CN102221874A - Computer system real-time power estimation method - Google Patents
Computer system real-time power estimation method Download PDFInfo
- Publication number
- CN102221874A CN102221874A CN2011101549775A CN201110154977A CN102221874A CN 102221874 A CN102221874 A CN 102221874A CN 2011101549775 A CN2011101549775 A CN 2011101549775A CN 201110154977 A CN201110154977 A CN 201110154977A CN 102221874 A CN102221874 A CN 102221874A
- Authority
- CN
- China
- Prior art keywords
- load
- computer system
- cpu
- power
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention discloses a computer system real-time power estimation method. Load intensity is equally divided into 10 regions; the powers of a target computer system under different load intensity are respectively tested; and power consumption data under the load intensity is recorded and used as reference power data. The load comprises a CPU (central processing unit) load, an I/O (input/output) load and screen luminance; then the reference power data is inputted in the target computer system to regularly obtain data of the CPU load, the I/O (input/output) load and the screen luminance of the computer system; and data really tested is combined to implement real-time power estimation. The invention also provides an effective reference energy consumption data test method, which has more accurate test results. The computer system real-time power estimation method can be realized by software, so that the cost can be effectively reduced.
Description
Technical field
The present invention relates to computer system energy consumption testing and estimation, particularly the realtime power evaluation method in the computer system operational process.
Background technology
Reduce carbon emission and become a global problem, energy-saving and emission-reduction also are important state basic policies of China.Computer system, especially Large Scale Computer System, the electric power that expends is considerable.To power consumption of computer systems estimation is to calculate operation cost, carries out the important evidence of business migration between different system.
Existing method mainly comprises three kinds, and a kind of is to carry out accumulation calculating according to nominal power, and second kind is to use special-purpose instrument to measure, and the third is the estimation of carrying out realtime power according to the resource operating position of electronic equipment.Method one (power calculation method) is that the nominal power according to each parts of computer system carries out addition, obtains the power consumption values of computer system.This method in the great master v2.76 of the Shandong of 360 companies, adopt (
Www.ludashi.com), specifically be to detect mainboard respectively, processor, video card, hard disk, internal memory, the model of display and other parts, adding up according to the default performance number of each parts draws the energy consumption estimated value.This method has significant limitation, is example with the notebook U35JC of Asus, and its estimating power is 107 watts, and actual measured power only is 47.Method two (method of testing) use power analyzer can obtain the accurate energy consumption of particular computer system, but the power analyzer range is generally between 1kw～10kw.In addition, the shortcoming of using instrumentation to measure be surveying instrument cost an arm and a leg and the number of devices that at every turn can connect limited.Be not suitable for the test of Large Scale Computer System.Method three (monitoring and estimation) was mentioned in the patent 201010122026 of Huawei Company's application, and concrete mode provides a kind of energy consumption evaluation method and device is monitored and assessed the real time energy consumption of each electronic equipment.The energy consumption apparatus for evaluating of the method realizes that by hardware this increases with regard to the cost that makes the computer real-time power budget.On the other hand, this method be at various electronic equipments carry out energy consumption assessment but not specially at computer system, and the method by the type of pick-up unit, is carried out the prediction of energy consumption again according to the energy consumption calculation rule of type of device and preset value when carrying out the energy consumption estimation.The method also is not quite similar for the energy consumption calculation rule of distinct electronic apparatuses, and its energy consumption is relevant with device type and CPU usage for portfolio and electronic equipment that CPU has strong correlation.Energy consumption rule in this way, computer system should range the electronic equipment that has strong correlation with CPU, that is to say according to the energy consumption of this kind method computer system to calculate according to cpu busy percentage.But find in reality test, only exist than large deviation that the energy consumption of miscellaneous part remains and cannot be left in the basket, the energy consumption of bringing as screen intensity energy consumption, I/O load etc. according to cpu load estimation computer system energy consumption.This just brings certain influence to the accuracy of energy consumption estimation.
Therefore, there are the following problems at least in the prior art: 1. the power in the computer system operational process is not to be constantly equal to peak power or nominal power, uses peak power or nominal power evaluates calculation machine system energy consumption inaccurate.2. take multinomial power-saving technology in computer system, the especially portable computer at present,, reduce method decreasing energy consumptions such as screen intensity, use the static power computing method can't consider these factors by reducing cpu frequency.3. the dynamic power method of testing is except using specialized equipments such as power analyzer and analysis chip, and the energy consumption data that additive method is predicted out is not very accurate.Just carry out estimating of computer system energy consumption according to the CPU behaviour in service, this has just neglected the consumption of other resources to energy.4. existing dynamic power method of testing does not provide the method for testing of energy consumption reference data, causes user and method supplier so probably because different and the energy consumption estimation results is exerted an influence to the understanding of reference data method of testing.
Summary of the invention
In order to overcome the deficiency that the prior art estimation is inaccurate or testing cost is higher, the invention provides a kind of energy consumption evaluation method based on load information, obtain the actual test data of separate unit target computer system, the various loads of dynamic monitoring computer system then or computer cluster and relevant configuration information are according to related load data-evaluation realtime power.
The technical solution adopted for the present invention to solve the technical problems is:
At first use the power of power analyzer test target computer system under different loads, load mainly comprises cpu load, I/O load and screen intensity.Experiment shows three kinds of intensity of loads of energy consumption and this relevant (nonlinear dependence).Therefore for different loads, intensity of load is divided into 10 intervals, tests the power under the different loads intensity respectively, and write down the power consumption data under each intensity of load.These data will be predicted real time power consumption as the benchmark power consumption data.
The test of three kinds of loads needs independently to carry out.At first carry out the test of cpu load, test respectively cpu busy percentage from 0%, 10%, 20%～100% o'clock the computer system power of totally 11 sampled points, and be designated as a
0, a
1,～a
10Totally 11 numerical value.These 11 numerical value are divided into 10 intervals with cpu load, are designated as interval 1 respectively to interval 10, and I/O load and brightness also are to be divided into 10 intervals equally.0%, 10%, 20%～100% o'clock computer power of test I/O load (per second data throughout) (also can produce cpu load when producing the I/O load owing to application program, when test, need keep cpu busy percentage to be stabilized in fixed value, as 30%), and the computing machine power consumption number is designated as b
0, b
1,～b
10Totally 11 numerical value.The test that screen intensity is divided into also is divided into according to brightness 11 point samplings, during CPU and I/O are not done any operation, draw computing machine power consumption c respectively
0～c
10To preserve as basic energy consumption data corresponding to different types of load and power consumption number, as the foundation of dynamic energy consumption estimation.
Secondly, input reference power consumption data in the target computer system of needs estimation dynamic power, use software approach (API that uses operating system to provide) regularly to obtain computer system CPU, I/O load and screen intensity data, carry out the realtime power estimation in conjunction with the data of actual measurement.Cpu busy percentage as target computer system is x%, be positioned at i interval, i.e. the * 10% of utilization factor i*10%＜x%＜(i+1); The I/O load is y%, is positioned at j interval, and screen intensity is z%, is positioned at k interval, and then the realtime power of target computer system is:
W＝a
i+(a
i+1-a
i)*(x/10-i)+(b
j-b
0)+(b
j+1-b
j)*(y/10-j)+(c
k-c
0)+(c
k+1-c
k)*(z/10-k)
X: object computer cpu busy percentage, span are 0～100;
Y: object computer I/O load utilization, span are 0～100;
Z: object computer screen intensity, span are 0～100;
I:CPU utilization factor interval of living in, computing method are to round behind the x/10, and span is 0～9;
J:I/O load interval of living in, computing method are to round behind the y/10, and span is 0～9;
K: screen intensity interval of living in, computing method are to round behind the z/10, and span is 0～9;
a
i, b
j, c
k: the power consumption values that corresponds respectively to the different loads situation in the benchmark power consumption data.
The invention has the beneficial effects as follows:
1. the present invention can estimate the computer system power of running status according to benchmark energy consumption data and the real-time load data of various computing systems;
2. the benchmark energy consumption data that target computer system is tested can be applicable to the computer system of many similar configuration.In many computer clusters of forming with configuring computer system, can use this method to estimate the energy consumption of each node respectively at each node, and obtain the energy consumption of whole computer cluster.
3. except benchmark energy consumption data test process, energy consumption is estimated the stage does not need specialized equipment, uses this method to calculate and estimates that the cost of energy consumption is very low.
The invention provides the method for testing of an effective benchmark energy consumption data, this is that prior art did not propose.Secondly, the present invention estimates the power of the computer system of running status according to the real time data of benchmark energy consumption data and cpu load, I/O load and screen intensity.This only carries out the energy consumption prediction by cpu load for prior art and has made very big improvement.At last, this appraisal procedure can realize by software in Large Scale Computer System, can reduce cost effectively with respect to the method for using hardware testing and assessment energy consumption.
The present invention is further described below in conjunction with drawings and Examples.
Description of drawings
Fig. 1 is energy consumption reference data measuring phases method of testing figure.
Fig. 2 is that synoptic diagram is obtained and predicted to energy consumption forecast period data.
Fig. 3 is the cpu load figure that utilizes this method to obtain.
Fig. 4 is prediction of energy consumption and the comparison diagram of surveying energy consumption.
Embodiment
This method implements to be divided into that the benchmark energy consumption data obtains and energy consumption two stages of prediction, and wherein the energy consumption forecast period needs data that the basic energy consumption stage of obtaining the obtain foundation as the energy consumption prediction.
The embodiment that the energy consumption reference data is obtained is with reference to Fig. 1.
1) tested computer power supply is connected to power measuring, power measuring is connected to test machine, and connect test machine and tested computing machine.
2) start tested computing machine, thereon the running load generator program.
3) start test machine, and running load is controlled and the energy consumption data capture program on rising.
4), in the different periods, tested computing machine is generated the load of dissimilar and intensity, and obtain the energy consumption data under the different loads situation by test machine control.
5) need test CPU, the I/O different loads information relevant with screen.Cpu load, I/O load and screen intensity.Power under the different loads number percent is tested in three kinds of loads respectively, and the record related data.
6) at first carry out the test of cpu load, test the power that cpu load is respectively 0%～100% computer-chronograph system respectively, and be designated as a
0, a
1,～a
10Totally 11 numerical value, cpu load is divided into 10 intervals.
7) 0%～100% o'clock computer power of test I/O load (per second data throughout) (needing to keep cpu busy percentage is constant value, as 30%), and be designated as b
0, b
1,～b
10Totally 11 numerical value.
8) test of screen intensity also is divided into according to brightness 11 point samplings, during CPU and I/O are not done any operation, draw c respectively
1～c
10
The specific implementation method of energy consumption forecast period is with reference to figure 2
1) goes up the installation load monitoring software at all computing machines (configuration needs similar) that need carry out the dynamic energy consumption estimation, comprising load detecting, the module that energy consumption data is estimated and communication is relevant, the energy consumption reference data that the first step is obtained is as the foundation of energy consumption estimation simultaneously.
2) the load detecting module is regularly obtained the CPU and the disk utilization factor of system, and the API (Application Programming Interface) that provides by operating system can realize.
3) dynamic energy consumption of the method computing computer system that lists according to formula 1 of energy consumption estimation module.
4) communication module sends to server storage and uniform and expression with the dynamic energy consumption of estimating.
The cpu load variation diagram that Fig. 3 is to use the API of Windows system to gather
Fig. 4 is an energy consumption of utilizing the present invention to predict, and is very approaching with the energy consumption of reality test
As mentioned above, the present invention can add the dynamic evaluation that software is realized the Large Scale Computer System energy consumption by a small amount of common hardware.
Claims (2)
1. a computer system realtime power evaluation method is characterized in that comprising the steps:
Intensity of load is divided into 10 intervals, the power of test target computer system under different loads intensity respectively, and write down power consumption data under each intensity of load as the benchmark power consumption data, described load comprises cpu load, I/O load and screen intensity;
Input reference power consumption data in target computer system is regularly obtained computer system CPU, I/O load and screen intensity data, carries out the realtime power estimation in conjunction with the data of actual measurement, and the realtime power of target computer system is:
W＝a
i+(a
i+1-a
i)*(x/10-i)+(b
j-b
0)+(b
j+1-b
j)*(y/10-j)+(c
k-c
0)+(c
k+1-c
k)*(z/10-k)
X: object computer cpu busy percentage, span are 0～100;
Y: object computer I/O load utilization, span are 0～100;
Z: object computer screen intensity, span are 0～100;
I:CPU utilization factor interval of living in, computing method are to round behind the x/10, and span is 0～9;
J:I/O load interval of living in, computing method are to round behind the y/10, and span is 0～9;
K: screen intensity interval of living in, computing method are to round behind the z/10, and span is 0～9;
a
i, b
j, c
k: the power consumption values that corresponds respectively to the different loads situation in the benchmark power consumption data.
2. computer system realtime power evaluation method according to claim 1, it is characterized in that: described cpu load, the test of I/O load and screen intensity needs independently to carry out, at first carry out the test of cpu load, test respectively cpu busy percentage from 0%, 10%, 20%～100% o'clock the computer system power of totally 11 sampled points, and be designated as a
0, a
1,～a
10Totally 11 numerical value, these 11 numerical value are divided into 10 intervals with cpu load, are designated as interval 1 to interval 10 respectively; The per second data throughout that keeps test I under the constant situation of cpu busy percentage/O load then is 0%, 10%, 20%～100% o'clock a computer power, and the computing machine power consumption number is designated as b
0, b
1,～b
10Totally 11 numerical value; The test that screen intensity is divided into also is divided into according to brightness 11 point samplings, during CPU and I/O are not done any operation, draw computing machine power consumption c respectively
0～c
10
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011101549775A CN102221874A (en) | 2011-06-09 | 2011-06-09 | Computer system real-time power estimation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011101549775A CN102221874A (en) | 2011-06-09 | 2011-06-09 | Computer system real-time power estimation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102221874A true CN102221874A (en) | 2011-10-19 |
Family
ID=44778442
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011101549775A Pending CN102221874A (en) | 2011-06-09 | 2011-06-09 | Computer system real-time power estimation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102221874A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105760270A (en) * | 2016-01-21 | 2016-07-13 | 西北工业大学 | Energy consumption estimation method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030135772A1 (en) * | 2002-01-11 | 2003-07-17 | Ncr Corporation | Methods and apparatus for conserving battery power in an electronic shelf label system |
| CN101807108A (en) * | 2010-03-08 | 2010-08-18 | 成都市华为赛门铁克科技有限公司 | Energy consumption evaluation method and device |
-
2011
- 2011-06-09 CN CN2011101549775A patent/CN102221874A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030135772A1 (en) * | 2002-01-11 | 2003-07-17 | Ncr Corporation | Methods and apparatus for conserving battery power in an electronic shelf label system |
| CN101807108A (en) * | 2010-03-08 | 2010-08-18 | 成都市华为赛门铁克科技有限公司 | Energy consumption evaluation method and device |
Non-Patent Citations (1)
| Title |
|---|
| 王振: "浅谈计算机的功率问题", 《新疆有色金属》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105760270A (en) * | 2016-01-21 | 2016-07-13 | 西北工业大学 | Energy consumption estimation method and device |
| CN105760270B (en) * | 2016-01-21 | 2019-01-01 | 西北工业大学 | A kind of method and device of Estimation of energy consumption |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Witkowski et al. | Practical power consumption estimation for real life HPC applications | |
| Noureddine et al. | A review of energy measurement approaches | |
| Ahmad et al. | A Review on mobile application energy profiling: Taxonomy, state-of-the-art, and open research issues | |
| Hackenberg et al. | Power measurement techniques on standard compute nodes: A quantitative comparison | |
| US9311209B2 (en) | Associating energy consumption with a virtual machine | |
| Lin et al. | A cloud server energy consumption measurement system for heterogeneous cloud environments | |
| US8086882B2 (en) | Energy measurement techniques for computing systems | |
| Xiao et al. | A system-level model for runtime power estimation on mobile devices | |
| Kim et al. | Enhancing online power estimation accuracy for smartphones | |
| US20120084028A1 (en) | Framework and Methodology for a Real-Time Fine-Grained Power Profiling with Integrated Modeling | |
| Lee et al. | PowerTrain: A learning-based calibration of McPAT power models | |
| CN101650685A (en) | Method and device for determining energy efficiency of equipment | |
| Kim et al. | FEPMA: Fine-grained event-driven power meter for android smartphones based on device driver layer event monitoring | |
| Minartz et al. | Simulation of power consumption of energy efficient cluster hardware | |
| Diouri et al. | Assessing power monitoring approaches for energy and power analysis of computers | |
| Salinas-Hilburg et al. | Unsupervised power modeling of co-allocated workloads for energy efficiency in data centers | |
| Tang et al. | NIPD: Non-intrusive power disaggregation in legacy datacenters | |
| US20130211752A1 (en) | Software power analysis | |
| Ghaleb | Software energy measurement at different levels of granularity | |
| Goel et al. | Techniques to measure, model, and manage power | |
| Jang et al. | Exploiting application/system-dependent ambient temperature for accurate microarchitectural simulation | |
| CN103293424A (en) | Data center power usage efficiency (PUE) determination method | |
| Quintiliani et al. | Understanding “workload-related” metrics for energy efficiency in Data Center | |
| Chinnici et al. | Data center, a cyber-physical system: improving energy efficiency through the power management | |
| CN102221874A (en) | Computer system real-time power estimation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20111019 |