KR102181065B1 - A Power Controlling Apparatus for Efficient Computer Power Management Based on Discrete Event System Specification - Google Patents

Abstract

The present invention relates to a discrete event system formalism based power control device for efficient power management of a computer. The discrete event system formalism based power control device for efficient power management of a computer comprises: a discrete event simulation engine that receives time series events generated from a CPU, a hard disk, a mouse, a keyboard, etc. and creates a list of events at regular intervals; and a discrete event simulation model that flexibly and precisely controls power according to a given event list even in a complex computing environment. The discrete case system model is consisting of a pure policy model, a process model for processing events related to CPU execution, a power prediction model for optimizing to satisfy requirements derived from the policy model and the process model as much as possible, and a power management model that can directly set system variables or call system functions of an operating system according to requests from the process model or power prediction model.

Description

A Power Controlling Apparatus for Efficient Computer Power Management Based on Discrete Event System Specification for efficient power management of computers

The present invention relates to a discrete event system specification (DEVS) formalism-based power control device for efficient power management of a computer, and more specifically, occurs in a CPU (Central Processing Unit), a hard disk, a mouse, a keyboard, etc. 1) Time-series events are received and divided into those related to pure power management policy and CPU performance. 2) After inputting each into a power model and a process model modeled in the form of a discrete event system, 3) the power model and It is based on the delivery of optimized control commands to the operating system to reflect the requirements of the process model as much as possible, and 4) the requirements of the process model requiring urgency are power control that is delivered directly to the operating system regardless of the power model. It relates to the device.

With the advent of the 4th industrial revolution era, the demand for artificial intelligence, big data, and blockchain technology is increasing, and the commonality of the above technologies is that they require fast calculation speed and large amount of calculation. To support this, high-end CPUs and GPUs appeared, and the era of electricity bills for personal computers ranging from millions of won to tens of millions of won per month has arrived.

For this reason, the need to effectively manage the power of the computer is emerging more than ever, and it is possible to increase the power efficiency of the hardware or enter the power saving mode by detecting the absence of user input, as well as to control the power consumption by software. I am looking for a way to be there. For example, reducing the number of devices constituting a computer, temporarily shutting off power according to the external environment (10-1833992, 10-1890672), or reducing standby power by using separate hardware (10-1759345, 10-2017) -0054616) Beyond the traditional method, a method of monitoring CPU usage with software and entering the power saving mode when there is no additional work have been proposed (10-1927872).

The conventional computer power management method detects a specific situation (eg, CPU or memory usage, keyboard/mouse input) to reduce the amount of power consumed by the computer device and makes an appropriate response. However, looking at the current computing environment, the types and characteristics of software installed and operated on computers are diversifying, and devices have an organic dependence. Therefore, it is difficult to understand the traditional method of managing power according to the usage of resources such as CPU and memory. Limits are being revealed.

For example, CPUs and GPUs on the market today can perform multi-threaded tasks based on multi-cores, so background processes that do not require fast processing speed depending on the situation of using computing resources are driven into one core. Performing this can be a way to lower the total power consumption, but the conventional method of detecting and processing only the CPU or memory usage cannot provide such a flexible and high-level control.

An object of the present invention is to provide a programmable power control device to perform efficient power management even in a complex multi-stage computing environment.

The computer is a system and the operation of the system can be described as the current state and the transition between states. In addition to external transitions due to events occurring from various devices while the computer is operating, even when there is no event, internal transitions according to the conditions of the current state occur simultaneously. In addition, the state transition of the system can be divided into a transition related to a process that requires CPU execution and a transition that does not.

In order to efficiently control the power of a system that causes state transitions in such a complex and simultaneous manner,

First, it can accurately express the current state of the system,

Second, it is possible to respond without missing out on various events generated from the devices constituting the system,

Third, while reflecting the power-saving policies of various scenarios, the state of the system is changed in an optimized manner according to the situation of the process performing work in the CPU.

Fourth, before controlling each device through the operating system, a software tool is needed that can analyze the results and predict the performance in advance.

To this end, the present invention provides a programmable power control device based on the Discrete Event System Specification (DEVS) formalism. Discrete event system refers to a system that changes state only at the time of occurrence of discrete events occurring at irregular time intervals due to internal conditions or external factors of the system, and the method of specifying this is called a discrete event system specification. According to the definition of a discrete event system, a computer corresponds to a discrete event system, and the processes running on the computer are another discrete event system running within the discrete event system. The state and transition rules of a discrete event system can be easily expressed by using a set theory based on discrete mathematics, but the mathematical framework for describing the state equation of a discrete event system as a set theory is a discrete event system formal theory. to be.

The present invention is a software invention that receives events occurring in devices such as CPU, hard disk, mouse, keyboard, etc. that make up a computer, and generates a policy-related event list required for a power management policy and a CPU execution-related event list of the executed process. It consists of'Discrete Event Simulation Engine' and'Discrete Event Simulation Model' that derives optimized power control items based on the policy model and process model programmed according to the discrete event system formalism based on the above event list and delivers it to the operating system. .

From the viewpoint of combination with hardware, the present invention is realized within the CPU and memory (RAM) of a computer. Specifically, the discrete event simulation engine refers to a table in which the types of events that can occur in real time, so that the code area memory Or, because it operates within a register of the CPU, and the event list is dynamically created at run-time, it uses the heap area memory, and the discrete event simulation model is programmed in the form of a function. Executed in stack memory that can push or pop the called function and its parameters.

By introducing the discrete event system formalism as above, not only did a tool to describe complex computing environments elaborately, but also the discrete state equations are programmed with software and can be replaced according to the circumstances of the computing environment and power management policies. You can verify and analyze various scenarios for and predict its performance.

When the power control device based on the discrete event system formalism of the present inventor is used, the optimal power control result in consideration of the CPU process situation is derived based on various power policy scenarios, resulting in efficient power management even in a complex multi-stage computing environment. Energy) can contribute to the ecosystem.

1 is an atomic model of the formal theory of a discrete event system.
2 is a combined model of a discrete event system formal theory.
3 is an example of a power saving policy expressed in the form of a discrete case system.
4 is an example in which the spooling process is expressed in the form of a discrete case system.
5 is an example in which the optimization of power saving only the monitor to maintain the ongoing download process is expressed in the form of a discrete case system.
6 is a block diagram of the present invention.
7 is an event table stored by the discrete event system engine.
8 is a list of events transmitted by the discrete event system engine to the discrete event system model.
9 is a conceptual diagram showing a memory area of a computer used in the present invention.

Since the present invention implements a power-related policy model, a CPU-related process model, and a power prediction model based on the discrete event system formalism, after explaining the state equation and the basic model, which are background knowledge of the discrete event system formalism, components of the present invention It will be described in detail (Fig. 6).

1. State Equation of Discrete Event System

The state variable of the discrete event system is expressed as q = (s, r), s is the discrete state variable, and r is a real value representing the elapsed time in the discrete state s. For example, the state variable q = (CPU_Busy, 5.3 seconds) representing the CPU state means that the CPU was staying in the busy state for 5.3 seconds. Discrete events Discrete events that change the state of the system can be divided into external events that occur according to external events and internal events that occur when internal conditions are satisfied.For example, in the case of a protocol system, external events are'message arrival'. , Internal event becomes'Time out'.

The state variable of the discrete event system and the transition between states can be expressed as the following state equation.

q'= δ _int * δ _ext (q,x)

q'= δ _int (q) represents the internal state transition when there is no input,

q'= δ _ext (q,x) represents the state transition when an external event in which x is input occurs.

The * symbol used in the above equation performs state transition by applying both functions δ _int (q) and δ _ext (q,x), but if there is a case to perform these simultaneously, they are sequentially determined according to a predetermined priority. It means performing as. The present invention implements specific details for efficient power management of a computer within the framework of the state equation of the discrete event system as described above.

2. Basic Model of Discrete Case System

The discrete event system formal theory sets the atomic model and the coupled model as the basic models of the discrete event system. The atomic model is an element that cannot be decomposed any more, and is expressed by the equation of state of the discrete event system. The bond model is a combination of two or more atomic models or basic models consisting of bond models.

In the basic model, the internal transition is indicated by a dotted arrow, and the external transition is indicated by a solid arrow. The state in bold solid line indicates the starting state of the model. '?' means that an external event has occurred, and'!' means that a message is transmitted from the discrete case system to the outside.

Conditions can be attached to the internal transitions of the atomic model. In the atomic model of Fig. 1, when condition 1 is satisfied,'state information' is sent and returned to the original state, and when condition 2 is satisfied, no message is transmitted and the state transitions to the original state.

The combined model is a combination of two or more atomic models. In Fig. 2, when the computer in the standby state receives a start command from the outside, an external transition is made from the standby state to the starting state, and the current state information is displayed in the starting state. The output internal transition continues. When a stop command is received in the starting state, it transitions back to the standby state.

3. Components of the present invention

end. Discrete Case System Engine (100)

The discrete event system engine receives events indicating the occurrence time of CPU, memory, mouse, keyboard, etc. from the operating system, and selects only events related to power management. To this end, a table listing various events that can occur in a computer is loaded into the code area memory or a register in the CPU and referred to in real time (Fig. 7).

The events related to power management are divided into pure policy related events and events related to CPU execution by referring to the same table (Fig. 7), and a'policy model that connects events occurring within a predetermined period (eg, 2 seconds) in chronological order. A related event list' and a'process model related event list' are respectively generated (Fig. 8). The event lists above are generated and destroyed dynamically at run-time, so they are processed in conjunction with the heap area of the memory area.

I. Policy Model (300)

Using discrete event system formalism, basic policies for computer power saving can be easily modeled. In Fig. 3, when the computer in the standby state receives the attendance message, it makes an external transition to the daily start state, and immediately transitions to the morning business state while controlling power to exert the highest performance. After 3 hours of working in the morning, it automatically transitions to the hibernation state, which saves power as much as possible. After 1 hour in hibernation state, power is controlled so that the highest performance can be achieved again, and it enters the afternoon business state, and after 5 hours in that state, the computer resumes hibernation while in standby state.

All. Process Model (400)

A process generally refers to a user program, a system program, that is, a running program processed by a CPU, and is also referred to as a job or task. The process state transition means that the state of the process changes while the process is in the system, and the power of the computer can be efficiently managed by properly scheduling the state of the process.

An example of process-related power control is spooling. Spooling is a direct input/output of data to be input/output to complement the sharing of input/output devices and the processing speed of relatively slow input/output devices and improve the performance of multiple programming systems. It is the process of saving to disk for later input/output without sending to the device. Rather than connecting to a printer line by line when there is information to be printed, it is possible to save the information to be printed on the hard disk at once and operate the printer with a slow processing speed, reducing processing time and reducing power consumption.

In Fig. 4, when the computer in the standby state receives the data output request event, it transitions to the spooling state, and the internal transition continues until the end of storing the hard disk. When an event indicating that the hard disk storage has been completed occurs, it returns to the standby state again, and the CPU process transitions to the data output state.

la. Power prediction model (500)

In a complex computing environment, it is necessary to comprehensively review the policy model 300 and the process model 400 to perform optimized power control to satisfy the maximum possible requirements. In the present invention, the power prediction model 500 performs its function. Perform.

For example, if you are downloading a large file over the Internet, you should not immediately put your computer to sleep just because a user input such as a keyboard or mouse has not been received for a certain period of time. That said, it's a waste of power if you keep all the devices on your computer active even though you're not doing anything other than downloading data. In this case, it is necessary to activate the communication unit or the hard disk required for Internet download and save power only on the monitor.

In FIG. 5, when the computer in the standby state receives the data download event, it is a task related to CPU execution, so it enters the Internet download state through the process model 400, and the internal transition to the same state continues while the data is being downloaded. Separately, if there is no user input such as a mouse or keyboard for 10 minutes, the policy model 300 transitions to the power saving mode. The power prediction model 500 transitions to an optimized power control state in which only the monitor saves power so as not to interfere with the download process by considering the requirements of the process model and the policy model as described above.

hemp. Power management model (600)

The power control action determined from the process model 400 or the power prediction model 500 is responsible for calling a system function of an operating system or setting a system variable directly. For example, in order to achieve'maximum performance', a system function that can set BasePriority for each process can be called through the Windows SDK (software development kit), and BasePriority can be set to IDLE for'maximum power saving'. In addition, it is possible to save power in the computing environment by calling the SetAffinity system function to distribute processes between cores with a complex power management policy within the multi-core CPU, moving all processes to one core and lowering the process priority.

Among the requirements of the process model 400, those requiring urgency are directly transmitted to the power management model 600 without consideration of the policy model 300, and when optimization with the policy model 300 is required, the power prediction model ( 500) is input to the power management model 600.

bar. About the combination with hardware

The discrete event simulation model 200 of the present invention is composed of a policy model 300, a process model 400, a power prediction model 500, and a power management model 600,

The policy model 300 and the process model 400 are each programmed in the form of a number of functions according to the discrete case system formalism, and a stack capable of pushing and popping the called functions and parameters. Is done in local memory,

The power prediction model 500 and the power management model 600 are each programmed in the form of a function in accordance with the formal theory of the discrete case system, and a stack area in which the called function and parameter can be pushed and popped. Run in memory.

Eventually, the discrete event system engine 100 generates an event list input to the policy model 300 and the process model 400 in the heap area memory while being executed in the code area memory or a register in the CPU, and functions The discrete event system model 200 implemented in the form is executed on a stack area memory (FIG. 9).

100: Discrete event simulation engine
200: Discrete event simulation model
300: policy model
400: process model
500: power prediction model
600: power management model

Claims (8)

In a power control device based on a discrete case system formalism for efficient power management of a computer,
A discrete event simulation engine 100 that receives time-series events of a device constituting a computer from an operating system and generates an event list every predetermined period; And
Consists of; a discrete event simulation model 200 that calculates power-related control contents by a program defined as a transition between the state of the computer and the state based on the event list and delivers it to the operating system,
The discrete event simulation model 200,
A policy model 300 for determining the types and contents of devices to be controlled according to policy-related scenarios;
A process model 400 for calculating power control items related to CPU performance;
A power prediction model 500 that comprehensively reviews the policy model 300 and the process model 400 to perform optimized power control; And
Consists of; a power management model 600 for calling a system function of an operating system or setting a system variable to reflect the result of the power prediction model 500,
The process model 400,
It manages the power of the computer efficiently by scheduling the state of the process, which is a running program processed by the CPU.
The power prediction model 500,
If your computer is downloading files over the Internet, you can activate the configuration required for Internet download and save only the monitor.
The power management model 600,
In order to achieve'maximum performance', a system function that can set BasePriority for each process is called through the Windows SDK (software development kit), and BasePriority is set to IDLE for'maximum power saving', and a multi-core CPU A power control device that saves power in the computing environment by moving all processes to one core by calling the SetAffinity system function to distribute processes between cores within, and lowering the process priority.
In claim 1,
In the event that the discrete event simulation engine 100 receives from the operating system, the device and time of occurrence are specified, and the discrete event simulation engine 100 selects only events related to power among the input events, Power control device that generates a list of policy-related events and process-related events listed in chronological order within a specified period by dividing into those related to and CPU performance
In paragraph 2,
The discrete event simulation engine selects an event related to power by referring to an event table stored in advance in a code area of memory or a register in the CPU, and then generates an event list related to power policy in the heap area of the memory, and the policy model 300 Power control device that transmits to the memory, generates a list of events related to CPU execution in the heap area of the memory, and delivers it to the process model 400
delete In claim 1,
The process model 400 can directly transmit the matters to be urgently controlled to the power management model 600 to request an immediate response of the operating system, and the matters determined by the policy model 300 according to the power policy Power control device that transfers matters that need to be optimized for power control in a way that is considered together to the power prediction model 500
In claim 1,
The policy model 300 and the process model 400 are implemented by programming matters to be controlled by one or more functions according to the discrete event system formalism, and the power prediction model 500 includes the policy model 300 and the A power control device that sets priorities between process models 400 or implements programming rules that can satisfy each requirement as much as possible
In claim 1,
The policy model 300, the process model 400, the power prediction model 500, and the power management model 600 are a power control device executed in a stack area of a memory.
In claim 1,
A power control device with a plug-in-plug-out function implemented to add a new function or replace or delete an existing function in the process of executing the policy model 300, the process model 400, and the power prediction model 500

Images