Movatterモバイル変換


[0]ホーム

URL:


CN112395150B - Method for sensing power consumption of computing node - Google Patents

Method for sensing power consumption of computing node
Download PDF

Info

Publication number
CN112395150B
CN112395150BCN201910764228.0ACN201910764228ACN112395150BCN 112395150 BCN112395150 BCN 112395150BCN 201910764228 ACN201910764228 ACN 201910764228ACN 112395150 BCN112395150 BCN 112395150B
Authority
CN
China
Prior art keywords
data
power consumption
sensing
band system
node
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.)
Active
Application number
CN201910764228.0A
Other languages
Chinese (zh)
Other versions
CN112395150A (en
Inventor
龚道永
刘沙
钱宇
宋长明
李伟东
张宏宇
刘睿涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuxi Jiangnan Computing Technology Institute
Original Assignee
Wuxi Jiangnan Computing Technology Institute
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wuxi Jiangnan Computing Technology InstitutefiledCriticalWuxi Jiangnan Computing Technology Institute
Priority to CN201910764228.0ApriorityCriticalpatent/CN112395150B/en
Publication of CN112395150ApublicationCriticalpatent/CN112395150A/en
Application grantedgrantedCritical
Publication of CN112395150BpublicationCriticalpatent/CN112395150B/en
Activelegal-statusCriticalCurrent
Anticipated expirationlegal-statusCritical

Links

Images

Classifications

Landscapes

Abstract

The invention discloses a power consumption sensing method for a computing node, which comprises the following steps: operating an out-of-band system; the data sensing and registering program module periodically calls the data sensing and converting module to acquire power consumption data of the nodes; the data sensing and converting module reads the original voltage and current data of the sensor and converts the original voltage and current data from the sensor into power consumption data of the node; the data sensing and registering program module writes the power consumption data acquired from the data sensing and converting module into a register of an in-band system CPU; after waiting for the time T, the data sensing and registering program module is transferred to the step two, and the process is circulated; when the out-of-band system needs to acquire the node power consumption data, the power consumption query interface calls the data sensing and conversion module to acquire the node power consumption data. The invention solves the problems that the existing computer node power consumption data sensing method highly depends on in-band and out-band network transmission, has poor timeliness and high cost and incomplete data.

Description

Method for sensing power consumption of computing node
Technical Field
The invention relates to a power consumption sensing method for a computing node, and belongs to the technical field of computers.
Background
With the increasing concern of people on the problem of computer power consumption, the demand of computer power consumption monitoring and scheduling management is increasingly strong. The power consumption of a computer system is not only related to hardware design, but also is inseparable from specific usage patterns, node loads, and the like, and therefore, it is becoming more and more important to control the power consumption by optimizing scheduling means. When the power consumption of the computer is optimized and scheduled, the premise and the foundation are accurate and timely power consumption monitoring data, and on the basis, the task process is scheduled through software service in the node, the resource use mode is optimized, and the like, so that the power consumption of the node in the computer node and the power consumption of the node outside the node need to be accurately sensed in time, and the targeted optimized scheduling management can be performed by combining the current node resource use mode, the node task load and the like.
At present, the following methods are mainly used for sensing the power consumption of a computer node: 1. the local power consumption data of the nodes are acquired through the sensors with the internal nodes, the sensors are usually carried out aiming at independent node components, and the corresponding power consumption cannot be obtained for the components without corresponding sensors, so that the complete power consumption data of the nodes cannot be obtained; 2. the method is obtained by a sensor in an out-of-band system, the nodes are taken as a whole at the moment, complete power consumption information can be obtained, but the information needs to be transmitted to the in-band system of the nodes through a network, the method depends on the network, and the timeliness is poor; 3. the complete power consumption data of the nodes can be obtained by externally connecting a power consumption analyzer on a power supply line of the computing nodes, but a special instrument needs to be externally connected, the cost is high, and the method is not suitable for rapidly acquiring data of a large number of high-density assembled systems. Therefore, the existing computer node power consumption data sensing method has the problems of high dependence on in-band and out-band network transmission, poor timeliness, high cost, incomplete data and the like, and the requirements of efficiently and accurately sensing node power consumption in-band and out-of-band and performing power consumption optimization scheduling are difficult to meet at the same time.
Disclosure of Invention
The invention aims to provide a power consumption sensing method for a computing node, which solves the problems of high dependence on in-band and out-band network transmission, poor timeliness, high cost and incomplete data of the conventional power consumption data sensing method for the computing node, meets the requirements of in-band and out-band rapid power consumption sensing, and is suitable for large-scale system batch resource power consumption monitoring.
In order to achieve the purpose, the invention adopts the technical scheme that: a power consumption sensing method of a computing node is based on the following configuration:
the readable and writable chip-level register is configured on the CPU of the in-band system and is used for the interaction of power consumption data between the in-band system and the out-of-band system;
establishing a proprietary access protocol between an out-of-band system and an in-band system CPU (central processing unit), so that the out-of-band system can read and write a register of the in-band system CPU;
the sensor is configured in the out-of-band system and used for sensing original voltage data and original current data of power consumption of the target node;
the data sensing and converting module is configured in the out-of-band system and used for reading and converting original voltage data and original current data from the sensor to obtain power consumption data of the node;
the data sensing and registering program module is configured in the out-of-band system and used for calling the data sensing and converting module and recording the power consumption data from the data sensing and converting module into a register of a CPU (central processing unit) of the in-band system;
the power consumption query interface is configured on the out-of-band system and used for calling the data sensing and conversion module to acquire the real-time power consumption data of the node;
the method for sensing the power consumption of the computing node comprises the following steps:
s1, operating an out-of-band system;
s2, the data sensing and registering program module periodically calls the data sensing and converting module to acquire power consumption data of the nodes;
s3, reading the original voltage and current data of the sensor by the data sensing and converting module, and converting the original voltage and current data from the sensor into power consumption data of the node;
s4, the data sensing and registering program module writes the power consumption data acquired from the data sensing and converting module into a register of an in-band system CPU;
s5, after waiting for the set time T, the data sensing and registering program module is switched to S2, and the process is circulated;
s6, when the out-of-band system needs to acquire the node power consumption data, the power consumption query interface calls a data sensing and converting module to acquire the node power consumption data;
and S7, when the in-band system needs to acquire the node power consumption data, the in-band system reads the node power consumption data stored in the register by calling a software interface.
The further improved scheme in the technical scheme is as follows:
1. in the above scheme, the proprietary access protocol is a JTAG protocol, a GMII protocol, or a custom serial protocol.
Due to the application of the technical scheme, compared with the prior art, the invention has the following advantages:
the method for sensing the power consumption of the computing node solves the problems that the conventional method for sensing the power consumption data of the computing node highly depends on in-band and out-band network transmission, has poor timeliness, high cost and incomplete data, does not depend on in-band and out-band networks, has low cost, meets the requirements of in-band and out-band rapid sensing power consumption, and is suitable for large-scale system batch resource power consumption monitoring; by customizing the out-of-band management system and adopting the modes of out-of-band source sensing and in-band and out-of-band reading and writing of the in-band CPU register, the power consumption data of the computer node can be completely acquired, in-band and out-of-band rapid acquisition of the power consumption data of the node can be supported, in-band and out-of-band network interaction is not needed, the cost is low, and the method is suitable for large-scale computer systems to perform batch power consumption sensing. It is measured that the data sensing overhead can be reduced from at least tens of milliseconds in the conventional manner to the microsecond level.
Drawings
FIG. 1 is a flow chart of a method for sensing power consumption of a compute node.
Detailed Description
Example (b): a power consumption sensing method of a computing node is based on the following configuration:
the readable and writable chip-level register is configured on the CPU of the in-band system and is used for the interaction of power consumption data between the in-band system and the out-of-band system;
establishing a proprietary access protocol between an out-of-band system and an in-band system CPU (central processing unit), so that the out-of-band system can read and write a register of the in-band system CPU;
the sensor is configured in the out-of-band system and used for sensing original voltage data and original current data of power consumption of the target node;
the data sensing and converting module is configured in the out-of-band system and used for reading and converting original voltage data and original current data from the sensor to obtain power consumption data of the node;
the data sensing and registering program module is configured in the out-of-band system and used for calling the data sensing and converting module and recording the power consumption data from the data sensing and converting module into a register of a CPU (central processing unit) of the in-band system;
the power consumption query interface is configured on the out-of-band system and used for calling the data sensing and conversion module to acquire the real-time power consumption data of the node;
the method for sensing the power consumption of the computing node comprises the following steps:
s1, operating an out-of-band system;
s2, the data sensing and registering program module periodically calls the data sensing and converting module to obtain power consumption data of the nodes;
s3, reading the original voltage and current data of the sensor by the data sensing and converting module, and converting the original voltage and current data from the sensor into power consumption data of the node;
s4, the data sensing and registering program module writes the power consumption data acquired from the data sensing and converting module into a register of an in-band system CPU;
s5, after waiting for the set time T, the data sensing and registering program module is switched to S2, and the process is circulated;
s6, when the out-of-band system needs to acquire the node power consumption data, the power consumption query interface calls the data sensing and converting module to acquire the node power consumption data;
and S7, when the in-band system needs to acquire the node power consumption data, the in-band system reads the node power consumption data stored in the register by calling a software interface.
The proprietary access protocol described above is the JTAG protocol.
The examples are further explained below:
first, designing a computer in-band system:
(1) And the CPU provides a readable and writable chip IO register (Rpower) for the in-band and out-of-band system to interact power consumption data. The in-band CPU core can directly read and write the register, and the out-of-band system can read and write the register through a maintenance protocol;
(2) And establishing a proprietary access protocol (such as JTAG) between the out-of-band embedded CPU and the in-band CPU, and using the proprietary access protocol for the out-of-band system to read and write the register of the in-band CPU.
Secondly, designing a node level out-of-band system:
(1) A sensor for sensing original voltage dv and current dI data of power consumption of a target node is arranged in the out-of-band system;
(2) A data sensing and registering program module is built in, and the data sensing and converting module is called periodically to read and convert the sensor data and record a result (dpower) into a result register (Rpower) of an in-band CPU;
(3) The out-of-band system provides a power consumption query interface, and a data sensing and conversion module is called in the query interface to acquire real-time power consumption data (dpower) of the node;
(4) The method comprises the following functions of supporting node power consumption sensor configuration, data sensing period t configuration and the like.
Thirdly, the data sensing, registering and accessing process based on the register:
(1) And the data perception and registration process of the out-of-band system:
1. the out-of-band system is always operated after being powered on;
2. the data sensing and registering program periodically calls the data sensing and converting module to obtain node power consumption data;
3. the data sensing and converting module reads the original voltage and current data (dv, dI) of the sensor;
4. the data perception and conversion module converts the original data into power consumption data (dpower = f (dv, dI)) of the nodes according to a formula;
5. the data sensing and registering program writes the power consumption data (dpower) acquired from the data sensing and converting module into a result register (Rpower) of the in-band CPU;
6. the data sensing and registration program waits for time t and then goes to 2 to continue the data sensing, conversion and registration process.
(2) And the power consumption data flow of the out-of-band system inquiry node:
1. calling a power consumption data query interface in the out-of-band system;
2. and the data query interface calls a data sensing and conversion module to acquire node power consumption data (dpower) as the node power consumption data.
(3) And the in-band system acquires the node power consumption data flow:
1. calling a software interface by an in-band software process to inquire power consumption data;
2. reading the value (dpower) of a result register (Rpower) in the software interface;
3. and the software interface returns the dpower as the node power consumption data.
To facilitate a better understanding of the invention, the terms used herein will be briefly explained as follows:
power consumption perception: a method for obtaining power consumption data of computer nodes.
Controlling power consumption: a method of adjusting power consumption data of a target resource by changing its state and usage pattern.
The calculation node: the computer host machine is used for bearing computing functions in the parallel computer system.
In-band system: the computer host is used for executing various application tasks, and comprises a CPU, an internal memory, a hard disk, a network card and the like.
Out-of-band system: the computer node is provided with an embedded system for controlling and managing the in-band system, and the system is independent of the in-band system.
When the method for sensing the power consumption of the computing node is adopted, the problems that the existing method for sensing the power consumption data of the computing node highly depends on in-band and out-band network transmission, has poor timeliness, high cost and incomplete data are solved, the method does not depend on in-band and out-band networks, has low cost and low cost, meets the requirements of in-band and out-band rapid sensing power consumption, and is suitable for large-scale system batch resource power consumption monitoring; by customizing the out-of-band management system and adopting the modes of out-of-band source sensing and in-band and out-of-band reading and writing of the in-band CPU register, the power consumption data of the computer node can be completely acquired, in-band and out-of-band rapid acquisition of the power consumption data of the node can be supported, in-band and out-of-band network interaction is not needed, the cost is low, and the method is suitable for large-scale computer systems to perform batch power consumption sensing. It is measured that the data perception overhead can be reduced from at least tens of milliseconds in the traditional way to the microsecond level.
The above embodiments are merely illustrative of the technical ideas and features of the present invention, and the purpose thereof is to enable those skilled in the art to understand the contents of the present invention and implement the present invention, and not to limit the protection scope of the present invention. All equivalent changes and modifications made according to the spirit of the present invention should be covered within the protection scope of the present invention.

Claims (2)

CN201910764228.0A2019-08-192019-08-19Method for sensing power consumption of computing nodeActiveCN112395150B (en)

Priority Applications (1)

Application NumberPriority DateFiling DateTitle
CN201910764228.0ACN112395150B (en)2019-08-192019-08-19Method for sensing power consumption of computing node

Applications Claiming Priority (1)

Application NumberPriority DateFiling DateTitle
CN201910764228.0ACN112395150B (en)2019-08-192019-08-19Method for sensing power consumption of computing node

Publications (2)

Publication NumberPublication Date
CN112395150A CN112395150A (en)2021-02-23
CN112395150Btrue CN112395150B (en)2022-10-04

Family

ID=74603438

Family Applications (1)

Application NumberTitlePriority DateFiling Date
CN201910764228.0AActiveCN112395150B (en)2019-08-192019-08-19Method for sensing power consumption of computing node

Country Status (1)

CountryLink
CN (1)CN112395150B (en)

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
US8103884B2 (en)*2008-06-252012-01-24International Business Machines CorporationManaging power consumption of a computer
CN102664742A (en)*2012-04-282012-09-12浪潮电子信息产业股份有限公司Power dynamic management method based on embedded server
CN105487634B (en)*2015-11-242018-04-10无锡江南计算技术研究所A kind of quantization power consumption control method towards isomery many-core chip
CN106227636A (en)*2016-07-202016-12-14国网安徽省电力公司信息通信分公司A kind of data center based on IPMI outband management system

Also Published As

Publication numberPublication date
CN112395150A (en)2021-02-23

Similar Documents

PublicationPublication DateTitle
CN101534021B (en)Multimode data acquisitions and processing method applied to power automation system
CN107695775B (en)Heavy digital control machine tool heat error compensation control system and thermal error compensation method based on CPS
CN103631596A (en)Configuration device and configuration method of business object data entry and updating rule
CN103200199A (en)Out of band (OOB) data collection system
CN101369142A (en) Remote I/O data acquisition system and method based on embedded platform
CN102122431A (en)Method and system for intelligently diagnosing faults of electricity information collection terminal
CN106443166B (en) A meter reading system based on the Internet of Things platform
WO2010047170A1 (en)Calculation device, system management device, calculation method, and program
CN112000659B (en) A basic application system suitable for intelligent terminal equipment in distribution area
CN102495591A (en)Method for monitoring numerical control machine
CN110275470A (en) A Configurable Remote Data Acquisition System Applicable to Various Working Conditions
CN112395150B (en)Method for sensing power consumption of computing node
CN110795480B (en)Traffic operation data processing method and device
CN101217438A (en) Electric energy harvesting hybrid networking communication technology
CN1313903C (en)Method and apparatus for obtaining physical layer chip condition information
CN219105371U (en)Multimode environment monitoring system of thing networking
CN103021088B (en)Intelligent cloud ammeter
CN202047868U (en)On-line mine dust monitoring system based on OPC (Organic Photoconductor)
CN205212551U (en)Internet of things platform electric power monitored control system
CN115549782B (en)Information query method of optical module and related components thereof
CN201549934U (en)Remote terminal device
CN117033952A (en)Distributed multi-data source acquisition implementation method
CN109922247A (en)A kind of smart camera and a kind of image processing method
CN210111691U (en)Distributed new energy storage system
CN205691633U (en)A kind of Intelligent reading list processing equipment reducing electric-power metering line loss

Legal Events

DateCodeTitleDescription
PB01Publication
PB01Publication
SE01Entry into force of request for substantive examination
SE01Entry into force of request for substantive examination
GR01Patent grant
GR01Patent grant

[8]ページ先頭

©2009-2025 Movatter.jp