Dynamic Power Management Has Significant Values - a Baidu Case Study
Jackson He, Intel Corporation
We have just completed a proof of concept (POC) project with Baidu.com, the biggest search portal company in China (60+% market share in China), using the IntelĀ® Dynamic Power Node Manager Technology (Node Manager) to dynamically optimize server performance and power consumption to maximize the server density of a rack. We used Node Manager to identify optimal control points, which became the basis to set power optimization policies at the node level. A management console - IntelĀ® Datacenter Manager (Datacenter Manager) was used to manage servers at rack-level to coordinate power and performance optimization between servers to ensure maximum server density and perform yield for given power envelope for the rack. We have shown significant benefit from the POC and the customer like the results:
At a single node level, up to 40W savings / system without performance impact when a optimal power management policy is applied
At rack level, up to 20% 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 datacenter operation at Baidu, by using Intel Node Manager, there could be a rack density increase 20~40% improvement
Some background of the technologies tested in this POC:
IntelĀ® Dynamic Power Node Manager (Node Manager)
Node Manager is an out-of-band (OOB) power management policy engine that is embedded in Intel server chipset. It works with BIOS and OS power management (OSPM) to dynamically adjust platform power to achieve maximum performance/power at node (server) level. Node Manager has the following features:
Dynamic Power Monitoring: Measures actual power consumption of a server platform within acceptable error margin of +/- 10%. Node Manager gathers information from PSMI instrumented power supply, provides real-time power consumption data (point in time, or average over an interval), and reports through IPMI interface.
Platform Power Capping: Sets platform power to a targeted power budget while maintaining maximum performance for the given power level. Node Manager receives power policy from an external management console through IPMI interface and maintains power at targeted level by dynamically adjusting CPU p-states.
Power Threshold Alerting: Node Manager monitors platform power against targeted power budget. When the target power budget cannot be maintained, Node Manager sends out alerts to the management console