Original URL: https://www.theregister.com/2009/05/05/intel_dcm_power_throttle/

Intel animates Data Center Manager

Power throttling for Nehalem boxen

By Timothy Prickett Morgan

Posted in Channel, 5th May 2009 20:47 GMT

Look out, system admins. Intel wants your job.

Intel has started shipping its Data Center Manager power management tool, something the techies at the chip maker have cooked up to help system makers and server administrators make better use of the energy conservation features of its new "Nehalem EP" Xeon 5500 processors.

The Nehalem EPs, of course, were launched at the end of March, marking Intel's transformation to a supplier of Opteron processors evolution of the Xeon chip to high-bandwidth, multicore processors that ditch the bottle-necked front side bus of the former Xeon architectures.

The Nehalem EPs also have all kinds of electronic gadgets baked into the chips to help them conserve energy and run more efficiently. One of them is called the Intelligent Power Node Manager, and the Data Center Manager program - a software development kit, really - hooks into these features to help monitor and manage power consumption at the chip, server, rack, and data center level.

All the major server chip makers have been baking similar features into their chips because a watt of juice is a terrible thing to waste. (With chips, you waste it twice, in fact: once to spin the chip, and once to cool it down). To help its x64 server OEM customers - some of which are not chip makers or systems software makers and therefore do not have such tools already - Intel cooked up the DCM SDK to give them something to go to market with to provide an edge against RISC/Itanium Unix boxes and mainframes that have fine-grained power throttling already.

The DCM tool can monitor power consumption and air inlet temperature on Nehalem EP two-socket servers and presumably will work on Intel's future "Nehalem EX octo-core chips for larger machines, due in late 2009 or early 2010. The tool aggregates data over servers, racks, groups of racks, and data centers and stores a year's worth of trend data. System administrators can also set up DCM to give them custom alerts based on power or thermal events and automate event handling by assigning policies to power or thermal events.

For instance, if the server temperature breaks through a threshold, the clock speed is automatically stepped down. The DCM tool can aggregate policies across groups of servers to allow fairly sophisticated responses to thermal and power conditions in a data center. It includes provisions for application service level agreements that override the power and thermal conditions, if necessary. And because DCM can cope with power at the server level, it can allow admins to allocate server capacity based on actual needs now, not on some theoretical average power consumption calculated on the back of a napkin in the lunchroom.

The DCM 1.0 server can manage up to 1,000 server nodes, and it complies with IMPI 2.0 BMC authentication for security. The toolkit has application programming interfaces that adhere to the Web Services Description Language (WSDK), which means it can be integrated into server makers' own tools, third-party management tools, or homegrown tools. Intel is allowing DCM to run on existing management servers if server makers want to do that, or it can run on its own dedicated server.

Intel is offering a 90-day free trial version if you want to kick DCM around, which you can get here. What Intel is not divulging is what it is charging for DCM, but server makers are encouraged to contact Intel to license the product. ®