In Nemertes' New Data Center research, enterprise data center managers identified power consumption and heat dissipation as top challenges.

This trend is driven both by the simple issue of limitations - data centers hitting a wall with respect to the amount of power they can bring in or distribute, or the amount of heat they can control - and by externalities like the rapid rise in energy costs and the prominent occurrence of more frequent brownouts and blackouts in some regions. Data center power load is exceeding capacity in some places, blowing up budgets in others.

So, the enterprise now needs to consider resource consumption more carefully. The two main threads of this concern are in the design phase, with companies designing data centers to handle far more power and cooling; and in the operations phase, with emerging use of active management techniques to minimize power consumption.

Data center design measures are aimed mainly at making sure the power and cooling systems are able to cope with the kinds of densities of current draw and heat throw that are now common and what is foreseeable beyond them. It is, in essence, a one-time, design-time attempt to forestall power and heat problems in the future.

Data center conservation measures are aimed at making sure those limits are not hit, and at saving money along the way if possible. This is an ongoing activity, not something that happens at design time and then only at long intervals during the life of the data center as limits begin to pinch and the unforeseen changes the landscape.

Optimizing for low power consumption in the operating system of a server is a good start, and well established, but this local optimization is not enough. The need for higher-level awareness becomes easily apparent when you look at a mixed workload running across a mass of servers.

It is easy for operating-systems-based power management to keep a particular system running at the lowest current draw possible, but the fact that multiple servers’ workloads can be combined onto a single physical server is not something the OS can see or act on.

Tools like VMware Vmotion make it possible in a virtualized environment to shift the workloads on the fly to expand and contract the number of processors involved in serving a workload, by stacking virtual servers more deeply when CPU loads allow. This works well, but it is a reactive management method.