CN109492264A - A kind of efficiency prediction technique across cloud data center - Google Patents
A kind of efficiency prediction technique across cloud data center Download PDFInfo
- Publication number
- CN109492264A CN109492264A CN201811208599.2A CN201811208599A CN109492264A CN 109492264 A CN109492264 A CN 109492264A CN 201811208599 A CN201811208599 A CN 201811208599A CN 109492264 A CN109492264 A CN 109492264A
- Authority
- CN
- China
- Prior art keywords
- data center
- cloud data
- efficiency
- model
- energy consumption
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/12—Protocols specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Hardware Design (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The efficiency prediction technique across cloud data center that the invention discloses a kind of, this method is using multiple cloud data center equipment energy consumption data of acquisition and its efficiency PUE value as training sample, the historical energy consumption data and its efficiency PUE value of cloud data center to be measured carry out feature extraction as calibration sample, to the energy consumption data of acquisition；To each cloud data center, the ridge regression model based on training sample is constructed respectively；Judge whether calibration sample is sky, if it is empty then directly carries out the fusion of domain adaptive model and then all ridge regression models are filtered with calibration sample if not empty, and filtered ridge regression model is subjected to the fusion of domain adaptive model；Efficiency prediction is carried out to cloud data center to be measured based on Fusion Model to distribute with adjustresources, and then improves the cloud data center.The present invention provides more personalized prediction, makes full use of data to carry out machine learning, improve forecasting efficiency in view of the otherness between different cloud data centers.
Description
Technical field
The present invention relates to cloud data center energy efficiency analysis method for air more particularly to a kind of efficiency prediction sides across cloud data center
Method.
Background technique
Cloud computing promotes the development of large data center, but simultaneously, because there is certain requirement to its computing capability, therefore cloud number
Huge energy consumption is also produced in cloud computing according to center.Elasticity service and the characteristics such as expansible due to cloud computing, data
The hardware size at center extreme expansion in recent years, so that the energy consumption problem of dispersion becomes concentration energy consumption problem, and energy in the past
The increase of growth and the network application of source price makes the efficiency problem of data center become very important.Therefore, cloud computing is studied
The efficiency prediction technique of data center has important application value to the energy management of data center.
Most distributed software develops to certain phase, and a data center can not meet demand.Usual one
City has multiple data centers, and by private line access, transmission delay is smaller for multiple data centers in a city.If demand
Amount is very big, for example company has user all over the world, and there will be very serious delays for transnational access service, and when a data
When center is not available, in order to guarantee normal offer service, it is necessary to the support of other data centers, therefore realize across data
The efficiency prediction at center has important practical significance.
Cloud computation data center is a kind of based on cloud computing framework, calculating, storage and Internet resources loose coupling, various IT
Device virtualization, modularization, the degree of automation are higher, the higher new types of data center of green energy conservation degree.General evaluation criteria
For energy use efficiency, i.e. efficiency, it is denoted as PUE, is defined as ratio (such as Fig. 2,3 of data center's total energy consumption Yu information technoloy equipment energy consumption
It is shown).PUE value is smaller, illustrates that data center's energy use efficiency is higher.
The characteristics of cloud data center is the virtualization of height first, among these include server, storage, network, using etc.
Virtualization, allows user to call various resources on demand；Followed by automatic management degree, including to physical server, virtual
The management of server, to the automatic flow management of related service, to automatic managements such as the charges of customer service；It is finally green
Color energy conservation, cloud computation data center meet green energy conservation standard in all respects, and general PUE value is no more than 1.5.
In recent years, increasingly pay attention to data center's efficiency both at home and abroad and optimize this project, carried out using machine learning techniques
The modeling of data center's efficiency and prediction have become research hotspot.The classical different types of calculation in machine learning and data mining
Method can be applied to the efficiency prediction of data center.
However most research there is also some problems in the efficiency modeling of cloud data center, directly against cloud data
The efficiency prediction model at center is fewer, and general advanced model is all based on CPU or server index, and cannot be guaranteed
The validity and accuracy of model.The energy consumption of different cloud data center components be it is discrepant, than if any main energy consumption group
Part is processor, and the main energy consumption component having is memory (as shown in Figure 4), therefore can not use unified machine learning model
It goes to predict all cloud data center efficiencies.
Summary of the invention
Goal of the invention: in view of the above problems, the present invention proposes a kind of across cloud data based on filtering domain adaptive model fusion
Center efficiency prediction technique provides personalized efficiency prediction to the cloud data center of different frameworks, has very high accuracy.
Technical solution: to achieve the purpose of the present invention, the technical scheme adopted by the invention is that: one kind is across cloud data center
Efficiency prediction technique, comprising steps of
(1) sensor is used to acquire N number of cloud data center equipment energy consumption data and its efficiency PUE value as training sample, to
The historical energy consumption data and its efficiency PUE value for surveying cloud data center are as calibration sample；
(2) feature extraction is carried out to the energy consumption data of acquisition；
(3) to each cloud data center, the ridge regression model based on training sample is constructed respectively；
(4) judge whether cloud data center calibration sample to be measured is sky, if not empty, thens follow the steps (5)；If it is empty,
Then directly by all ridge regression models of building alternately model, jumps and execute step (6)；
(5) all ridge regression models are filtered to obtain alternative model with calibration data；
(6) alternative model is subjected to the fusion of domain adaptive model；
(7) efficiency prediction is carried out to cloud data center to be measured based on Fusion Model, is distributed with adjustresources, and then improving should
Cloud data center.
The utility model has the advantages that 1, for the different feature of the main energy consumption component of different cloud data centers, be based on domain adaptive model
Fusion, it is contemplated that the otherness between different cloud data centers, provides personalized prediction, obtain good prediction effect；
2, it is directed to cloud data center efficiency Direct Modeling, abundant maintenance data and statistical machine learning method, improves cloud data to be measured
The accuracy rate and efficiency of center efficiency prediction.
Detailed description of the invention
Fig. 1 is the flow diagram of the method for the present invention；
Fig. 2 is data center's energy flow graph；
Fig. 3 is certain consumption of data center distribution map；
Fig. 4 is the main energy consumption composition figure of two kinds of servers.
Specific embodiment
Further description of the technical solution of the present invention with reference to the accompanying drawings and examples.
As shown in Figure 1, a kind of efficiency prediction technique across cloud data center of the present invention, comprising steps of
(1) N number of cloud data center equipment energy consumption data and its efficiency PUE value are acquired as training sample, cloud data to be measured
For the historical energy consumption data and its efficiency PUE value of center D as calibration sample, PUE is defined as data center's total energy consumption and information technoloy equipment
The ratio of energy consumption, the corresponding PUE value of each energy consumption data.Y indicates corresponding efficiency PUE value, and X indicates the energy of training sample
Consume eigenmatrix, XzThe energy consumption characters matrix for indicating z-th of cloud data center, shown in following formula:
X=(X1；X2；X3；……；XN),
Wherein, n is the number of data that z-th of data center includes.
The energy consumption characters matrix of calibration sample, shown in following formula:
Wherein, m is the item number of known energy consumption data in cloud data center D to be measured.
(2) feature extraction is carried out to the energy consumption data of acquisition:
The present embodiment is extracted 19 features relevant to energy consumption: director server IT load；Total nuclear network room IT load；
Water pump operation sum；Water pump frequency averaging driving speed；Compress water pump operation sum；Pressurized water pump frequency average driving speed；
Cooling tower operation sum；Cooling tower leaves the average set temperature of water temperature；Cooling-water machine operation sum；Dry and cold machine operation sum；Note
Enter the operation water pump sum of cold water；The water pump of injection cold water is averaged set temperature；The mean temperature of heat exchanger；Outdoor air is wet
Ball temperature；Outdoor air dry-bulb temperature；Outdoor air enthalpy；Outdoor air relative humidity；Outdoor wind speed；Outdoor wind direction.
The enthalpy of air refers to the insulation amount contained by air, usually on the basis of the unit mass of dry air.Enthalpy symbol
Number i indicates that unit is kj/kg dry air.
Enthalpy calculation formula are as follows:
I=1.01t+ (2500+1.84t) d
Wherein, t is air themperature DEG C；D is the water capacity kj/kg dry air of air.
(3) to the energy consumption characters matrix X of each cloud data centerzAnd corresponding efficiency PUE matrix Yz, find a recurrence system
Number βz, meet and minimize loss function
βz=((Xz)TXz+λI)-1(Xz)TYz,
Wherein, z indicates the number of ridge regression model serial number and corresponding known cloud data center, and z=1...N, λ are given ridge
Parameter, λ > 0, λ meet following four condition, and are completed by cross validation:
1) make each regression coefficient β1,...,βNRidge estimaion it is basicly stable；
2) when doing least-squares estimation, due to least square method β=(XTX)-1XTY can not seek the broad sense of non-non-singular matrix
It is inverse, ridge regression plus λ I this, allow matrix to become non-singular matrix and solve regression coefficient, regression coefficient becomes to close
Reason；
3) make βz>=0, meet practical significance at this time；
4) make residual sum of squares (RSS)Increase relatively small；
(4) judge whether cloud data center D calibration sample to be measured is sky, if not empty, then follow the steps (5) and carry out model
Filtering；If it is empty, then without filtering, directly by all ridge regression models of building alternately model, execution step is jumped
(6)；
(5) all ridge regression models are filtered to obtain alternative model with calibration data, comprising steps of
(5.1) using obtained N number of ridge regression model, it is calculated to the historical energy consumption data of cloud data center to be measured respectively
Efficiency PUE predicted value；
(5.2) obtained N number of PUE predicted value is subjected to mean absolute error MAE with the PUE true value as check and correction sample
Assessment, shown in following formula:
Wherein, m is calibration sample capacity,For the efficiency PUE predicted value of historical energy consumption data, yjFor PUE true value；
(5.3) the minimum number a of threshold values σ, model effective integration is set, and removal MAE is greater than the ridge regression model of threshold values, protects
Remaining M model is stayed, if M < a, retains the smallest a ridge regression model of MAE.
In order to guarantee the validity of Model Fusion, the present invention claims M >=3, if final MAE is less than the ridge regression model of σ
Number is less than 3, then compared with illustrating the energy consumption data at data with existing center and the calibration sample otherness of cloud data center D to be measured all
Greatly, retain the smallest three ridge regression models of MAE.
(6) alternative model is subjected to the fusion of domain adaptive model, following formula:
Wherein, βfinalFor final mask parameter；
(7) final mask parameter beta is utilizedfinalThe efficiency PUE value of one group of energy consumption data of cloud data center to be measured is carried out
Prediction, shown in following formula:
yPUE=ATβfinal,
WhereinFor energy consumption characters vector, n is the number extracted to cloud data center energy consumption characters to be measured,
α1,…,αnFor n energy consumption characters respective value, gross data is provided for the resource allocation of data center；For model
Parameter；For the efficiency predicted value of cloud data center to be measured, it is used for the comprehensive assessment data center, carries out resource point
The improvement matched, to improve and obtain the cloud data center of efficiency optimization.
Claims (6)
1. a kind of efficiency prediction technique across cloud data center, which is characterized in that comprising steps of
(1) sensor is used to acquire N number of cloud data center equipment energy consumption data and its efficiency PUE value as training sample, cloud to be measured
The historical energy consumption data and its efficiency PUE value of data center are as calibration sample；
(2) feature extraction is carried out to the energy consumption data of acquisition；
(3) to each cloud data center, the ridge regression model based on training sample is constructed respectively；
(4) judge whether cloud data center calibration sample to be measured is sky, if not empty, thens follow the steps (5)；If it is empty, then directly
It connects all ridge regression models of building alternately model, jumps and execute step (6)；
(5) all ridge regression models are filtered to obtain alternative model with calibration data；
(6) alternative model is subjected to the fusion of domain adaptive model；
(7) efficiency prediction is carried out to cloud data center to be measured based on Fusion Model to distribute with adjustresources, and then improve the cloud number
According to center.
2. a kind of efficiency prediction technique across cloud data center according to claim 1, it is characterised in that: in step (2),
The feature extraction, extract in cloud data center equipment including information technoloy equipment, refrigeration equipment, power Transmission equipment with efficiency
Relevant feature, including cpu busy percentage, air-conditioner temperature, air wind direction, for realizing the efficiency prediction to data center.
3. a kind of efficiency prediction technique across cloud data center according to claim 1, it is characterised in that: in step (3),
The ridge regression model is, to the energy consumption characters matrix X of each cloud data centerzAnd corresponding efficiency PUE matrix Yz, find one
Regression coefficient βz, meet and minimize loss function
βz=((Xz)TXz+λI)-1(Xz)TYz,
Wherein, z indicates the number of ridge regression model serial number and corresponding known cloud data center, z=1...N；λ is given ridge ginseng
Number, the selection of value can be completed by cross validation, λ > 0.
4. a kind of efficiency prediction technique across cloud data center according to claim 1, it is characterised in that: in step (5),
It is described filtering comprising steps of
(1) using obtained N number of ridge regression model, its efficiency is calculated to the historical energy consumption data of cloud data center to be measured respectively
PUE predicted value；
(2) obtained N number of PUE predicted value is subjected to commenting for mean absolute error MAE with the PUE true value as check and correction sample
Estimate, shown in following formula:
Wherein m is calibration sample capacity,For the efficiency PUE predicted value of historical energy consumption data, yjFor PUE true value；
(3) MAE calculated according to (2), the artificial minimum number a that threshold values σ, model effective integration is arranged, the setting of threshold values σ will be protected
The accuracy and precision of model of a syndrome, prevent model over-fitting, and removal MAE is greater than the ridge regression model of threshold values, retains remaining M
Model retains the smallest a ridge regression model of MAE if M < a.
5. a kind of efficiency prediction technique across cloud data center according to claim 1, it is characterised in that: in step (7),
The domain adaptive model merges following formula:
Wherein, βfinalFor final mask parameter.
6. a kind of efficiency prediction technique across cloud data center according to claim 1, it is characterised in that: step (7), institute
It states and is predicted as utilizing final mask parameter betafinalThe efficiency PUE value of cloud data center to be measured is predicted, shown in following formula:
yPUE=ATβfinal
Wherein, yPUE, A be respectively cloud data center to be measured efficiency predicted value and energy consumption characters vector, n be to cloud data to be measured
The number of power consumption feature extraction, α1,…,αnFor n energy consumption characters respective value, reason is provided for the resource allocation of data center
By data, total prediction of obtained cloud data center to be measured can valid value yPUEFor the overall merit data center, resource point is carried out
The improvement matched, and then improve and obtain the cloud data center of efficiency optimization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811208599.2A CN109492264A (en) | 2018-10-17 | 2018-10-17 | A kind of efficiency prediction technique across cloud data center |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811208599.2A CN109492264A (en) | 2018-10-17 | 2018-10-17 | A kind of efficiency prediction technique across cloud data center |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109492264A true CN109492264A (en) | 2019-03-19 |
Family
ID=65691429
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811208599.2A Withdrawn CN109492264A (en) | 2018-10-17 | 2018-10-17 | A kind of efficiency prediction technique across cloud data center |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109492264A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110781068A (en) * | 2019-11-07 | 2020-02-11 | 南京邮电大学 | Isomorphic decomposition method-based data center cross-layer energy consumption prediction method |
| CN110826784A (en) * | 2019-10-28 | 2020-02-21 | 腾讯科技（深圳）有限公司 | Energy use efficiency prediction method and device, storage medium and terminal equipment |
| CN110866592A (en) * | 2019-10-28 | 2020-03-06 | 腾讯科技（深圳）有限公司 | Model training method and device, energy efficiency prediction method and device and storage medium |
| CN110866528A (en) * | 2019-10-28 | 2020-03-06 | 腾讯科技（深圳）有限公司 | Model training method, energy consumption use efficiency prediction method, device and medium |
| CN111582604A (en) * | 2020-05-20 | 2020-08-25 | 中国工商银行股份有限公司 | Data processing method and device, electronic device, and medium |
| CN111723342A (en) * | 2020-06-22 | 2020-09-29 | 杭州电力设备制造有限公司 | Transformer top layer oil temperature prediction method based on elastic network regression model |
| CN113778215A (en) * | 2021-04-28 | 2021-12-10 | 龙坤(无锡)智慧科技有限公司 | Method for realizing data center PUE prediction and consumption reduction strategy based on big data |
| CN115164361A (en) * | 2022-06-13 | 2022-10-11 | 中国电信股份有限公司 | Data center control method and device, electronic equipment and storage medium |
-
2018
- 2018-10-17 CN CN201811208599.2A patent/CN109492264A/en not_active Withdrawn
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826784A (en) * | 2019-10-28 | 2020-02-21 | 腾讯科技（深圳）有限公司 | Energy use efficiency prediction method and device, storage medium and terminal equipment |
| CN110866592A (en) * | 2019-10-28 | 2020-03-06 | 腾讯科技（深圳）有限公司 | Model training method and device, energy efficiency prediction method and device and storage medium |
| CN110866528A (en) * | 2019-10-28 | 2020-03-06 | 腾讯科技（深圳）有限公司 | Model training method, energy consumption use efficiency prediction method, device and medium |
| CN110866592B (en) * | 2019-10-28 | 2023-12-29 | 腾讯科技（深圳）有限公司 | Model training method, device, energy efficiency prediction method, device and storage medium |
| CN110826784B (en) * | 2019-10-28 | 2023-12-12 | 腾讯科技（深圳）有限公司 | Method and device for predicting energy use efficiency, storage medium and terminal equipment |
| CN110866528B (en) * | 2019-10-28 | 2023-11-28 | 腾讯科技（深圳）有限公司 | Model training method, energy consumption use efficiency prediction method, device and medium |
| CN110781068A (en) * | 2019-11-07 | 2020-02-11 | 南京邮电大学 | Isomorphic decomposition method-based data center cross-layer energy consumption prediction method |
| CN110781068B (en) * | 2019-11-07 | 2022-11-08 | 南京邮电大学 | Data center cross-layer energy consumption prediction method based on isomorphic decomposition method |
| CN111582604A (en) * | 2020-05-20 | 2020-08-25 | 中国工商银行股份有限公司 | Data processing method and device, electronic device, and medium |
| CN111723342B (en) * | 2020-06-22 | 2023-11-07 | 杭州电力设备制造有限公司 | Transformer top layer oil temperature prediction method based on elastic network regression model |
| CN111723342A (en) * | 2020-06-22 | 2020-09-29 | 杭州电力设备制造有限公司 | Transformer top layer oil temperature prediction method based on elastic network regression model |
| CN113778215A (en) * | 2021-04-28 | 2021-12-10 | 龙坤(无锡)智慧科技有限公司 | Method for realizing data center PUE prediction and consumption reduction strategy based on big data |
| CN115164361A (en) * | 2022-06-13 | 2022-10-11 | 中国电信股份有限公司 | Data center control method and device, electronic equipment and storage medium |
| CN115164361B (en) * | 2022-06-13 | 2024-06-07 | 中国电信股份有限公司 | Data center control method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109492264A (en) | A kind of efficiency prediction technique across cloud data center | |
| Fonseca et al. | Integrated model for characterization of spatiotemporal building energy consumption patterns in neighborhoods and city districts | |
| Gmach et al. | Profiling sustainability of data centers | |
| Hedegaard et al. | Energy system investment model incorporating heat pumps with thermal storage in buildings and buffer tanks | |
| Islam et al. | Paying to save: Reducing cost of colocation data center via rewards | |
| CN104516768B (en) | The real-time migration of virtualization system | |
| Comaniciu | Nonparametric information fusion for motion estimation | |
| CN103150003B (en) | The power-saving control method of information handling system, this information handling system and device | |
| Amos-Abanyie et al. | Effects of Thermal Mass, Window Size, and Night‐Time Ventilation on Peak Indoor Air Temperature in the Warm‐Humid Climate of Ghana | |
| Ruddell et al. | Reducing a semiarid city’s peak electrical demand using distributed cold thermal energy storage | |
| CN103269364B (en) | A kind of distributed cloud computing data migration method | |
| US20100256958A1 (en) | Method for predicting cooling load | |
| Li et al. | Experimental validation of the simulation module of the water-cooled variable refrigerant flow system under cooling operation | |
| CN105452984B (en) | Calculate the power supply service efficiency in data center | |
| CN104035542B (en) | A kind of virtual machine migration method and system calculated with energy consumption for cooling balance | |
| CN109974231A (en) | Ice-storage air-conditioning ice gauge draws method, apparatus, equipment and medium | |
| US20160283844A1 (en) | Load predictor for a cooling system | |
| CN108808730A (en) | Consider the distribution network system reserve capacity for load variation in power computational methods and system of photovoltaic time space distribution | |
| Joe et al. | Virtual storage capability of residential buildings for sustainable smart city via model-based predictive control | |
| CN113806017A (en) | Method, system, equipment and storage medium for initial deployment of virtual machine in cloud computing | |
| Boehm | An approach to decreasing the peak electrical demand in residences | |
| CN114357668A (en) | Matching method and device for on-frame position of machine room equipment | |
| CN209342847U (en) | Distribution line fault outage prediction meanss and system based on weather and XBoost algorithm | |
| Rehman et al. | Technical implementation | |
| CN115455524A (en) | Energy-saving design method and system of prefabricated building based on BIM |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190319 |
|
| WW01 | Invention patent application withdrawn after publication |