CN104298536A - Dynamic frequency modulation and pressure adjustment technology based data center energy-saving dispatching method - Google Patents
Dynamic frequency modulation and pressure adjustment technology based data center energy-saving dispatching method Download PDFInfo
- Publication number
- CN104298536A CN104298536A CN201410526996.XA CN201410526996A CN104298536A CN 104298536 A CN104298536 A CN 104298536A CN 201410526996 A CN201410526996 A CN 201410526996A CN 104298536 A CN104298536 A CN 104298536A
- Authority
- CN
- China
- Prior art keywords
- task
- data
- server
- cluster
- 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.)
- Pending
Links
Landscapes
- Power Sources (AREA)
Abstract
The invention provides a dynamic frequency modulation and pressure adjustment technology based data center energy-saving dispatching method. The dynamic frequency modulation and pressure adjustment technology based data center energy-saving dispatching method comprises firstly performing DVFS (Dynamic Voltage And Frequency Scaling) technology based performance model and energy consuming model modeling work on machines in a cluster; forecasting energy consumption data after task dispatching through a performance model and an energy consuming model; guiding real-time dispatching of an online task in the integral cluster according to the forecasted energy consuming data. According to the dynamic frequency modulation and pressure adjustment technology based data center energy-saving dispatching method, single-point dynamic frequency modulation and pressure adjustment technology of a traditional server and the dispatching task in the integral cluster are combined, a central management node in the integral cluster can obtain an analysis result according to real-time data of every server to perform online task distribution which is more beneficial to energy saving, and finally energy consumption of the integral data center can be saved.
Description
Technical field
The present invention relates to method for scheduling task, the particularly performance of one based on dynamic frequency voltage-regulating technique (Dynamic Voltage and Frequency Scaling, DVFS) and the method for scheduling task of energy consumption model in psychological field scape in the data.The method can distribute current task in the cluster also for processor arranges rational frequency of operation, reaches energy-conservation target.
Background technology
The energy consumption problem of data center has become more serious in recent years.Data show: 1 percent five of global total energy consumption and 2 percent of the whole America total energy consumption consume by data center, and this number percent also goes up gradually in the arrival along with large data age.Wherein, the energy consumption that processor (CPU) works occupies nearly 50% of total energy consumption, improves the efficiency of processor, becomes the importance reducing energy consumption.Existing resource management and scheduling system is paid close attention to and is reduced enliven machine quantity by improving server resource utilization factor, and the quantity namely run by reducing processor saves the energy, does not pay close attention to the running status of processor.In fact, modern processors can work in different frequencies, and the higher travelling speed of frequency of operation is faster, and corresponding energy consumption is higher, and therefore reasonably set of frequency effectively can control energy consumption.Dynamic frequency voltage-regulating technique (Dynamic Voltage and Frequency Scaling, DVFS) be a kind of technology of the efficient utilization of power of modern processors, DVFS makes processor be in the setting of different frequency and voltage, namely different P-state (Performance states), decided the set of frequency of current processor (CPU) by current task workload, thus reach the object of saving power consumption of processing unit.At present, the research work based on the energy-conservation aspect of DVFS technology launches round mobile computing and embedded system, for alleviating electric energy storage limited in these mobile systems.And in the environment of the contour power consumption of the heart in the data, the energy-saving distribution technology based on dynamic frequency pressure (DVFS) is not yet applied, also without the method for allocating tasks effectively based on DVFS.
Summary of the invention
Of the present inventionly mainly propose a kind of data center's method for scheduling task based on dynamic frequency voltage-regulating technique, the method effectively can reduce the energy consumption of whole cluster.
Data center's method for scheduling task based on dynamic frequency voltage-regulating technique of the present invention, data center comprises front-end server, some cluster servers, and for storing the network stored data storehouse of desired data of executing the task, wherein front-end server includes task receiver module, task scheduling modules, company-data administration module and front end communications module; Every platform cluster server is provided with data obtaining module, the back end communications module be connected with front end communications module; The process of method for scheduling task is:
1) when its required service of user's forward end server request, first on front-end server, its performance model and energy consumption model is set up according to task,
2) every platform cluster server is by the data of data obtaining module Real-time Obtaining self, comprise the setting of current dominant frequency and voltage, the utilization factor of processor, and the company-data administration module passing through the communication module forward end server of front end and rear end upgrades real time data;
3) resource request of task is collected by the task receiver module of front-end server, again in conjunction with Real-Time Cluster data in company-data administration module, with the performance model established in step 1) and energy consumption model, in cluster, the single-point performance bottleneck of each server is as restrictive condition, the total energy consumption of cluster server determines as optimization aim to carry out real-time scheduling to task by the optimal deployment point of task.
In addition, also comprise step 4), under multi-task state, due to concentrating of different task, the real time resources of task is made to require the change changed or server heating factor causes, the dominant frequency voltage of real-time adjustment cluster server, the change comprising processor host frequency, utilization factor general data index caused, obtained by the real-time information acquisition module of cluster server, then upgraded the resources occupation situation of server real time data in cluster server and deployment task by first and second communication module.
Above-mentioned steps 1) in the foundation of performance model be, when each server to different dominant frequency is arranged down in adjustment cluster server, in conjunction with the performance index mainly considered during task run, comprise the execution time, processor occupancy, goes out corresponding performance model according to these data fittings;
The foundation of energy consumption model is also by regulating the dominant frequency of each server in cluster server and voltage to different combinations, obtain the data message of power and corresponding dominant frequency, the theoretical formula of associative processor power consumption carries out the Fitting Analysis of data, sets up corresponding energy consumption model; The theoretical formula of above-mentioned power consumption of processing unit and the energy consumption of processor
with the dominant frequency of processor
and voltage
relation:
It is generally discrete P-state that voltage to frequency due to modern processors is arranged, above-mentioned formula can be reduced to shape as
formula of reduction,
Wherein
with
for constant.The energy consumption that can be gone out by actual measurement
and dominant frequency
carry out data fitting.
The invention has the beneficial effects as follows: combine the scheduler task in traditional server single-point dynamic frequency voltage-regulating technique and whole cluster, make the analysis result that the central management node of whole cluster can draw according to the real time data of each server, online task is made and is more beneficial to energy-conservation task distribution, thus finally can save the energy consumption of whole data center.
Be described in detail below in conjunction with accompanying drawing.
Accompanying drawing explanation
Fig. 1 is scene structure of the present invention,
Fig. 2 is module frame chart of the present invention,
Fig. 3 is operational scheme of the present invention.
Embodiment
First lower two key concepts of the present invention are introduced:
1, based on the performance model of DVFS:dVFS technology changes dominant frequency and the voltage of the central processing unit of server, can have an impact to the computing power of processor.It is generally acknowledged, the dominant frequency reducing processor can make processor execution instruction number per second present the downtrending of equal proportion.Under the pattern of different application, this downtrending can show difference.Usually, in the task of computation-intensive, what we can be desirable thinks that the decline of dominant frequency is equal to the decline of computing power.But be directed to specific tasks, the concrete resource restriction can revealed due to application table is as different in I/O or network.
2, based on the energy consumption model of DVFS:the advantage that DVFS technology reduces after the dominant frequency of processor and voltage is, concerning this processor, its energy consumption can reduce greatly.The energy consumption of processor in theory
with the dominant frequency of processor
and voltage
relation be
It is generally discrete P-state (dominant frequency combinations of voltages) that voltage to frequency due to modern processors is arranged, above-mentioned formula can be reduced to shape as
formula of reduction.Wherein
with
for constant.The energy consumption that can be gone out by actual measurement
and dominant frequency
carry out data fitting.
With the structure of the scene of typical data center Processing tasks as shown in Figure 1, user, to its required service of front-end server request of data center, is processed this task by certain in this front-end server selection cluster server or certain server.The data that required by task is asked store or database information is organized by the network storage.Our method for scheduling task is performed by front-end server, and front-end server is also the central management node of whole cluster.
As shown in Figure 2, the modular design of system is mainly made up of task receiver module, task scheduling modules, company-data administration module, communication module, data obtaining module etc.Wherein, task receiver module, task scheduling modules and cluster management module operate in front-end server and central management node, and data obtaining module is on every platform cluster server.Task receiver module is responsible for the request and the corresponding resource requirement of this request that obtain online task; Task scheduling modules is the execution module of our dispatching method, and it decides the optimal deployment point of task according to the cluster real time data that the task requests of task receiver module submission and cluster management module provide; Company-data administration module is the real time data of being responsible for collecting whole cluster server.Data obtaining module on cluster server is the data of Real-time Obtaining server, is submitted to the cluster management module of front-end server by the communication module of front and back end.
The process flow diagram of system cloud gray model flow process as shown in Figure 3.Concrete enforcement comprises following several aspects: 1, based on the foundation of DVFS performance model and energy consumption model; 2, the real-time information of cluster server obtains; 3, based on the task scheduling of DVFS; 4, the data real-time update of front-end server.
1. based on the foundation of DVFS performance model and energy consumption model
For the task of being submitted to data center, before the dispatching method performing us, we need on corresponding server type, to set up its performance model and energy consumption model according to task.Its concrete method is under the support of DVFS, and regulation server to different dominant frequency is arranged down, the performance index mainly considered during Real-time Obtaining task run, as the execution time, and processor occupancy etc. information.Corresponding performance model is gone out according to these data fittings.
Energy consumption model based on DVFS technology is also by regulating the dominant frequency of processor and voltage to different combinations, real-time detection is carried out by equipment such as power meters, obtain the data message of power and corresponding dominant frequency, the theoretical formula of associative processor power consumption carries out the Fitting Analysis of data, sets up corresponding energy consumption model.
2. the real-time information of cluster server obtains
Pass through data obtaining module, the general data of the every station server Real-time Obtaining self in cluster, the such as setting of current dominant frequency and voltage, utilization factor of processor etc. leading indicator, and carry out renewal real time data by the company-data administration module of communication module forward end server.
3. based on the task scheduling of DVFS
After online task arrives cluster, the resource request of task is collected by the task receiver module of front-end server, again in conjunction with Real-Time Cluster data in cluster management module, with the performance model established in step 1 and energy consumption model, in cluster, the single-point performance bottleneck of each server is as restrictive condition, the total energy consumption of cluster server determines as optimization aim to carry out real-time scheduling to task by the optimal deployment point of each task.
4. the data real-time update of front-end server
The concentrating due to different task occurred in the real time execution of task, the real time resources of task requires the change that change or server heating etc. factor causes, the dominant frequency voltage of real-time adjustment cluster server, the change as the general data such as processor host frequency, utilization factor index caused is obtained by the real-time information acquisition module of cluster server, is then upgraded the resources occupation situation of server real time data in cluster server and deployment task by communication module.So not only can meet the change that the current mission performance caused due to real-time change requires, also can better complete the online task scheduling of next time.
Embody rule approach of the present invention is a lot, and the above is only the preferred embodiment of the present invention, should be understood that; for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvement, these improvement also should be considered as protection scope of the present invention.
Claims (3)
1. the data center's method for scheduling task based on dynamic frequency voltage-regulating technique, data center comprises front-end server, some cluster servers, and for storing the network stored data storehouse of desired data of executing the task, wherein front-end server includes task receiver module, task scheduling modules, company-data administration module and front end communications module; Every platform cluster server is provided with data obtaining module, the back end communications module be connected with front end communications module; It is characterized in that, the process of method for scheduling task is:
1) when its required service of user's forward end server request, first on front-end server, its performance model and energy consumption model is set up according to task,
2) every platform cluster server is by the data of data obtaining module Real-time Obtaining self, comprise the setting of current dominant frequency and voltage, the utilization factor of processor, and the company-data administration module passing through the communication module forward end server of front end and rear end upgrades real time data;
3) resource request of task is collected by the task receiver module of front-end server, again in conjunction with Real-Time Cluster data in company-data administration module, with the performance model established in step 1) and energy consumption model, in cluster, the single-point performance bottleneck of each server is as restrictive condition, the total energy consumption of cluster server determines as optimization aim to carry out real-time scheduling to task by the optimal deployment point of task.
2. the data center's method for scheduling task based on dynamic frequency voltage-regulating technique according to claim 1, characterized by further comprising step 4), under multi-task state, due to concentrating of different task, the real time resources of task is made to require the change changed or server heating factor causes, the dominant frequency voltage of real-time adjustment cluster server, that causes comprises processor host frequency, the change of utilization factor general data index, obtained by the real-time information acquisition module of cluster server, then the resources occupation situation of server real time data in cluster server and deployment task is upgraded by the communication module of front end and rear end.
3. the data center's method for scheduling task based on dynamic frequency voltage-regulating technique according to claim 1 and 2, it is characterized in that in step 1), the foundation of performance model is, when in adjustment cluster server, each server to different dominant frequency is arranged down, in conjunction with the performance index mainly considered during task run, comprise the execution time, processor occupancy, goes out corresponding performance model according to these data fittings;
The foundation of energy consumption model is also by regulating the dominant frequency of each server in cluster server and voltage to different combinations, obtain the data message of power and corresponding dominant frequency, the theoretical formula of associative processor power consumption carries out the Fitting Analysis of data, sets up corresponding energy consumption model; The theoretical formula of above-mentioned power consumption of processing unit and the energy consumption of processor
with the dominant frequency of processor
and voltage
relation:
It is generally discrete P-state that voltage to frequency due to modern processors is arranged, above-mentioned formula can be reduced to shape as
formula of reduction,
Wherein
with
for the energy consumption that constant can be gone out by actual measurement
and dominant frequency
carry out data fitting.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410526996.XA CN104298536A (en) | 2014-10-09 | 2014-10-09 | Dynamic frequency modulation and pressure adjustment technology based data center energy-saving dispatching method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410526996.XA CN104298536A (en) | 2014-10-09 | 2014-10-09 | Dynamic frequency modulation and pressure adjustment technology based data center energy-saving dispatching method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104298536A true CN104298536A (en) | 2015-01-21 |
Family
ID=52318276
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410526996.XA Pending CN104298536A (en) | 2014-10-09 | 2014-10-09 | Dynamic frequency modulation and pressure adjustment technology based data center energy-saving dispatching method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104298536A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104808770A (en) * | 2015-04-23 | 2015-07-29 | 南京大学 | Data center energy consumption management method and system based on dynamic frequency modulation |
| CN107197323A (en) * | 2017-05-08 | 2017-09-22 | 上海工程技术大学 | A kind of network video-on-demand server and its application based on DVFS |
| CN114546096A (en) * | 2022-02-10 | 2022-05-27 | Oppo广东移动通信有限公司 | Base band chip, task scheduling method and terminal equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110173329A1 (en) * | 2010-01-13 | 2011-07-14 | Nec Laboratories America, Inc. | Methods and Apparatus for Coordinated Energy Management in Virtualized Data Centers |
| CN102650957A (en) * | 2012-04-09 | 2012-08-29 | 武汉理工大学 | Self-adaptive energy-saving dispatching method in isomorphic cluster system based on dynamic voltage regulation technology |
| CN103076870A (en) * | 2013-01-08 | 2013-05-01 | 北京邮电大学 | Application fusing scheduling and resource dynamic configuring method of energy consumption drive in data center |
| CN103235640A (en) * | 2013-01-08 | 2013-08-07 | 北京邮电大学 | DVFS-based energy-saving dispatching method for large-scale parallel tasks |
| CN103279184A (en) * | 2013-06-05 | 2013-09-04 | 浪潮电子信息产业股份有限公司 | Cloud server energy-saving system based on data mining |
| CN103902379A (en) * | 2012-12-25 | 2014-07-02 | 中国移动通信集团公司 | Task scheduling method and device and server cluster |
-
2014
- 2014-10-09 CN CN201410526996.XA patent/CN104298536A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110173329A1 (en) * | 2010-01-13 | 2011-07-14 | Nec Laboratories America, Inc. | Methods and Apparatus for Coordinated Energy Management in Virtualized Data Centers |
| CN102650957A (en) * | 2012-04-09 | 2012-08-29 | 武汉理工大学 | Self-adaptive energy-saving dispatching method in isomorphic cluster system based on dynamic voltage regulation technology |
| CN103902379A (en) * | 2012-12-25 | 2014-07-02 | 中国移动通信集团公司 | Task scheduling method and device and server cluster |
| CN103076870A (en) * | 2013-01-08 | 2013-05-01 | 北京邮电大学 | Application fusing scheduling and resource dynamic configuring method of energy consumption drive in data center |
| CN103235640A (en) * | 2013-01-08 | 2013-08-07 | 北京邮电大学 | DVFS-based energy-saving dispatching method for large-scale parallel tasks |
| CN103279184A (en) * | 2013-06-05 | 2013-09-04 | 浪潮电子信息产业股份有限公司 | Cloud server energy-saving system based on data mining |
Non-Patent Citations (1)
| Title |
|---|
| 马艳: "基于能耗和成本的资源管理与调度策略研究", 《万方数据企业知识服务平台》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104808770A (en) * | 2015-04-23 | 2015-07-29 | 南京大学 | Data center energy consumption management method and system based on dynamic frequency modulation |
| CN104808770B (en) * | 2015-04-23 | 2017-04-12 | 南京大学 | Data center energy consumption management method and system based on dynamic frequency modulation |
| CN107197323A (en) * | 2017-05-08 | 2017-09-22 | 上海工程技术大学 | A kind of network video-on-demand server and its application based on DVFS |
| CN114546096A (en) * | 2022-02-10 | 2022-05-27 | Oppo广东移动通信有限公司 | Base band chip, task scheduling method and terminal equipment |
| CN114546096B (en) * | 2022-02-10 | 2024-02-06 | Oppo广东移动通信有限公司 | Baseband chip, task scheduling method and terminal equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104991830B (en) | YARN resource allocations and energy-saving scheduling method and system based on service-level agreement | |
| CN103345298B (en) | Method of data center energy saving system based on virtual IT resource distribution technology | |
| CN103384272B (en) | A kind of cloud service distributive data center system and load dispatching method thereof | |
| CN104142860A (en) | Resource adjusting method and device of application service system | |
| CN107168770B (en) | Low-energy-consumption cloud data center workflow scheduling and resource supply method | |
| CN102508714A (en) | Green-computer-based virtual machine scheduling method for cloud computing | |
| Mao et al. | A multi-resource task scheduling algorithm for energy-performance trade-offs in green clouds | |
| CN103297505B (en) | The On-Line Control Method of data center's multiple-energy-source and system under dynamic cloud service request | |
| CN104021040A (en) | Cloud computing associated task scheduling method and device based on time constraint | |
| CN103645795A (en) | Cloud computing data center energy saving method based on ANN (artificial neural network) | |
| CN107423133B (en) | Data network load distribution method among data centers for reducing power grid loss | |
| CN114710563A (en) | Cluster energy saving method and device | |
| CN104298536A (en) | Dynamic frequency modulation and pressure adjustment technology based data center energy-saving dispatching method | |
| CN102855157A (en) | Method for comprehensively scheduling load of servers | |
| CN103617090A (en) | Energy saving method based on distributed management | |
| Terzopoulos et al. | Bag-of-task scheduling on power-aware clusters using a dvfs-based mechanism | |
| CN108429784B (en) | Energy efficiency priority cloud resource allocation and scheduling method | |
| Peng et al. | Energy-efficient management of data centers using a renewable-aware scheduler | |
| CN103176850A (en) | Electric system network cluster task allocation method based on load balancing | |
| Gu et al. | Lowering down the cost for green cloud data centers by using ESDs and energy trading | |
| CN104796673A (en) | Energy consumption optimization-oriented cloud video monitoring system task access method | |
| Yuan et al. | An Online Energy Saving Resource Optimization Methodology for Data Center. | |
| CN103246497B (en) | A kind of real time data method for parallel processing based on Data Placement | |
| CN102521715B (en) | A kind of method and system controlling application system Resourse Distribute | |
| CN103631659A (en) | Schedule optimization method for communication energy consumption in on-chip network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20150121 |