Patent Application
20060143486
1. Technical Field
One or more embodiments of the present invention generally relate to power management. In particular, certain embodiments relate to profiling unconstrained power of a processor.
2. Description of Related Art
As the trend toward advanced central processing units (CPUs) with more transistors and higher frequencies continues to grow, computer designers and manufacturers are often faced with corresponding increases in power and energy consumption. Increased power consumption can lead to overheating, which may negatively affect performance, and can significantly reduce battery life.
Modern CPUs have internal power control mechanisms that are responsible for constraining the CPU power usage and for preventing thermal overrun. Internal power control mechanisms usually consist of one or more on-die temperature sensors that are used as “thermal overrun indicators” and a power reduction mechanism that causes a processor to alternately move between low and high voltage/frequency operating points. A technique where a power reduction mechanism causes a processor to switch to a low voltage/frequency operating point once in a while may be referred to as throttling. A typical activation method of the above control mechanism is to determine a threshold temperature and to activate the power reduction mechanism for short periods of time as many times as it takes to maintain the temperature below the threshold.
A significant constraining factor in the design of CPU-based systems is the CPU's average power usage, rather than maximal power usage, because overheating of a CPU is more likely to be caused by extended periods of high power usage rather than short bursts of high power usage followed by a low usage or idle operation. The reduction of the CPU's average power usage (e.g., optimizing the power usage) may be addressed at all system levels: from the application level through the operation system (OS) and down to the CPU design itself.
Power usage optimizations at software layers (e.g., OS, application, etc.) usually require both access to internal CPU activity indicators and an ability to accurately estimate power usage for an application in order to be able to make power-performance tradeoff decisions (either in real time or for subsequent runs of the same software). The power usage pattern of an application may also be referred to as the power profile of an application. For example, many computer applications cause the CPU to consume relatively high power at high performance for short periods of time, while requiring relatively low power operation the rest of the time (e.g., idle while waiting for user input). For such applications, it may be inefficient to constrain the CPU power based on the peak power usage by the application, as it may unnecessarily reduce performance of the application. On the other hand, some computer applications cause the CPU to consume relatively high power for extended periods of time. The latter type of computer application may require tightening the power constraining mechanism for the duration of the extended peak power usage and relaxing the constraints during the low power usage periods. Ability to accurately power profile an application may allow precise tailoring of the power constraining mechanism to the application's power needs.
The embodiments of the present invention are illustrated by way of example and not limited by the accompanying drawings, in which like references indicate similar elements and in which:
Effective power/performance optimization at the software level may require accurate CPU power profiling capabilities. In order to allow runtime optimizations of software applications, power profiling data should be available in real time. In order to accurately power profile an arbitrary application, it may be necessary to determine the unconstrained power usage (or simply unconstrained power) of that application. Unconstrained power is a measure of usage of power in a system, provided there are no power constraining mechanisms. Unconstrained power with respect to an application utilizing a CPU (application's unconstrained power) may also be referred to as CPU's unconstrained power. There are currently available techniques to perform measurements of CPU's unconstrained power. The existing techniques, however, may require additional hardware and often may be used only in laboratory conditions
The use of temperature sensors as “thermal overrun indicators” is an efficient way for avoiding power (thermal) overrun. Temperature sensors, however, do not necessarily provide a way to accurately monitor the CPU's unconstrained power. One reason is that a temperature reading only provides a momentary thermal status. Furthermore, when a CPU activates its power constraining mechanisms in order to prevent thermal overrun, it also changes the actual power dissipation and temperature. Therefore, the temperature reading is not always an accurate indicator of the CPU's unconstrained power.
A more reliable approach to estimate CPU's unconstrained power may include determining a rate with which the CPU's power constraining mechanism is activated for a particular thermal threshold and identifying unconstrained power that corresponds to the determined rate. This approach may be implemented, in one embodiment, utilizing existing power-constraining components of a CPU. For example, the amount of throttling that is required in order to maintain the temperature of a CPU below a particular threshold can be accurately translated into the real-time unconstrained CPU power. In one embodiment, a real time estimator of unconstrained power may monitor the unconstrained power of a CPU by obtaining the thermal threshold (e.g., by obtaining the temperature reading from the CPU's temperature sensor), measuring the amount of throttling within a given timeframe, and consulting a lookup table containing correlation information.
The idea behind the real time estimator of unconstrained power is based on the following observation: monitoring the activation of the CPU's power constraining mechanism over a certain timeframe, while knowing the constraining mechanism's temperature target, may be accurately translated into the unconstrained power of the CPU. By way of analogy, this idea may be thought of as a mechanism to monitor the energy consumed by a car by monitoring the speed of the car and the extent to which the car brakes were used. The more frequently the brakes are used in order to maintain a particular speed, the greater is the unconstrained energy of the car.
A method utilizing temperature reading and the rate of power constraining (e.g., throttling) within a given timeframe may provide an accurate real-time measure for the CPU's unconstrained power, while exploiting the existing on-die components. Thus, the unconstrained power may be estimated and an arbitrary application may be power profiled without significant modifications to existing hardware and software and, in one embodiment, with no need for sophisticated and expensive measurement equipment.
The method and system to estimate a CPU's unconstrained power may be utilized, for example, to power profile applications in real-time in order to optimize power-performance at runtime, to power profile applications offline for offline compiling-based power-performance optimizations, for laboratory studies that require power profiling, as well as for other purposes. It will be noted that, in on embodiment, the method and system described below may be utilized for any kind of a processor that possesses a throttle meter.
In operation, according to one embodiment of the present invention, the power estimator manager 132 determines the time frame (e.g., a predetermined period of time) for monitoring activation of the power constraining control mechanism 120 (e.g., throttling) and communicates the time frame to the throttle meter 134. The throttle meter 134 communicates with the throttling control logic to determine the rate at which the power constraining mechanism 120 is activated (e.g., by counting throttling events). The power estimator manager 132 reads the result produced by the throttle meter 134 (e.g., the rate of activation of the power constraining mechanism 120 or throttling rate) at the end of the time frame and translates it into unconstrained power values. The translation of the result produced by the throttle meter 134 into unconstrained power values may be accomplished using a pre-computed lookup table. Such lookup table may include correlation data between unconstrained power of a processor and a throttling rate for a particular thermal threshold. In one embodiment, the throttle meter 134 counts throttling events over the timeframe designated by the power estimator manager 132 and returns the number of throttling events to the power estimator manager 132. Thus, the unconstrained power of a processor is estimated in real time, which may allow for power-performance tradeoff decisions to be made based on the resulting power profile of the application utilizing the processor.
It will be noted that the translation of the result produced by the throttle meter 134 into unconstrained power values may be performed offline, and the resulting power profile of the application may be utilized for optimizing power usage of the applications during subsequent runs of the application.
Referring to
The method illustrated in
Referring to
The exemplary computer system 300 includes a processor 302 (e.g., a central processing unit (CPU) a graphics processing unit (GPU) or both), a main memory 304 and a static memory 306, which communicate with each other via a bus 308. The computer system 300 may further include a video display unit 310 (e.g., a liquid crystal display (LCD) or a cathode ray tube (CRT)). The computer system 300 also includes an alphanumeric input device 312 (e.g., a keyboard), a cursor control device 314 (e.g., a mouse), a disk drive unit 316, a signal generation device 318 (e.g., a speaker) and a network interface device 320.
The disk drive unit 316 includes a machine-readable medium 322 on which is stored one or more sets of instructions (e.g., software 324) embodying any one or more of the methodologies or functions described herein. The software 324 may also reside, completely or at least partially, within the main memory 304 and/or within the processor 302 during execution thereof by the computer system 300, the main memory 304 and the processor 302 also constituting machine-readable media.
The software 324 may further be transmitted or received over a network 326 via the network interface device 320.
While the machine-readable medium 322 is shown in an exemplary embodiment to be a single medium, the term “machine-readable medium” should be taken to include a single medium or multiple media (e.g., a centralized or distributed database, and/or associated caches and servers) that store the one or more sets of instructions. The term “machine-readable medium” shall also be taken to include any medium that is capable of storing, encoding or carrying a set of instructions for execution by the machine and that cause the machine to perform any one or more of the methodologies of the present invention. The term “machine-readable medium” shall accordingly be taken to include, but not be limited to, solid-state memories, optical and magnetic media, and carrier wave signals.
Thus, a method and system to profile unconstrained power of a processor have been described. Although the method and the system have been described with reference to specific exemplary embodiments, it will be evident that various modifications and changes may be made to these embodiments without departing from the broader spirit and scope of the disclosure. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense.
