• <Läs mer på Intel.com

Baidu, Intel: Power Optimization for High Server Density Racks

Den här PDF-filen är tyvärr endast tillgänglig i hämtningsformat

Baidu, Intel: Power Optimization for High Server Density Racks
Executive Summary
Intel partnered with Baidu.com and conducted a proof of concept (PoC) project using the Intel® Intelligent Power Node Manager (Node Manager) and Intel® Data Center Manager to intelligently optimize server performance and power consumption to optimize server density. The engineers implementing the PoC used Node Manager to identify optimal control points, which became the basis to define node level power optimization policies. Intel® Data Center Manager (Data Center Manager) Reference Graphical User Interface was used to manage servers as a group to carry out rack power policies while minimizing performance impact of all the managed nodes in the rack. The policies enable increasing rack density and workload yield even when the whole rack is subject to power limitations.

The PoC was conducted initially on site at Baidu in Q1’ 2008 at Intel-Baidu joint lab with a configuration close to production configurations running Baidu search workloads. The runs in the first round were carried out with Bensley generation platforms. An updated set of runs was carried out in Q1’2009 using pre-production units with the more advanced Nehalem platform.
The test servers used were provisioned with dual 5560 processors. These processors feature eleven ACPI p-states, allowing the frequency to vary from 2.8 GHz to 1.6 GHz instead of the three states available in the Bensley generation. The additional p-states allow finer grained power control and an expanded power dynamic range.
Here are some global results:
• For the Bensley generation it was possible to trim power consumption by 40 W running the Baidu search workload while staying within the acceptable performance boundaries determined by Baidu. For the Nehalem platform the range has been expanded to 70 watts, allowing Baidu increased operational flexibility.
• At rack level, up to 20 percent additional capacity increase could be achieved within the same rack-level power envelope when aggregated optimal power management policy is applied
• Comparing with today’s data center operation at Baidu, by using Intel Node Manager and Intel® Data Center Manager, there could be a rack density increase from 5 to 7/8 servers – a 40+ percent improvement

Read the full Baidu, Intel: Power Optimization for High Server Density Racks Case Study.