CN103955266A - Low power consumption design method based on Android mobile Sink load prediction - Google Patents

Abstract

A low power consumption design method based on Android mobile Sink load prediction belongs to a low power consumption design method for a mobile Sink node in a wireless sensor network, and particularly is a low power consumption design method for prediction through CPU load; the method comprises the steps that through a CPU driver of Android, the current load load-cur of a CPU is obtained, a historical load value in a history_load [3] array is updated, whether CPU historical load values history_load [1] equals history_load [2] of the first two time periods of current time period in history_load [3[ array is judged, respective current optimal weights Alpha best are calculated, then a CPU load value load_ next of next time period is obtained through calculation, finally a function _cpufreq_driver_target is called, and the working efficiency of the CPU is adjusted based on the predicated CPU load value load_ next. The low power consumption design method based on the Android mobile Sink load prediction can realize the dynamic adjustment of the weights to further improve predication accuracy so as to select more appropriate CPU calculated frequency to reduce power consumption.

Description

Based on Android, move the low power consumption design method of Sink load estimation

Technical field

The low power consumption design method that moves Sink load estimation based on Android belongs to the mobile Sink node low power consumption design method in wireless sensor network, the low power consumption design method of especially predicting by cpu load.

Background technology

At present, aspect cpu load prediction, what the system based on Android adopted is very simple and practical PAST algorithm, yet the cpu load predicated error of this algorithm is larger, thereby the frequency of most suitable CPU work can not be selected according to the load of prediction, so cannot reduce power consumption in guaranteed performance; In order to improve load estimation precision, the exponential average of Weight number adaptively (Dynamic Exponential average, DEXP) algorithm is suggested and carries out cpu load prediction, precision of prediction increases, but the peak value place that loads on of this algorithm predicts changes milder, error is still larger, although reduced power consumption, take that to sacrifice performance be cost; In order further to improve load estimation precision, linear prediction (Linear Prediction, LP) algorithm has been carried out, although this algorithm has improved load estimation precision effectively, but its weights are fixed, can not dynamically adjust with the variation of load, cannot guarantee that its weight is current optimum.

Summary of the invention

In order to address the above problem, the invention discloses a kind of low power consumption design method that moves Sink load estimation based on Android, the method can realize the dynamic adjustment of weights, further improves load estimation precision, thereby selects more suitably CPU calculated rate to reduce power consumption.

The object of the present invention is achieved like this:

The low power consumption design method that moves Sink load estimation based on Android, comprises the following steps:

Step 1: obtain the load load_cur of current C PU by the CPU driver of Android, enter step 2;

Step 2: the cpu load load_cur that obtains by step 1 upgrades history_load[3] historical load value in array, update mode is history_load[1] assignment is to history_load[2], history_load[0] assignment is to history_load[1], load_cur assignment is to history_load[0], make history_load[3] preserve all the time the cpu load value of current slot and the first two time period in array, enter step 3;

Step 3: judge history_load[3] the first two time period cpu load history value history_load[1 of current slot in array] and history_load[2] whether equate, if:

To ask current optimum weights α by following formula _best,

${\mathrm{\α}}_{\mathrm{best}}=\frac{\mathrm{history}\_\mathrm{load}\left[0\right]-\mathrm{history}\_\mathrm{load}\left[2\right]}{\mathrm{history}\_\mathrm{load}\left[1\right]-\mathrm{history}\_\mathrm{load}\left[2\right]}$

No, set α _bestvalue be 0.5,

Enter step 4;

Step 4: the α that utilizes step 3 to obtain _best, by the load value of following formula prediction next time period of CPU,

load_next＝α _best×history_load[0]+(1-α _best)history_load[1]

Obtain, after the cpu load value load_next of next time period, entering step 5;

Step 5: call _ _ cpufreq_driver_target function, regulate CPU frequency of operation according to the cpu load value load_next of prediction.

The present invention is based on Android and move the low power consumption design method of Sink load estimation, can realize the dynamic adjustment of weights, further improve load estimation precision, thereby select more suitably CPU calculated rate to reduce power consumption.

Accompanying drawing explanation

Fig. 1 the present invention is based on the low power consumption design method process flow diagram that Android moves Sink load estimation.

Embodiment

Below in conjunction with accompanying drawing, the specific embodiment of the invention is described in further detail.

The present embodiment based on Android, move the low power consumption design method of Sink load estimation, process flow diagram is as shown in Figure 1.The method comprises the following steps:

Step 1: obtain the load load_cur of current C PU by the CPU driver of Android, enter step 2;

Step 2: the cpu load load_cur that obtains by step 1 upgrades history_load[3] historical load value in array, update mode is history_load[1] assignment is to history_load[2], history_load[0] assignment is to history_load[1], load_cur assignment is to history_load[0], make history_load[3] preserve all the time the cpu load value of current slot and the first two time period in array, enter step 3;

Step 3: judge history_load[3] the first two time period cpu load history value history_load[1 of current slot in array] and history_load[2] whether equate, if:

To ask current optimum weights α by following formula _best,

${\mathrm{\α}}_{\mathrm{best}}=\frac{\mathrm{history}\_\mathrm{load}\left[0\right]-\mathrm{history}\_\mathrm{load}\left[2\right]}{\mathrm{history}\_\mathrm{load}\left[1\right]-\mathrm{history}\_\mathrm{load}\left[2\right]}$

No, set α _bestvalue be 0.5,

Enter step 4;

Step 4: the α that utilizes step 3 to obtain _best, by the load value of following formula prediction next time period of CPU,

load_next＝α _best×history_load[0]+(1-α _best)history_load[1]

Obtain, after the cpu load value load_next of next time period, entering step 5;

Step 5: call _ _ cpufreq_driver_target function, regulate CPU frequency of operation according to the cpu load value load_next of prediction.

Above step is done to finer explanation: the frequency adjustment work of Android system all completes in CPUFreq subsystem, in this subsystem, defined conservative, ondemand, userspace, five kinds of frequency modulation strategies of powersave and performance (cpufreq_policy), every kind of strategy has the frequency modulator (cpufreq_governor) of oneself, these frequency modulators all can call the work queue of a delay of dbs_timer_init () function initialization, every identical delay, just the work of calculating cpu load is joined in this queue, and cpu load calculating is called dbs_check_cpu () realization by do_dbs_timer (), the realization of SAWDLP algorithm is also in dbs_check_cpu () function, the load value of the current C PU getting in this function is cur_load, load Weight number adaptively linear prediction algorithm code is as follows:

For fear of real arithmetic, when calculating optimum weights, amplified 10 times, then when prediction load, dwindle 10 times.Next_load is the cpu load value of next time period of prediction, by this load, selects then a call _ _ cpufreq_driver_target function of suitable frequency to realize frequency modulation and operate.

Contrast with existing method, result shows: the cpu load value of the linear prediction algorithm prediction by Weight number adaptively of the present invention is compared percentage error with actual cpu load be 20.03%, than PAST algorithm 55.01%, 20.24% and DEXP algorithm of LP algorithm 22.65% all low.

Claims (1)

1. based on Android, move the low power consumption design method of Sink load estimation, it is characterized in that, comprise the following steps:

Step 1: obtain the load load_cur of current C PU by the CPU driver of Android, enter step 2;

Step 2: the cpu load load_cur that obtains by step 1 upgrades history_load[3] historical load value in array, update mode is history_load[1] assignment is to history_load[2], history_load[0] assignment is to history_load[1], load_cur assignment is to history_load[0], make history_load[3] preserve all the time the cpu load value of current slot and the first two time period in array, enter step 3;

Step 3: judge history_load[3] the first two time period cpu load history value history_load[1 of current slot in array] and history_load[2] whether equate, if:

To ask current optimum weights α by following formula _best,

${\mathrm{\α}}_{\mathrm{best}}=\frac{\mathrm{history}\_\mathrm{load}\left[0\right]-\mathrm{history}\_\mathrm{load}\left[2\right]}{\mathrm{history}\_\mathrm{load}\left[1\right]-\mathrm{history}\_\mathrm{load}\left[2\right]}$

No, set α _bestvalue be 0.5,

Enter step 4;

Step 4: the α that utilizes step 3 to obtain _best, by the load value of following formula prediction next time period of CPU,

load_next＝α _best×history_load[0]+(1-α _best)history_load[1]

Obtain, after the cpu load value load_next of next time period, entering step 5;

Step 5: call _ _ cpufreq_driver_target function, regulate CPU frequency of operation according to the cpu load value load_next of prediction.