Today I happened to wander across this article on simplifying for a happier data center. It’s a nice sentiment but I think it misses the correct definition of simplify.
Once you get past the new age happy babble, the gist of the article is to simplify by refusing “features”. While it is true sometimes that adding features badly leads to headaches from buggy results, rejecting new features that may make life easier in the name of simplicity is a non-starter.
So, what is the secret to simplifying the data center? Simple, identify which items are most likely to be a headache and eliminate them or use alternatives. A highly complex system that never ever gives a problem is not the problem. What do I classify under this type of heading? Any mechanical infrastructure systems (chillers, legacy HVAC, generators, etc), any electrical infrastructure that has a potential to catastrophically fail or where a catastrophic failure can turn a failure of a single redundant system into a single point of failure for the entire facility (consider the recent rash of transformer explosions and fires in data centers).
So, how do we simplify these things? Try augmenting legacy cooling with free air cooling (or moving to free air cooling completely). This will reduce the chances of a service impacting cooling failure dramatically, as well as reducing maintenance requirements to virtually nil. For generators, consider alternative green power sources or long run time static back up systems. My rule of thumb is, any time you introduce an internal combustion engine into the picture, reliability goes down and headaches go up (just ask IBM and Air New Zealand).
As I’ve written before in previous posts, keep oil filled transformers outside and away from the facility. If you are not operating your own local generation, it’s hard to avoid these, so know in advance they ARE going to be a headache sooner or later and act to minimize it.
Of course, from the internal data center standpoint, good network design that doesn’t add a lot of superfluous equipment and the most efficient use of servers possible never goes wrong either. Most of this is just good engineering for maximum reliability of the data center systems.
There you go, data center simplicity in a nutshell, and you didn’t even have to climb to a mountaintop guru to get it!
Vern, SwiftWater Telecom