At a high level, it is easy to call the proliferation of virtual machines “sprawl” which connotes a negative use of resources. I think there is subtly here that implies not all this "sprawl" is actually bad.
In the past, to harness and use significant computing power in your office required lots of space, physical servers, power, cooling, time, server administrators, paper work etc. The bar was pretty high to get a brand new multi-thousand dollar server approved through the CFO. Once approved, you had to wait a significant period of time for server deployment on the data center floor. The cost, the process time, and the process steps were all long and complex. Only applications with the highest ROI and criticality theoretically made it through the gauntlet.
Virtualization has effectively lowered the cost, time and process bar to deploy a server. This is not all bad. In fact, it is the opposite. It means more and more computing power can be effectively deployed by an organization to its competitive advantage. Take the mainframe to open systems change. With the mainframe, only the crème de la crème of applications were deployed. The rest were done, well, manually. Open systems changed that and lowered the bar for application deployment and hence the ability to deploy compute power. We are seeing a similar change today with virtualization. The bar is lower and more compute power can be deployed to solve a company’s problems. That is a good thing. Hence I am not sure sprawl is the right term for these applications.
At some point, however, even in a virtualized world, the value to an organization to deploy an application is less than the cost. In other words, there is no ROI to an application. But how do you find these applications? The answer for the virtualization admin is that you can’t unless it exhibits Zombie-like characteristics.
This is where the business leadership needs to be involved to make the hard choices. Enabling business leadership with the data to understand the true cost of applications is the key step in this process. The way to accomplish this is to provide meaningful showback of cost data. Notice I used the word showback and not chargeback, although chargeback is certainly possible.
Showback implemented in a basic model of either allocated costs or utilized costs grouped by application name and owner provides the data IT leadership needs to question which applications should be deployed or not. VKernel provides chargeback and showback modules along with other resources on the Chargeback Resources page that can get you started in this process. Whether you use our tools, build your own, or use someone elses, does not matter. What is key is to get started on the showback process understanding your cost structure.
So VM sprawl is not as cut and dry as it seems. Last week, we saw how Zombie VMs are the epitome of sprawl. This week, we saw how the simple deployment of computing power is not necessarily a bad thing provided there is sufficient business visibility into the environment to make the right deployment decisions.