Most in the data center industry know Jonathan Koomey from a 2007 Stanford University and Lawrence Berkeley National Laboratory study, which estimated worldwide server power consumption. The study helped get the federal Environmental Protection Agency (EPA)and the U.S. Congress involved and interested in data center energy consumption, which eventually led to the EPA's Energy Star programs for IT equipment and data centers.
Next month, Koomey will deliver a keynote address at the Uptime Symposium 2010 on how cloud computing can reduce total data center energy consumption. We talked with Koomey about his upcoming presentation.
Explain the power-saving advantages of hosting applications in the cloud as opposed to running them in your own data center.
Jonathan Koomey: Energy is a big driver of data center costs, and there are some advantages that cloud computing [has] over small data centers. One is diversity. If you have an in-house data center, you have a load profile of who is using the data center at certain times. If you have a homogenous group of users (running the same kind of applications), they're using it all at the same time. Google has so many users of different kinds that it can spread the use, and utilization is higher.
The second important advantage is economies of scale. If you want to improve efficiency, there is a fixed cost to doing that. An example is something like implementing the metrics in your facility to track what the energy use is. If you have a big facility, you can spread these fixed costs over more computers.
The third advantage is what I call flexibility. Most cloud computing centers are managing virtual servers, and the software is such that it allows you to manage them and route around physical servers that die. So you're redefining what we mean by reliability. Instead of obtaining a higher tier of reliability from redundant power equipment, you can create a system that's as reliable from nonredundant components.
The last one is enabling structural change. If you're a user and facing institutional barriers to reducing the cost of computing in your data center, it can be easier to ship [the application] to the cloud rather than fix the institutional problems in the IT shop.
Which applications are best suited for cloud computing?
J.K.: Well, Lawrence Berkeley Labs just announced that they're switching to Gmail. This is a high-tech place with big supercomputers, but [the lab] found advantages to not running email in house. There is no reason why any company shouldn't switch email to a cloud provider. The pressure will be to shift a lot more computing to the cloud.
So when will IT become like another utility bill to pay?
J.K.: I think it's all tied up into what applications are going to ship there. For some applications, it will never be like a utility bill. They will have to build their own facilities, and that is that. But they still may learn from the cloud computing centers. Instead of having one server for every application, they may figure out a way to design the application to run in an internal cloud.
Mark Fontecchio can be reached at firstname.lastname@example.org.