componentization methodology for specifying energy requirements
Original Publication Date: 2009-Oct-08
Included in the Prior Art Database: 2009-Oct-08
Described is a methodology which enables the accurate prediction of daily power requirements for computer systems.
The methodology described in this article describes an approach for enabling the accurate prediction of daily power requirements for computer systems, also known as servers. Because of increasing energy costs, power consumption has become a key element in the decision process when purchasing a server. While it is relatively easy to estimate a minimum, a maximum, or possibly even a "typical" power requirement for a server, it is much more difficult to accurately predict a server's power requirements based on how the server and its components are being utilized given numerous significant factors that could affect the power requirements. Due to the large power requirements that servers can demand, ranging into the tens of thousands of watts, the ability to accurately predict power requirements is important for planning purposes (computer labs, IT and power budgets, etc.) and represents a competitive advantage to the server provider who can demonstrate the ability to accurately do so.
There are existing tools in the industry that predict power requirements for servers. The drawbacks of known solutions are that they have not demonstrated accuracy in their ability to reflect the varying power requirements for a given server based on how its components are utilized. A more effective solution would provide a method to enable the power requirements to be more accurately estimated based on a given customer's expected use of the server along with other significant factors. Such a methodology would incorporate information for each of the key components of the server to reflect how those components consume power differently based on numerous factors such as the component's utilization, server software settings, temperature, altitude, use patters, etc.
The core idea of the new methodology described herein is to represent the key components of a server in a manner that enables the ability to specify numerous attributes for each component which can be used to generate an accurate power estimate for that component based on its expected use. Some components, such as processors, can have a linear type behavior such that the power requirement increases somewhat proportionately with the utilization of the component. Even with a linear type behavior, the slope of the curve will vary by component and differences between the minimum and maximum power values will vary greatly by component. These are some of the more obvious attributes that would help describe a component.
Much invention is happening in the area of hardware design and software methodology for managing and optimizing power consumption on servers. These attributes must also be captured and characterized for each server component as is appropriate. Software settings can greatly vary the behavior of components by introducing different minimums and maximums, setting power consumption thresholds, imposing step-like functions instead of slope behaviors for...