Pay for What You Use -- Are you sure?
One of the big stated benefits of moving your environments into "on-demand" platforms has always been that you only pay for what you use. I have to say, though, that I think that this particular value is dubious at best.
I can't speak for my counterparts across Australia, but from a management point of view, it's pretty important for me that I know what something is going to cost me. Every year, I prepare a budget and my boss in Tokyo approves a certain level of spend. Once I have my budget approved, I really try to avoid any nasty cost-overage surprises.
Now, I still a lot of value of these "on-demand" platforms. If I need to double my capacity for a short period, I can. I have the ability to choose to increase my planned spend if it makes business sense. I just want to avoid the after-the-fact, this is how much something costed me.
If you think about it, there are a lot of parallels between current the pricing for these "on-demand" services and mobile-phone pricing a number of years ago. When mobiles were far less wide-spread, the phone companies offered companies a "pay-for-what-you-use" pricing plan. Companies were charged for each call received and made and for each byte uploaded or downloaded.
This led, quite naturally, to much higher phone bills than companies had budgeted. Companies received their phone bills and started trying to manage when people could use their phones. The CFO's weren't very happy with these plans.
The phone companies responded by offering fixed costs for a certain level of usage and charging more for overages. They also allowed the companies to control who could use more than their quota and warned customers when additional costs were going to be incurred.
This new approach worked a lot better for the phone companies' customers. They could deliver the required service and plan for their costs. They could also bursttheir usage when necessary.
(The one big area that has remained "pay for what you use" has been overseas roaming and this is the last area that creates a problem for our companies' bean counters. Even this, though, is slowly being addressed)
I have got to believe that the compute-on-demand market will go a similar way. As companies increasingly use these platforms, they will find that they prefer to pay a fixed and known cost so that they can plan accordingly but maintain the right to increase usage through an explicit decision if they think that doing so makes sense.
So, when evaluating an offer from an "on-demand" provider, ask yourself how important is "pay for what you use" relative to cost predictability and the right to burst. You may find that this long-held value of these platforms isn't that valuable after all.