I got into an argument awhile back with a client over thin-provisioning. His argument, while valid, represented a very conservative approach while I say if you’re not living on the edge you’re taking up too much room.
Thin provisioning is what happens when you provide as much of a resource as requested, and not reserve all that is promised. Say you’re planning to have some friends over. You are aware than an adult human could technically consume a 6-pack of beer at one party. When they walk in the door do you have them a 6-pack? Or just one, expecting to hand them another when they finish the first? Voila – thin provisioning.
Over-subscribing is taking the next step: promising everyone who shows up a 6-pack when you only buy enough for each person to have 3. Now the options when you run out are either go buy more or start pouring into cups and adding water to stretch that Keystone as far as possible (really, who’ll notice more water?).
The assumption is, on average you’ll meet demand and hope your infrastructure can scale rapidly. Not one to abandon a metaphor lightly, let’s say you live next to a 24/7 liquor store, now oversubscribing could be quickly rectified.
The benefits of these methods are lower up-front and on-going hardware costs to maintain a lower level of resources. Traditionally, there were not many ways for the IT industry to implement this even if an admin subscribed to the idea. Before x86 virtualization (thanks VMware) the average WIntel server averaged <10% of resources used. Virtualization allowed thin-provisioning of hardware resources while maintaining OS integrity and separation. You wind up spending less money and getting greater utilization of your IT $$.
The downside is administration. If you make it possible to use more than you have, you’d better make sure you always know before you run out. You want to make sure you know who grabs the last can from the fridge, preferably before they do. Monitoring, trend analysis, planning etc suddenly become crucial, and it’s crucial 24/7.
The larger the pool of resources, the more economical it becomes. The greater the hardware utilization, the lower the costs of hardware, electricity, HVAC, etc. Complex monitoring systems and a staff with in-depth knowledge becomes more affordable the more systems you can spread your costs across.
Thus the underlying premise of the “Cloud”
Now, who needs a refill?