At DeployCon in NYC last week, a few of us were having a discussion over dinner about quantifying the value of IT Infrastructure. There’s the old joke that the best way to save money in IT is to simply turn off all of the servers. Aside from being impractical, it turns out that this won’t save you much money anyway (as James Hamilton points out in his terrific talk at Mix 2010). Perhaps we’ve been looking at the cost and utilization problems backwards. Let me explain.

Most enterprise customers implementing chargeback/showback in their private infrastructure do so based upon how much capacity you have used. Either this is a function of the actual utilization on a weekly/daily/hourly basis or it’s based upon how much capacity you have been allocated, regardless of how much you actually use. In either case, most business units will have to give IT some sense of what your utilization is likely to be for capacity planning and budgeting purposes. Enterprises are really bad at predicting this type of stuff and tend to over-estimate the resources that they will need. Even within highly virtualized environments, the average server utilization still hovers in the 30-35% range. The wasted capacity hurts the economics of private cloud and current chargeback systems create financial incentives to exacerbate this problem – the less you use, the less you pay. (Note: In public clouds situations, as long as the unused capacity is ‘sold’, it’s not really considered waste from the business perspective of the provider.)

What if You Charged Users for What They Didn’t Use?

If servers and infrastructure are underutilized sunk-costs for the enterprise, shouldn’t you reward users for finding business value for this capacity? This is especially true since the marginal cost of running a server at 30% utilization vs. 80% utilization is so incredibly small. As a business unit, I ask IT for a specific amount of capacity and the closer that I come to 100% utilization of that capacity (or whatever target is set in place), the less I pay per unit of that capacity. This scheme creates a financial incentive to maximize the investments already in place and reduce over-provisioning infrastructure. It also focuses on measuring value and efficiency rather than simply reducing direct costs of the infrastructure. I would envision utilization threshold “cliffs” which, once exceeded, drop per unit capacity costs to lower pricing tiers with linear reductions in per unit costs between cliffs. In a future post, we’ll explore a detailed example to compare various approaches and models.

Anybody implementing this today? Would love to get feedback.