Best Practices That Introduce Proactive Power Management
If you read parts one and two, you recall this series about energy in the data center started with money—wasted money, to be exact. Approximately $24.7 billion is spent each year on server management, power and cooling for unused servers. This final installment circles back to money.
Read about practical how-to advice for introducing best practices that drive up energy efficiencies and therefore reduce operating costs. The resulting savings can contribute to your bottom line and strengthen the long-term business case for transformative, holistic energy-management approaches.
Best Practice: Leveraging New Energy-Management Tools
Visibility is the best antidote for wasted energy and for overprovisioning power and cooling. Most energy-management tools, however, restrict monitoring abilities to returned-air temperature at the air-conditioning units, or power draw per rack in the data center. Some vendors promote model-based approaches for energy management in lieu of real-time visibility, but the estimations that drive these models cannot achieve the accuracy required to accurately predict energy trends and identify issues before they disrupt service.
A new class of energy-management solution has emerged in response to the need for accurate visibility and analysis of energy. Holistic in nature, the new tools focus on server inlet temperatures to introduce more-fine-grained monitoring and control. Data is aggregated to give data center managers views of the thermal zones and energy-usage patterns for a row or group of racks, or the entire room.
Figure 1. Shows a thermal map of the “holistic” and real-time monitoring abilities
Best Practice: Fully Exploit Thermal and Energy Data
By adopting a leading energy-management solution, data center managers can automate the collection and aggregation of data from data center power and cooling equipment, servers, storage devices and power-distribution units. The analysis capabilities support drill-down into the data and extraction of insights that let the data center manager do the following:
- Identify potential hot spots early to pinpoint causes and correct problems proactively and thereby extend the life of equipment and avoid service disruptions.
- Log server and storage device power consumption to optimize rack provisioning, determine realistic de-rating factors for vendor power specifications and enable more-accurate capacity planning.
- Introduce power capping per rack to avoid harmful power spikes.
- Intelligently allocate power during emergencies and extend operation during outages.
- Analyze trends to fine-tune power distribution, cooling systems and airflow in the data center to ultimately drive up energy efficiency.
Best Practice: Set Limits and Price Services Accordingly
With accurate monitoring and trending, IT managers can take the next step and introduce controls that enforce “green” policies for energy conservation and sustainable processes. This can include introducing alerts when power limits are exceeded to discover which users and workloads are consuming more than their fair share of power. Alerts can also be used to trigger automatic adjustments of workload.
Energy charge-backs can also encourage conservation. With power-based monitoring and logging, enterprises can attach energy costs to services and raise awareness about environmental impact and energy expenses related to individual data center workloads.
Best Practice: Balancing Power and Server Performance
Power capping for individual servers or groups of servers can directly affect server performance and quality of service. To make power capping a practical option for limiting power consumption, advanced energy-management solutions dynamically balance power and server performance. Processor operating frequencies are adjusted to tune performance and keep power below threshold levels. These approaches have been proven to reduce server power consumption by as much as 20 percent without affecting performance.
Companies like EMC currently rely on this control to meet their business targets. When EMC designed a cloud-optimized storage (COS) solution as the foundation for its Atmos service, it set targets for power and energy efficiency. The company now uses an energy-management platform that lets it control power for groups of Atmos servers. The power-management architecture calibrates voltage and frequency to keep servers below the power threshold. In a proof of concept for its power-constrained regime, EMC validated that power caps and related frequency adjustments are an effective combination that does not affect user services.
Baidu, the largest search company in China, similarly applied power capping and dynamic frequency adjustments, but in this case, the goal was an increase in server density. The company’s data center hosting provider was charging Baidu by the rack. The rack power limit (10 amps, or 2.2kW at 220V) left many racks with as much as 75% unused space, and an intelligent energy-management solution allowed Baidu to increase rack loading by 40% to 60%.
Best Practice: Agentless Energy Management
To avoid degradation of servers and services, an energy-management solution must be able to support agentless operation. Today’s best-in-class solutions offer full support for Web Services Description Language (WSDL) APIs and transparently coexist with other enterprise applications on physical or virtual hosts.
For compliance with data center regulations, data center managers should also look for an energy-management solution that enables secure communications with managed nodes.
Data center energy consumption represents one of the fastest (if not the fastest) growing cost of doing business today. Blindly over-provisioning and other past practices are exacerbating the problem, which can only be solved with a forward-looking, holistic approach to energy management.
The latest generation of tools and solutions continues to gain momentum, with measureable results validating the above best practices with a range of immediate and long-term cost-cutting benefits. Besides curbing runaway costs, intelligent energy management can also avoid power spikes and equipment damaging hot spots. Management will appreciate the contribution to the bottom line, and end users will appreciate the increase in service quality. And everyone will appreciate a more earth-friendly attitude in the form of responsible energy policies and practices.
 White Paper, PoC at EMC, “Using Intel Intelligent Power Node Manager to Minimize the Infrastructure Impact of the EMC Atmos Cloud Storage Appliances,” http://software.intel.com/sites/datacentermanager/atmospoweropt1_3c.pdf
 White Paper, PoC at Baidu, “Intelligent Power Optimization for Higher Server Density Racks,” http://software.intel.com/sites/datacentermanager/intel_node_manager_v2e.pdf
Leading article photo courtesy of Tom Raftery
About the Author
Jeffrey S. Klaus is the director of Data Center Manager (DCM) at Intel Corporation.