can be automated.
Finally, in this, the era of cloud, the policy engine is delivering on its promise… and you can save your time for higher-order tasks.
* 2001: A Space Odyssey; Star Trek: The Next Generation; and for the esoterically oriented, Red Dwarf.
Dr. Cloud answers cloudy questions on Tuesdays (or when he's late, Wednesdays). To reach the good doctor, email firstname.lastname@example.org