Power Management at the Core


AMD and Intel have put in a lot of time and effort devising CPU power management schemes for their multicore devices aimed at trimming the energy draw of the server farm.


But new questions are being raised as to whether these techniques are the most effective means to lower energy bills, particularly as virtualization becomes more popular throughout the enterprise.


Both AMD and Intel use some form of on-chip processing to shut down idle cores, allowing energy to be diverted elsewhere. Intel's platform provides for individual control of processors -- ostensibly to preserve laptop battery life -- and it is tying it to a server power management system that allows admins to direct power within individual server racks.


AMD offers a number of power-saving options, such as the PowerNow system, which lowers consumption across the entire processor, and Independent Dynamic Core Technology, in which each core can vary its operating frequency based on application needs.


But according to some critics, enterprises shouldn't expect much from the on-chip power-saving tools, particularly those that seek to manage idle cores. As tech journalist Tom Henderson points out on Computerworld, increased use of virtualization will make an idle core a rarity in the data center. The entire point of virtualization, after all, is increased utilization, so as more and more partitions go up, processors tend to start running at full speed with little or no idle time. In addition, the mere act of monitoring and allocating system resources adds more work to the CPU, so there will be fewer and fewer opportunities to shut down idle cores as more systems go virtual.


To their credit, neither chip maker is relying on idle-core management alone to cut energy costs. And a number of third-party vendors are broadening the supply of data center energy management systems that work on the server and network levels. One of the latest is Power Assure, which provides the Holistic Power Management system that aims to shut down idle servers and then automatically restart them when loads increase.


At best, individual core management is useful for organizations that need to overprovision their server resources in anticipation of scaling up operations, allowing them to switch on individual cores as needed. But you should know going in that the efficiencies gained will diminish as virtualization causes cores to switch on, and stay on.