Q&A: Delivering Real-Time Power Data
New servers deliver real-time power data, but it’s difficult to uncover.
Data center managers must cope with their environments reaching physical capacity as well as maxing out the amount of energy available to them. In an effort to cut costs and still power their equipment, more managers are turning to power-management solutions. We explore the difficulties of collecting data about energy consumption, the role of new Energy Star regulations, and the role DCIM will play with Jeff Klaus, director at Intel's Data Center Manager.
Enterprise Systems: With many priorities vying for attention in the data center, why is the collection of power/energy usage data important?
Jeff Klaus: Power consumption in the data center is a rapidly growing expense that is consuming a larger share of the IT budget. In most cases, enterprises are now outgrowing the power availability within their data centers long before they run out of space. Given the increasing costs of provisioning new power to a data center, or even constructing a new data center due to power limitations, data center managers need to get the most out of every square foot and kilowatt of power.
ENERGY STAR compliance requires new servers to deliver more power and thermal data than ever before, presumably making power usage data more readily available. What makes it difficult to uncover?
Although real-time, server-level power data has been available for servers for only the last couple of years, collecting the data has typically required a complex, manual effort. It also requires a deep understanding of a variety of low-level firmware protocols of the server. Microprocessors are consuming the lion’s share of power in the data center. Although some server manufacturers can provide this power data across their equipment for a fee, these solutions are proprietary to their devices only, preventing a comprehensive view across the data center.
What are the current methods for collecting power data?
Power is currently monitored at various sources throughout the data center, including the Uninterruptible Power Supply (UPS), the room power distribution unit (PDU), and within the rack. These data points are consolidated and either reported using a manual process that plugs in the data in Excel or similar spreadsheets, “home grown” reporting systems, or within an enterprise data center infrastructure management (DCIM) tool. The nascent DCIM market, according to industry analyst firms such as Gartner, IDC, Forrester Research, and The 451Group, addresses the need for better power and cooling management in the data center.
Why aren’t these enough?
Much of the available power data is based on estimated or manufacturers’ power ratings and not “real time” power consumption. This data can deviate from real consumption by as much as 40 percent. As a result, data centers are generally over provisioning, incurring high operational costs; have difficulty optimizing power, cooling, and physical capacity; and are left without a solution for managing power and cooling. In addition, many existing solutions provide data only down to the rack/circuit level and not at the server level. Without this level of granularity, data centers are missing many opportunities for optimizations that require server-level data.
What are the obstacles to unified data collection?
Currently there is no consistently adopted industry standard for this data, so it comes in different and incompatible formats from the various server manufacturers. In fact, some OEMs view power management as a differentiation for their hardware, and thus have compelling reasons to keep their interfaces proprietary.
Proprietary protocols being one obstacle to unified data collection, is there any work being done to standardize protocols?
Intel DCM is trying to bridge this obstacle by providing a unified way for solution providers to access servers’ power and thermal data, regardless of manufacturer.
What are the consequences of collecting partial or incomplete data?
One of the consequences is dramatic overprovisioning of power by as much as 40 percent. Another is the annual cost of overprovisioning. As one example, powering an idle 400-watt server can be $800; multiply that by hundreds of servers in a data center and the cost is no longer trivial.
Finalization of the next revision (2.0) of ENERGY STAR requirements for data center servers is scheduled for July 2012. What impact might this have on power usage data availability or collection?
Currently in development, the next version of ENERGY STAR for servers could do much to drive data center efficiency. There is speculation that there will be a requirement for a server to be able to report power consumption to be ENERGY STAR compliant. This will make more data available; however, there will still be issues of collecting that data across multiple pieces of equipment, as well as the issue of data incompatibility, to obtain a comprehensive view of power consumption across the entire data center.
The nascent market segment, Data Center Infrastructure Management (DCIM), has been grabbing headlines lately. What role will DCIM products play in real-time power data collection and analysis?
The growing costs of data center power consumption and cooling are driving the emerging DCIM space such that The 451Group predicts that DCIM will grow from $245 million in 2010 to $1.3 billion by 2015, or at an annual compound interest rate of 39 percent (according to a November 2011 report). The ability to monitor and adjust power and cooling to the server level and in real-time is a critical element to the comprehensive DCIM solutions on the market today. It will be important for any company evaluating a DCIM solution to ensure that the solution provides this micro-level data to ensure optimal operational efficiency.
Because many of today’s servers have the capability to report power consumption in real time, there is no longer a need to overprovision power provided they have the appropriate tools. Adjustments can be made immediately to ensure that there is never a risk of under-provisioning the power delivery to data center devices.
What initiatives does Intel have in supporting DCIM?
With the large amount of Intel silicon in today’s servers, we have a deep understanding of server power and thermodynamics. Additionally, we understand the highly complex and changing low-level protocols from various server OEMs and we are agnostic and work with the broadest range of OEMs (Dell, HP, IBM, and Cisco in addition to the regional OEMs). We also provide power data for UPS and PDU devices through our Web services interface, contributing to a broader understanding of the data center power footprint.
Several years ago, we saw a strong and unmet need for a software solution that provides real-time, server-level data on the power and thermal conditions of a wide range of data center servers and other equipment. To address this, we developed a number of power-capping patents that went into the creation of Intel Data Center Manager (DCM), which is provided as a software developer kit (SDK) to ISVs, OEMs, and end-user data center management.