“Nowadays, the requirements for the performance of Electronic system equipment are getting higher and higher. When weighing the performance and power consumption of the electronic system, the performance of the electronic system often gets more attention. Batteries with limited capacity are the only source of energy for portable devices, and the increase in battery capacity is obviously not as fast as the increase in performance of the central processing unit. Therefore, how to use limited electrical energy to provide the highest performance for portable devices is the main source of power management in portable devices. Target. In addition, power management also takes into account stability and heat dissipation.The power management module is on the programmable power management device, providing application programming to realize various power consumption modes for power management
Nowadays, the requirements for the performance of electronic system equipment are getting higher and higher. When weighing the performance and power consumption of the electronic system, the performance of the electronic system often gets more attention. Batteries with limited capacity are the only source of energy for portable devices, and the increase in battery capacity is obviously not as fast as the increase in performance of the central processing unit. Therefore, how to use limited electrical energy to provide the highest performance for portable devices is the main source of power management in portable devices. Target. In addition, power management also takes into account stability and heat dissipation. The power management module is a software module that provides an application programming interface that implements various power consumption modes for power management on a programmable power management device.
There are two types of power consumption: static power and dynamic power. Static power consumption is mainly transistor leakage (leakage) power; dynamic power consumption is derived from the activation of circuit validity, such as the activation of register lines caused by the input of address lines or data lines. The power consumed by the switched capacitor is the most important part of dynamic power consumption, that is, the power consumed by the switched capacitor during the charging and discharging process at the output of the circuit:
From formula (1), it can be known that Pdynamic depends on the following four parameters: C (capacitance), V (voltage), f (signal frequency) and a (variable factor). where a is related to the number of 0-1 transitions that occur in the chip. Approaches to reducing dynamic power consumption are accordingly divided into 4 categories:
① Reduce the electric capacity or the stored power of the circuit.
② Reduce the activity of the switch. As computer chips are packaged with more and more complex functionality, the switching activity of the chip is enhanced, so reducing the switching activity plays an increasingly important role in reducing the dynamic power consumption of the system. Clock gating (time gating) technology is currently popular technology to reduce switch activity, this technology makes the clock signal does not pass through the idle application unit. Because the clock network is a major part of the chip’s power consumption, this technique can effectively reduce power and power consumption in the processor.
③ Reduce the signal frequency.
④ Reduce the voltage.
The latter two methods are at the expense of reducing system performance, but are also the main means to reduce system power consumption. System power management is to use reasonable power management strategies to balance the relationship between improving system performance and reducing system power consumption by recognizing system tasks and states, and providing system applications with the lowest power consumption. optimum performance. This paper discusses the management of reducing system power consumption from the perspectives of task information and system state, and proposes a system-level power management module architecture based on the highest decision.
1 Analysis of Power Management Technology
1.1 System Status, Task and Power Management Policy Information
The premise of power management is accurate detection and management of system device status and task information, as well as accurate efficiency statistics for power management strategies.
System states include Running (working) state, Idle (idle) state and Sleep (sleep) state. Some systems can provide a multi-mode working state, the difference is mainly in the processor operating frequency, operating voltage and device combinations. Real-time metrics for tasks include response time, latency, and task deadlines. Hard real-time tasks have hard requirements on these indicators. When the system cannot meet these indicators, the provided data or services will fail completely, and even cause catastrophic consequences; soft real-time tasks have only soft requirements for them, which cannot be achieved. The consequence of the indicator is simply not being able to provide the required quality of service. In addition to real-time metrics, task information also includes the equipment components used to perform the task. In the case of a multi-working mode system, also include the lowest working mode to perform the task. Many operating systems and processors provide good detection modules and functional units. For example, the timer function of Linux can provide system status monitoring timing, and the PMU (Performance Monitoring Unit) unit of the Intel XScale processor can be used to monitor the work of the XScale platform. The detection and management of system status and task information is one of the important parts in the management module.
Power management policy information includes core algorithms and work efficiency. The efficiency of the power management strategy can be measured by calculating the “competitive ratio” and the “error rate”. D. RamanathanThe metric “competition ratio” is used in analyzing power management strategies using competitive analysis methods. The premise of the competitive analysis method: It is assumed that the problem under study has a competitor, and this competitor can influence the input of the problem. The contention ratio is defined as the ratio between the resources consumed by the online strategy and the least possible resources to complete the task. Here, the online strategy is a power management strategy for systems with unknown loads. Because in the actual system, it is impossible to predict the arrival time of the next task request of the system completely and correctly, and the least resource that may be consumed is the resource consumed under the premise that the power management strategy can completely and correctly predict the arrival time of the next task request . The resources here can be simply replaced by power consumption, or it can be combined with the delay of system execution tasks, that is, system performance. The prediction error rate is an efficiency indicator for the prediction strategy, which is equal to the ratio between the number of prediction errors and the total number of predictions, and a Boolean number can be used to determine the evaluation function of the prediction error rate.
1.2 Power management method in Running state
When the system is in the Running state, the power management module achieves the purpose of reducing power consumption by converting the state or working mode of the system equipment on the premise of completing the task according to the task information. For example: Tasks are classified according to computation-intensive and access-intensive tasks. When performing computation-intensive tasks, the system power consumption can be reduced by reducing the bus frequency while ensuring the real-time requirements for task completion; tasks, the power consumption can be reduced by reducing the operating frequency of the processor.
When the system performs multiple workload tasks or performs multiple tasks at the same time, an effective power management strategy is a combination of task scheduling and task deadlines. The basic idea of this power management strategy is: Group tasks according to the equipment used and task sets, list all scheduling possibilities, exclude scheduling other than constraints (complete within the deadline), and try to concentrate the same group of tasks within the task deadline. Implemented so that system idle time is as concentrated as possible for dynamic power management. The task scheduling process applied to this strategy is shown in Figure 1.
Figure 1 Task scheduling process combined with task deadlines
The principles of scheduling tasks based on task scheduling and task deadline power management strategies are:
① The lower the scheduling energy consumption, the higher the priority; the tasks in the same group are sorted by deadline; the earlier the deadline of the first task of each group, the higher the priority of the group’s scheduling; the last task of each group has a higher priority; The earlier the deadline, the higher the priority of the group’s schedule.
② For scheduling with exactly the same scheduling energy consumption as the deadline, the one that arrives first has a higher priority.
③ When an external task requests to use the dormant device, the power management module rearranges the priority of the task.
Assuming a continuous function P(s), if the system equipment runs at a speed s, the power it consumes is P. According to the cube root principle of devices based on CMOS process, there are:
For the convenience of analysis, the relationship between power consumption and system equipment operating speed is expressed as the following formula:
This is a strictly convex function that conveys the message that the slower the task, the more power is saved. This is the basic starting point for power management strategies based on task deadline constraints to task completion. At present, there are many power management strategies based on task deadlines to constrain task completion, such as the simplified online strategy AVR(Average Rate), OA (Optimal Available) and BKP strategiesEtc. is typical of this type of strategy. Among them: AVR strategy assumes that only one task is being executed in the system; OA strategy assumes that no new tasks will be scheduled; and BKP strategy can reduce power consumption well when c is relatively large.
1.3 Power management method in Idle state
After the system device completes the task, the state transition of the system device in the Idle state is the main method of power management in this state. Mainstream strategies include Timeout strategy, prediction-based management strategy and random-based management strategy. Among them, the Timeout strategy is the simplest and easiest to implement. The strategy flow is shown in Figure 2.
Figure 2 Timeout strategy flow
After the system completes all tasks, when the duration in the Idle state exceeds the threshold, the power management module transitions the system to the Sleep state, and then wakes up the system until a new task request arrives. In this way, the purpose of reducing the power consumption of the system equipment is achieved. The time interval can be set by the timing module provided by the system, and the setting of the time threshold Tth is determined by the following formula:
In the formula: Etran is the total energy consumed by the known system from the Idle state to the Sleep state and then to wake-up, a total of two state transitions; PI is the power consumed by the system in the Idle state.
Figure 3 shows the loss of two performances in the Timeout strategy. In the figure, E is the Running (working) state time, I is the Idle state time, F is the time threshold, D is the state transition time, S is the sleep state time, and W is the system device wake-up time. The strategy is simple, but the disadvantages are also obvious. As shown in Figure 3, when I>F+D, the setting of the waiting time threshold is likely to lose more power reduction opportunities, and at the same time, due to the time-consuming and energy-consuming system state wake-up transition, it will inevitably cause task waiting delay; even when When F+D>I>F, the delay will be greater than the wake-up time, which will cause great performance loss; at the same time, the delay of task execution time will directly shorten the duration of the next Idle state. In this way, the power management strategy based on the prediction of the idle state time after the task is completed and the arrival time of the next task is very efficient.
Figure 3 Loss of two properties in the Timeout strategy
The prediction-based power management strategy is to predict the duration Tpred that the system will be in the Idle state according to system information (including historical information and user habits, etc.). Comparing Tpred and Tth, when Tpred ≥ Tth, the system is switched to the sleep state immediately after the task is completed; otherwise, the system Idle state is maintained. The prediction moment and the prediction interval in the Idle state are determined by the specific strategy.
The core of the prediction-based power management strategy is which algorithm to use to use the system feedback information to update the prediction basis of the algorithm. To make accurate predictions that conform to the user habits and task requests of system equipment users, it is necessary to continuously deepen the understanding of user habits, and to have comprehensive statistics on system task information and policy history information. Adaptive learning tree ALT (Adaptive Learning Tree) strategy, PBALT (ProbabilityBased ALT) strategy, and AR (AutoRegressive) model-based predictive control feedback PCF (Predictive Control Feedback) prediction strategy are all optimized prediction strategies. The PBALT strategy uses probability to reflect the accuracy rate, which strengthens the correlation between the sub-trees and the learning ability of the ALT method; but the boundary conditions of this strategy restrict its application range. The adaptability of the PCF prediction strategy is controlled by its feedback module; however, the prediction strategy itself is unstable when it is requested for tasks in a non-stationary state. At the same time, the prediction strategy basically only considers that the system has one working mode, all of which limit the its application.
Stochastic-based power management strategy is an optimization strategy with uncertainty, which originates from the abstraction of the system model. The random-based power management strategy not only specifies when to perform state transition, but also specifies which operating mode to switch to, so it is suitable for system devices with multiple operating modes. It treats dynamic power management as a stochastic optimization problem, instead of removing the uncertainty of task requests through a predictive approach like a predictive power management strategy. The stochastic decision dynamic power management strategy based on CTMDP (Continuous Time Markov Decision Process) gives an optimal decision for system power management, but this optimization is based on an uncertain model, namely The optimal decision obtained by this algorithm can only obtain an expected value of the performance and power consumption of the system, and cannot be guaranteed to be applicable in a specific system device, and the establishment of the Markov process mathematical model also requires careful analysis.
2 Power management strategy based on top decision
From the above analysis of the system power management strategy, it can be seen that the power management of the system equipment runs through each state of the system equipment. Therefore, a power management method should be proposed to combine multiple power management strategies to manage the system power consumption collaboratively. There is a set of strategies in the power management framework, each strategy has its own priority, and each strategy is used for multi-strategy power management according to the needs. However, there are also problems with this architecture: First, the tasks of a complex system are likely to be diverse, and the power management strategies have different power reduction efficiencies for different tasks. Only the priority of the power management strategy is used to determine the use of the power management strategy. Targeted; in addition, each policy information should be counted in the process of executing system tasks, and its priority should be changed adaptively.
A power management framework based on the highest decision management module is proposed here. This system equipment power management framework includes five main parts: the highest decision module, the task information statistics module, the strategy collection module, the information detection module and the control module, as shown in Figure 4.
Figure 4 Architecture of power reduction management module based on the highest decision
Information detection module: used to detect system status information and newly arrived task information.
Task information statistics module: It is used to count the task information performed by the system equipment and interpret it into accurate task information parameters.
Strategy collection module: Calculate the efficiency of the power management strategy through dynamic statistics on system status and task information, update the power management strategy information and interpret it into accurate power management strategy parameters.
The highest decision-making module: According to the received task and system status information, select the optimal power management strategy or power management strategy group in the strategy set, and perform power management on the system equipment through the control module.
The task information is received in real time; the system state information is provided by the information detection module to the highest decision-making module every time the system state changes; the information of the power management strategy refers to the calculated power management efficiency and the system to which the power management strategy applies. status and tasks. For example, when a new task arrives, there must be a prediction strategy that is most efficient in predicting the duration of the idle state after the task is completed. During power management policy control, the success or failure of each decision changes the priority weighting parameters of the power management policy. In this way, the highest decision-making module decides to adopt the optimal power management strategy or power management strategy group according to the system state and task information, so that each part of the system equipment can obtain the optimal power management.
The core of power management in today’s portable devices is the power management strategy. The key of the power management framework based on the highest decision in this paper is to preselect the set of power management strategies. Regarding the power management strategy, there are two issues that need to be discussed and studied: First, weigh the performance and power consumption of the system equipment. In the process of system power management by the power management strategy, although the power management strategy tries to avoid delays, such delays are unavoidable. The system user’s trade-off between performance and power consumption directly affects the choice of the power management strategy and the preset of specific parameters in the power management strategy. Second, trade-off power management effectiveness and complexity. The larger the size of the policy set and the task information set, the more complete the statistical information, and the more accurate the decision of the power management strategy, but at the same time the complexity of the power management module also increases, which is directly related to the complexity of its engineering implementation. In addition, establishing power management policy standards, providing power management policy packages and task information packages, and standardizing system status and task information will also be beneficial to the development of portable device power management technologies.