TW201344391A - CPU working frequency control method for mobile devices - Google Patents

Abstract

CPU working frequency control method for mobile devices comprises the steps of: When change in resource utilization model is detected, CPU working frequency in the new resource utilization model is predicted, according to historical records for resource utilization models and CPU work frequencies used in respective models; Control the CPU to work in the predicted working frequency; And modify the record of the CPU working frequency for the new model according to the actual utilization of the CPU in the new model. The modified CPU working frequency may be used as reliable reference when predicting needed CPU working frequency in the following steps.

Description

Mobile device CPU operating frequency control method

The present invention relates to a mobile device frequency control method for a mobile device, and more particularly to a control method for adjusting a CPU operating frequency according to a state in which a mobile device uses resources at the time.

Mobile devices have also become an indispensable tool for everyone every day. In order to meet the different needs of users, mobile devices have been designed to provide multiple functions, and their hardware components are more powerful and easy to use. When people pursue better human-machine interfaces and complex application services, next-generation mobile devices will continue to significantly improve their computing, display and communication capabilities. However, these trends have made mobile devices require a lot of energy to provide mobile applications, so their use time is facing severe challenges.

Among the applications provided by mobile devices, more and more mobile users are addicted to multimedia streaming applications and the ability to distribute audio and video via social networking communities. A 2010 report predicts that mobile data traffic will grow at multiples each year over the next few years, and in 2014 video streams will account for almost 66% of data traffic. The behavior of such users will significantly increase the energy consumption of the mobile device, especially if the user strongly desires a larger, higher resolution screen. Recent studies of user behavior of mobile devices have indicated that most of the power usage of mobile devices is consumed by backlights that provide a screen source for mobile devices. In addition to the backlight, the power consumption of the central processing unit (CPU) is also a problem. When receiving and playing an online or offline video stream, the central processor must perform intensive operations. In this case, the intensive operation of the central processing unit will shorten the battery life, thus shortening the use time of the mobile device.

Regarding the power consumption of the central processing unit and the communication components, a solution has also been proposed. Some recommendations dynamically adjust the operating frequency based on, for example, the workload of the central processing unit. This solution can be called Dynamic Voltage Frequency Scaling (DVFS) to dynamically adjust the CPU supply bottom voltage and operating frequency to save CPU power consumption. However, most DVFS methods do achieve power savings when the mobile device receives and/or plays a known video stream, but these solutions are in a mode of user interaction with the mobile device. It is difficult to help save energy.

In recent years, there have also been proposed ways to save CPU power consumption in the use of mobile devices intensively interacting with users. For example, the method proposed by Yan et al. is based on the degree of delay that the user can feel under the interactive application, and adjusts the voltage supplied to the CPU to save power. See Yan et al., "User-Perceived Latency Driven Voltage Scaling for Interactive Applications", published in Proc. of IEEE/ACM DAC, 2005, pp. 624-627. Mochocki et al. published a technique for using 3D interactive game applications that uses a signature-based prediction technique to predict the CPU workload in a three-dimensional animation to adjust the CPU operating frequency. See "Signature-Based Workload Estimation for Mobile 3D Graphics" by MOCHOCKI, B. C. et al., Proc. of IEEE/ACM DAC, 2006, pp. 592-597. Gu et al. proposed to establish feedback theory based on the last few prediction errors. See Gu and Chakraborty, 2008: "Control Theory-Based DVS for Interactive 3D Games", published in Proc. Of IEEE/ACM DAC, pp. 740-745. Basically, the above techniques are based on the animation frame of the animation game, and the proposed power-saving adjustment technology can only be applied to specific application fields. As for the dynamic adjustment technology that can be used in the field of unlimited applications, few papers have been proposed. More relevant technologies include the AutoDVS technology proposed by Gurun and Krintz in 2005, which differentiates the use of mobile devices into general application types and course grain applications, and uses predictive techniques to predict applications. Type to determine how to adjust the operating frequency of the CPU. See Gurun, S. and Krintz, C., 2005, "AutoDVS: An Automatic, General-Purpose, Dynamic Clock Scheduling System for Hand-Held Devices", published in Proc. of IEEE/ACM EMSOFT, pp. 218-226. In addition, the Governor Technology, which was published by Pallipadi and Starikovskiy in 2006, adjusts the operating frequency of the CPU according to several rules related to CPU applications during each sampling period. This technology is adopted by Linux 2.6 and is supported by the Android system version 1.5 and later. See Pallipadi, V. and Starikovskiy, A., 2006: "The on-demand governor: Past, present and future", published in Proc. of Linux Symposium, Vol. 2, pp. 223-238.

In the currently available literature, if the CPU operating frequency adjustment is based on predictions, most of the technology is based on the past application state of the CPU as a basis for prediction.

It is currently necessary to provide an innovative mobile device CPU operating frequency control method to provide mobile devices with efficient and efficient power saving.

It is also necessary to provide a mobile device frequency control method for the mobile device, which can determine the CPU operating frequency of the mobile device through simple calculations to save power.

It is also necessary to provide a mobile device frequency control method for the mobile device, which can dynamically adjust the CPU operating frequency of the mobile device according to the actual demand of the mobile device application, so as to effectively save power consumption.

It is an object of the present invention to provide an innovative mobile device CPU operating frequency control method to provide an efficient and efficient power saving for mobile devices.

Another object of the present invention is to provide a mobile device frequency control method for a mobile device, which can determine the CPU operating frequency of the mobile device through simple calculations to save power.

Another object of the present invention is to provide a mobile device frequency control method for a mobile device, which can dynamically adjust the CPU operating frequency of the mobile device according to actual needs of the mobile device application, thereby effectively saving power consumption.

Another object of the present invention is to provide a mobile device operating frequency control device for a mobile device, which can be applied to a mobile device, and controls the CPU operating frequency of the mobile device in the above manner to achieve an effective power saving.

The invention provides a novel mobile device frequency control method for a mobile device, which can dynamically adjust the working frequency of the CPU according to the state of use of the mobile device at the time, so as to save the CPU energy consumption.

According to an embodiment of the present invention, the mobile device CPU operating frequency control method includes: detecting a state of using a resource by the system; and detecting a change in a resource usage state of the system, predicting a new state according to a change in resource usage state before the system CPU operating frequency; using the predicted CPU operating frequency as the operating frequency of the CPU in the new state; and causing the CPU to operate at the operating frequency.

In some embodiments of the invention, the resource usage status includes a status of the system using and not using resources at a particular time. In still other embodiments of the invention, the resource usage state change includes the system for a plurality of consecutive time periods, individual resource usage states. The time period may be a certain length or a different length of time. In a preferred embodiment of the invention, the time period used is that the same usage state lasts for a long time before the system resource usage state changes. The method of the present invention may further comprise the step of periodically detecting the state of the resource used by the system.

In an embodiment of the invention, the resource usage status includes a perimeter provisioning type that the mobile device uses and does not use at a particular point in time. In other examples of the invention, the resource usage status and the operating frequency of the CPU in the resource usage state. The method of the present invention may also include the step of recording a history of changes in the state of use of the system resources and the operating frequency data of the CPU in each resource usage state. The method of the present invention may further comprise the steps of calculating the usage rate of the CPU in a period of time, and calculating an updated CPU operating frequency according to the calculation result, replacing the CPU operating frequency data. Wherein, the usage rate of the CPU can be normalized by the highest operating frequency of the CPU.

The method of the present invention may also include the step of using the highest operating frequency of the CPU as the operating frequency of the CPU in the state of use when the new state of use of the resource is different from all of the recorded states of use. In such an example, the step of replacing the CPU's highest operating frequency data with the updated CPU operating frequency calculated based on the usage rate of the CPU during the state may also be included after the end of the state.

In the embodiment of the present invention, the CPU operating frequency of the new state is predicted by setting the CPU operating frequency of the new state to the average CPU operating frequency recorded in the resource usage states before the system. In other embodiments, the CPU operating frequency of the new state is predicted by setting the CPU operating frequency of the new state to the maximum of the CPU operating frequencies recorded in the resource usage states of the system.

Several preferred embodiments of the invention are described below by way of examples. The description will be referred to in the description. It should be understood that the description of the preferred embodiments of the present invention and the drawings are intended to illustrate the various aspects of the present invention, and that those of ordinary skill in the art can practice the invention. The drawings and detailed description are not intended to limit the scope of the invention.

While not necessarily bound by any theory, the inventors have discovered that mobile devices have their resource usage patterns for different applications and applications during different applications. The term "resource" refers to peripheral hardware devices equipped with mobile devices, such as displays, network devices, random access memory, flash memory, and amplifiers. CPU usage can also be considered as a resource application for mobile devices. The resource usage mode refers to the resource usage state of the mobile device during a specific application, that is, the state of using or not using a specific peripheral device or CPU, and can be classified into several modes.

The inventors have also found that the CPU of the mobile device has a corresponding usage rate under different resource usage states/modes. For example, when the data transfer rate is high, the CPU usage is also increased, and the frequency at which the CPU accesses the memory device is also increased. In addition, when the mobile device plays the video data stored in the memory card, the CPU only periodically accesses the memory card and the random access memory. In the former case, the CPU must operate at a higher frequency, while in the latter case, the CPU can operate at a lower frequency. Therefore, dynamically adjusting the operating frequency of the CPU according to the resource usage state of the mobile device can save power consumption of the CPU during periods when no high operating frequency is required. The present invention has thus been proposed.

Fig. 1 is a system diagram showing the CPU operating frequency control device of the mobile device of the present invention. The mobile device CPU operating frequency control device dynamically adjusts the operating frequency of the CPU according to the resource usage mode of the mobile device under different applications to reduce power consumption. As shown, the mobile device CPU operating frequency control device 10 of the present invention includes a resource monitor 11 for periodically sampling the usage status of most resources; and a plurality of resource state machines 12 for generating an application for a specific application. Pattern development chart. A resource state machine is enabled when a mobile device (not shown) begins to enable a new application (or is called in the foreground). The mobile device CPU operating frequency control device identifies the resource usage state at the time according to the updated state information of the newly enabled resource state machine, and the relationship between the usage rate of the CPU and the resource usage mode, and predicts and adjusts the operating frequency of the CPU. .

Further, in order to be able to periodically monitor the usage status of most resources, the resource monitor 11 is connected to a timer 13 to activate the resource monitor 11 at a predetermined time.

The resource monitor 11 generates a usage status code for most resources, including the usage status of the flash memory R1, the network R2, the RAM R3, the screen R4, and the like. In the representation of the code, if the resources are in the use state, the code is coded as 1 and vice versa. Therefore, the sampled resource usage state combination can be represented by a string of one-dimensional code called a bitmap or a bit string. A specific bitmap (also known as a state collection) represents a specific resource usage state pattern. For example, if the resources used by the mobile device include flash memory R1, network R2, RAM R3, screen R4, etc., as shown in FIG. 1, only the flash memory R1 and RAM R3 are used in the mobile device. The code can be programmed to 1010. When using network resource R2, RAM R3 and screen R4, the code can be programmed as 0111. The rest of the way. Of course, it is also feasible to use other coding methods.

Each state mode is coupled to a correlation table 14 for recording/representing the information needed to predict the CPU usage of the next state. When the resource usage state combination is changed, the system first predicts the resource usage mode of the new state, and records the current CPU usage rate and the next state code, and stores them in the original state correlation table 14. It then enters the new state. In this new state, the system predicts the operating frequency of the CPU and establishes a correlation table associated with the new state. In the new state, the CPU operates at the predicted operating frequency, but the operating frequency is later corrected by the system based on the information in the correlation table for the new state. The corrected data replaces the predicted working frequency and is recorded in the correlation table of the new state as the basis for predicting the CPU operating frequency.

Since the state machine 12 is sampling the dynamic change of the application mode in an on-line manner, predicting the next state and correcting the operating frequency of the CPU, the state machine 12 will be fully integrated with the particular application being executed at that time after a period of execution. Therefore, the operating frequency of the CPU can be accurately adjusted to effectively save power consumption.

The function of the resource usage state machine 12 is to track the resource usage pattern change of the mobile device to determine the relationship between the CPU usage rate and the resource usage status of the particular application. Thus, the definition of each state mode can include: a bitmap representing the usage state mode; and an indicator indicating the corresponding dependency table. For example, the resources shown in FIG. 1 include four types of flash memory, network, RAM, and screen. In state 3, both network and RAM resources are used. Therefore, the code is 0110. The correlation table 14 and records that its next state may be S1, S2, and S4.

Fig. 2 is a flow chart showing a method of controlling the CPU operating frequency of the mobile device of the present invention. As shown, in step 201 the mobile device begins executing an application. At this time, the resource state machine 12 records the current resource usage status in the correlation table 14 in step 202. At this time, the resource state machine 12 records only one state mode, that is, resources used by the application executed during the start period, such as State 1, code 1010. In step 203, the resource monitor 11 detects a change in usage status, such as adding and/or reducing one or more hardware devices. At this point in step 204, the resource state machine 12 generates a new state mode, such as State 2, code 1011. The resource state machine 12 determines in step 205 whether the state combination (State 2) sampled at that time is present in the record? If the result of the determination is negative, then the new state mode (State 2) is recorded in step 206 and recorded as the next possible state mode of the original state (State 1). If the result of the determination is affirmative, then in step 207, only the next possible state mode in which the new state mode (State 2) is the original state (State 1) is recorded. Thereafter, the control device 10 predicts the CPU operating frequency required for the new state (State 2) based on the information recorded in the correlation table of the original state (State 1) in step 208, and records the correlation table 14 recorded in the new state. in.

Before the mobile device enters the new state (State 2) from the original state (State 1), the resource usage state machine 12 calculates the usage rate of the CPU in the original state (State 1) in step 209, and calculates according to step 210. The resulting CPU usage is corrected for the CPU operating frequency recorded in the correlation table of the original state (State 1). In step 211, the resource state machine 12 replaces the originally recorded CPU operating frequency with the corrected CPU operating frequency, and records the code of the next state. Finally, in step 212, the control device 10 issues a control command to cause the CPU to operate at the predicted operating frequency until the resource monitor 11 detects a resource usage state change (step 213) or the application execution is complete (step 214) So far. After the resource monitor 11 detects the resource usage status change again, the process returns to 204. After the execution of the program is detected, the job ends.

In the application of the present invention, if the number of resources of the sampled object is large, the number of state patterns recorded by the resource state machine 12 may increase considerably after a period of time. To address this problem, in an embodiment of the invention, the resource state machine 12 can index the Hertz sequence. The purpose of using the Hertz sequence is to save space for the memory to record the next state of each state, and to make the state look more efficient. The method includes encoding the bit combination of each state as a sequence or corresponding to a one-hertz value. The combination of states with the same he-order value is added to the same Hertz sequence. When the state combination of resource usage changes, it is easy to find the corresponding new state combination according to the his-order value of the bitmap.

Figure 3 is a block diagram showing the structure of the correlation table in an embodiment of the present invention. As shown, in some embodiments of the present invention, the correlation table 14 may include three fields, respectively tracking the next state of the state (N column), CPU usage in this state (U column) And the likelihood value of each state in a single or most likely next state. The dependency table continuously tracks the relationship between CPU usage and resource usage in each state.

For example, in the embodiment shown in FIG. 3, the association table 14 is an association table of the state S3, and it is described that the state of the past ten times of the state S3 is changed according to the past state change record (Path Window). It changes to states S4, S4, S1, S1, S4, S1, S1, S1, S2, and S2. That is, it is only possible to transition from state S3 to states S1, S2, and S4. Further, from this history, it is understood that the probability of transitioning from state S3 to state S1 is 50%, the probability of transitioning to state S2 is 20%, and the probability of transitioning to state S4 is 30%. And if the state S3 is changed to the state S1, the CPU usage rates of the state S3 before the transition are 0.2, 0.2, 0.3, 0.3, and 0.2, respectively; and the average CPU usage is 0.24. In the embodiment of the present invention, the number of records in the next state of each state is restricted to a specific number, so the correlation table can be maintained on a small scale, but the next state that can be predicted by the state can be effectively predicted. .

According to the results of the actual use of the mobile device resources, it can be found that the P column and the U column can display two characteristics in the correlation table of one state: one is the approximation of the state transition (P column), and the other is The regularity of resource use (U column). The so-called state transition approximation means that in a specific application, when a time transitions from one state to the next state, it usually switches to a certain state. The so-called regularity means that the amount of resources used by mobile devices will be close during the time, so the CPU usage will remain at the same level during a period.

Based on the above observations, the resource usage status of the mobile device can be predicted from the past usage state changes. And according to the predicted result, the CPU operating frequency will roughly match the operating frequency required by the CPU. In one embodiment of the present invention, the prediction of the new resource usage state and its CPU operating frequency is determined based on the CPU usage rate in a predetermined number of resource usage states in the past. For example, in the example of FIG. 3, in the past several resource usage states, the number of "utilization window" of the CPU usage rate is set to five, and the average value of the five CPU usage rates is calculated as 0.24.

In calculating the usage rate of the CPU in a specific resource usage state in a specific application, the present embodiment counts the ratio of the busy time to the elapsed time accumulated by the CPU during the state, and The CPU's highest operating frequency is normalized and the results obtained. This calculation saves memory space used to record CPU usage. However, other calculation methods are also applicable to the present invention. In this example, for example, the CPU has a maximum operating frequency of 800 MHz, but during the busy period, only 40% is used at a frequency of 200 MHz, and the CPU usage should be 0.1, that is, 40% X (200/800). The calculated result according to the above method can be used to correct the actual operating frequency of the CPU in this state as a basis for predicting the operating frequency of the CPU.

The present invention provides several prediction methods when predicting CPU operating frequencies. For example, when the new state does not exist in the record of the resource state machine 12, the CPU operating frequency used by the new state can be set to the highest operating frequency of the CPU. This is not a technical limitation, but it ensures that the application at the time can be executed with the highest CPU support. The CPU operating frequency required for this state can be corrected in the future according to the usage rate of the CPU in this state to achieve or approach the lowest operating frequency that meets the actual demand.

In an example of the invention, the frequency stability of the CPU is a primary consideration. At this time, the CPU operating frequency of the new state can be set to the highest operating frequency of the CPU, and the value of the result is calculated by the following formula:

Where Ns is the number of records in the correlation table, Pi is the P value of the i-th record in the correlation table, and Ui is its U value.

On the other hand, if the user's experience is the main consideration, the highest of the CPU usage records recorded in the correlation table of the state is used as the predicted CPU usage. In other words, the predicted CPU usage can be calculated using the information in the correlation table based on the desired or desired conditions of the system.

Fig. 4 is a table showing comparison of the power saving results of the present invention with the prior art. The test results of the present invention are represented by "RD-DVFS", and the comparative example is the result of setting the three power saving modes of Performance, Conservative, and Ondemand in the above Governor technology. The application of the experiment includes: receiving and playing the movie "Monga" (MW) by WiFi, reading and playing the movie "艋舺" (MS) from the SD memory card, receiving and playing the movie "Avatar" by WiFi. (AW), read and play the movie "Avatar" (AS) from the SD memory card. The experimental results show that the method and device of the present invention are superior in effect.

10. . . Mobile device CPU operating frequency control device

11. . . Resource monitor

12. . . Most resource state machines

13. . . Timer

14. . . Correlation table

Fig. 1 is a system diagram showing the CPU operating frequency control device of the mobile device of the present invention.

Fig. 2 is a flow chart showing a method of controlling the CPU operating frequency of the mobile device of the present invention.

Figure 3 is a block diagram showing the structure of the correlation table in an embodiment of the present invention.

Fig. 4 is a table showing comparison of the power saving results of the present invention with the prior art.

Claims (19)

A mobile device operating frequency control method for a mobile device includes the following steps: detecting a state of using a resource of the system; and detecting a change in resource usage state of the system, predicting a CPU operating frequency of the new state according to a change in resource usage state before the system; Taking the predicted CPU operating frequency as the operating frequency of the CPU in the new state; and causing the CPU to operate at the operating frequency. The mobile device operating frequency control method of the mobile device according to claim 1, wherein the resource usage state includes a state of the resource used and unused by the system at a specific time. The mobile device CPU operating frequency control method of claim 1, wherein the resource usage state change comprises an individual resource usage state of the system in a plurality of consecutive time periods. The mobile device operating frequency control method of the mobile device of claim 3, wherein the time period can be a certain length or a different length of time. The mobile device operating frequency control method of the mobile device of claim 3, wherein the same usage state lasts for a long time before the system resource usage state is changed. The mobile device operating frequency control method of the mobile device of claim 1, wherein the detecting of the resource status of the system includes timing detection. The mobile device CPU operating frequency control method of claim 1, wherein the resource usage status comprises a peripheral equipment type used and not used by the mobile device at a specific time point. The mobile device operating frequency control method of the mobile device according to claim 7, wherein the resource usage state includes the operating frequency of the CPU in the resource usage state. For example, the mobile device CPU operating frequency control method of claim 1, 7, or 8 includes a step of recording the history data of the system resource usage status and the operating frequency data of the CPU in each resource usage state. For example, the CPU operating frequency control method of the mobile device of claim 9 further includes calculating the usage rate of the CPU in a time period, and calculating an updated CPU operating frequency according to the calculation result, instead of changing the history of the resource usage state. In the data, the step of recording data relative to the CPU operating frequency of the time period. The mobile device operating frequency control method of the mobile device of claim 10, wherein the usage rate of the CPU is a ratio of a usage time of the CPU to the time period. The mobile device CPU operating frequency control method of the mobile device of claim 10 or 11, further comprising the step of normalizing the update usage rate of the CPU to the highest operating frequency of the CPU. The mobile device operating frequency control method of the mobile device of claim 9, wherein the prediction of the CPU operating frequency of the new state includes the CPU when the new usage state of the resource is different from all recorded usage states. The highest operating frequency is the step of the CPU operating frequency of the usage state. The mobile device operating frequency control method of the mobile device of claim 9, wherein the prediction of the CPU operating frequency of the new state comprises setting the CPU operating frequency of the new state to be the CPU recorded in the resource usage state before the system. Average operating frequency. The mobile device operating frequency control method of the mobile device of claim 9, wherein the prediction of the CPU operating frequency of the new state comprises setting the CPU operating frequency of the new state to be the CPU recorded in the resource usage state before the system. The maximum value in the operating frequency. For example, the method for controlling the operating frequency of the mobile device CPU of claim 13, 14 or 15 further includes calculating the usage rate of the CPU in a period of time, and calculating an updated CPU operating frequency according to the calculation result, instead of using the resource. The change history record of the state, the step of recording data relative to the CPU operating frequency of the time period. The mobile device operating frequency control method of the mobile device of claim 16, wherein the usage rate of the CPU is a ratio of a usage time of the CPU to the time period. The mobile device CPU frequency control method of the mobile device of claim 16 further includes the step of normalizing the update usage rate of the CPU to the highest operating frequency of the CPU. The mobile device CPU operating frequency control method of claim 17 of the patent application, further comprising the step of normalizing the update usage rate of the CPU to the highest operating frequency of the CPU.