CN102208986B - Cluster power consumption distribution and control method - Google Patents
Cluster power consumption distribution and control method Download PDFInfo
- Publication number
- CN102208986B CN102208986B CN201110068692.XA CN201110068692A CN102208986B CN 102208986 B CN102208986 B CN 102208986B CN 201110068692 A CN201110068692 A CN 201110068692A CN 102208986 B CN102208986 B CN 102208986B
- Authority
- CN
- China
- Prior art keywords
- power consumption
- cluster
- upper limit
- working group
- 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
Links
Images
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a cluster power consumption distribution and control method which comprises the following steps of determining and dividing the power consumption of clusters, distributing the cluster power consumption and finally adjusting the power consumption. The power consumption of a cluster server can be effectively controlled by using the distribution and control method so as to achieve the aim of effectively controlling and managing the cluster power consumption; the power consumption of the cluster sever is effectively saved without influencing the application performance, thereby brining convenience for management of users.
Description
Technical field
The present invention relates to cluster power consumption management domain, be specifically related to a kind of cluster power consumption and distribution and control method.
Background technology
Along with the scientific and technological progress of data center products, the renovation day by day of design concept, the power density of data center strengthens year by year, and more powerful data center's disposal ability is the target of pursuing always.
But the today of growing with each passing day at energy consumption, the another side of processor and Development of storage technology is to consume more resource.Along with the continuous increase of server density, power demands is also in corresponding increase, and produced thus more heat.In fact, data center's power supply and cooling under-supply problem, become the yoke that many companies realize its IT target.
" according to the investigation of IDC, show, concerning the computer hardware cost of each dollar, about 50 cents is to spend in electricity usage, and this numeral is estimated also will increase by 54% in 4 years from now on." " before 18 years, IBM adopts environmental protection and energy-conservation solution at Osaka, Japan, to build the huge data center of 28000 square metres at Osaka, Japan, and over 18 years, this data center has saved the energy consumption cost that surpasses 50%.”
Meanwhile, along with the application of blade server, cluster server is more and more universal, the power supply of high-density data center and refrigeration problem become more outstanding.
The various device at available data center manages separately mostly, for example coil battle array, server, UPS, air-conditioning etc., the author thinks that its management mode will develop to both direction, will change and the fluctuation of load be unified task scheduling and be regulated various device according to external environment condition on the one hand, realizes the optimized scheduling in broad sense more; On the other hand, the granularity of management can be thinner, for example, the task of every node is adjusted, and the dominant frequency of each CPU core is adjusted separately, carries out subregion cooling etc.
The application of server operation generally can frequently not change, and the real time data of load and power consumption can be preserved to history of forming data, and feature and the development trend that supervisory control system accordingly can automatic analysis load also made corresponding adjustment.Existing automatic control technology should guarantee in principle that power consumption is divided and match the quick response of the fluctuation of load, avoid again adjusting too frequently the waste that causes electric power resource.Perfect power consumption allocation strategy can make keeper that energy is concentrated in the analysis of load properties, and need not expend great effort for details such as its concrete numerical value and adjustment times.
Summary of the invention
The present invention considers from angle and two aspects of server component of service application, the power consumption that realization takes device to cluster is controlled, the control strategy of power consumption is more suitable in actual application, solved in real server application, only pay close attention in parts the blindness that only relies on judgement cpu instruction to carry out power adjustment, fundamentally solved the user in application to the demand of computer capacity and the object of consumption reduction in time when idle.
Distribute and a control method, step is as follows:
A, the setting cluster power consumption upper limit, judge that whether this upper limit is reasonable, if rationally use, if unreasonable or not setting of keeper, according to the Rule cluster power consumption upper limit;
B, for each working group specifies a group policy, comprise maximum performance, minimum power consumption, adjusts automatically; According to monitor data, be that working group distributes power consumption;
C, each node obtain the power consumption upper limit separately according to cluster power consumption control strategy, and by adjusting cpu power, the power supply status of kernel switch and parts regulates power consumption.
A preferred technical solution of the present invention is: in steps A, by rule, obtaining cluster power consumption upper limit process is, according to historical power consumption load value, calculate and obtain the cluster power consumption upper limit, if without historical basis use Cluster Theory maximum power dissipation 80% as the cluster power consumption upper limit.
Another kind of optimal technical scheme of the present invention is: working group's power consumption assigning process is:
B1, from historical data, obtain the peak value of each working group's power consumption in a upper adjustment cycle;
If the power peak sum of each working group of B2 is less than the power consumption upper limit of cluster;
B3, according to the power consumption peaks of each working group, carry out static first sub-distribution;
If the power consumption sum of each working group of B4 is greater than or equal to the power consumption upper limit of cluster;
B5, according to average power consumption, carry out just sub-distribution;
B6, according to working group's priority, carry out sub-distribution again.
Another optimal technical scheme of the present invention is: working group's interior nodes power consumption assigning process is:
C1, obtain the peak value that each node power consumes in an adjustment cycle;
If the power peak sum of each node of C2 is less than the power consumption upper limit of cluster, according to the power consumption peaks of each node, carry out static first sub-distribution, in whole adjustment cycle according to the dynamic power consumption request of actual node loading condition responsive node;
If the power consumption sum of each node of C3 is greater than or equal to the power consumption upper limit of working group, according to average power consumption, carry out just sub-distribution, according to node priority, carry out sub-distribution again.
Power consumption that can efficient dominating set group server by method of the present invention, has reached that cluster power consumption is controlled and the effective object of management, and the power consumption of effectively having saved cluster server in the situation that not affecting application performance, has facilitated user's management
Accompanying drawing explanation
Fig. 1 cluster power consumption control flow chart
Fig. 2 working group power consumption control flow chart
Fig. 3 is by request dynamic allocation flow figure
Fig. 4 presses average power consumption allocation flow figure
Fig. 5 is dynamic assignment flow chart according to priority
Embodiment
This method is regarded power consumption as a kind of resource of server, becomes quiescent dissipation and dynamic power consumption to carry out unified management, distribution and scheduling power consumption resource division.Quiescent dissipation is for meeting the fixedly needs of user job, and dynamic power consumption is for meeting the emergency requirement of different nodes.The distribution of power consumption according to the specified parameter of user, is user's application load historical data according on the one hand on the other hand.
1, cluster power consumption determines and divides
First obtain the cluster power consumption upper limit, by cluster administrator, specified, first determine that whether the power consumption upper limit of appointment is reasonable, judge that it is whether between the maximum and minimum value in cluster power consumption, cluster power consumption maximum and minimum value are recorded by actual.If cluster power consumption higher limit rationally, use it, if unreasonable or not appointment of user makes to obtain in the following method the cluster power consumption upper limit.
The automatic acquisition of the cluster power consumption upper limit
If there is no historical data, use the cluster power consumption upper limit peaked 80% as the cluster power consumption upper limit, otherwise calculate and obtain the cluster power consumption upper limit according to historical power consumption load value.Projectional technique is as follows:
The power consumption upper limit of the historical data analysis cluster by user:
User's practical application operation long period of time T;
T is equally divided into n equal portions, is respectively t1, t2 ..., tn;
Record clustering is at each adjustment cycle t1, t2 ..., the peak power w1 of tn, w2 ..., wn(makes W=(w1+w2+ ... + wn)/n);
Record clustering is at each adjustment cycle t1, t2 ..., the average power consumption △ w1 of tn, △ w2 ..., △ wn(makes △ W=(△ w1+ △ w2+ ... + △ wn)/n);
The final power consumption upper limit: wlimit=△ W+ (W-△ W)/2;
2, the distribution of power consumption
First for each working group specifies power consumption strategies in: maximum performance, minimum power consumption or automatically adjustment.The power consumption of working group distributes by the priority size of working group successively to distribute.When wherein power consumption strategies is maximum performance, working group's priority is maximum, and minimum power consumption is taken second place, and self-adjusting priority is carried out according to tactful designated value, if do not have to specify, by the sequencing of group, distributes.
Secondly working group's power consumption is distributed according to monitor data (peak power of a upper adjustment cycle of each group and average power consumption etc.).
The total power consumption upper limit of cluster is W;
In cluster, there is n working group;
The polling cycle of management work group is t;
The power consumption adjustment cycle of management work group is T; (integral multiple that T is t)
The power consumption peaks of a upper adjustment cycle of each working group is respectively: w1 ', and w2 ' ..., wn ';
The next adjustment cycle of each working group is respectively the quiescent dissipation upper limit being set up: w1, and w2 ..., wn;
Make W '=w1 '+w2 '+... + wn ', △ W=W-W ';
W* ' represents to press after just sub-distribution of average power consumption, the power consumption upper limit sum that each working group distributes.
Its allocation flow is described below:
1) from historical data, obtain the peak value of each working group's power consumption in a upper adjustment cycle.
2) if the power peak sum of each working group is less than the power consumption upper limit of cluster
3) according to the power consumption peaks of each working group, carry out static first sub-distribution;
4) if the power consumption sum of each working group is greater than or equal to the power consumption upper limit of cluster
5) according to average power consumption, carry out just sub-distribution (from the minimum working group of priority);
6) according to working group's priority, carry out sub-distribution again.
The allocation flow of working group's interior nodes power consumption is as follows:
The power consumption upper limit of working group is W, total n node in working group;
The power consumption peaks of a upper adjustment cycle of each node is respectively: w
1', w
2' ..., w
n';
The next adjustment cycle of each node is respectively the quiescent dissipation being set up: w
1, w
2..., w
n;
Make W '=w
1'+w
2'+... + w
n', △ W=W-W ';
W* ' represents to press after just sub-distribution of average power consumption, the power consumption upper limit sum that each node distributes.
Its allocation flow is as follows:
1) obtain the peak value that in an adjustment cycle, each node power consumes.
2) if the power peak sum of each node is less than the power consumption upper limit of working group
A) according to the power consumption peaks of each node, carry out static first sub-distribution;
B) in whole adjustment cycle according to the dynamic power consumption request of actual node loading condition responsive node.
3) if the power consumption sum of each node is greater than or equal to the power consumption upper limit of working group
A) according to average power consumption, carry out just sub-distribution (from the minimum node of priority);
B) according to node priority, carry out sub-distribution again.
3, power consumption adjustment
This method by operating system, adjust the frequency of CPU, the switch of kernel and parts (for example hard disk) thus the power consumption of power supply status knot modification.When cluster total power consumption is controlled, each node obtains the power consumption upper limit separately according to the power consumption control strategy of cluster, and the power consumption by various control interface knot modifications is to meet the demands, simultaneously, by the monitoring to node load and power consumption, can be dynamically to management node application power consumption.
Claims (3)
1. cluster power consumption distributes and a control method, it is characterized in that: step is as follows:
A, the setting cluster power consumption upper limit, judge that whether this upper limit is reasonable, if rationally use, if unreasonable or not setting of keeper, according to the Rule cluster power consumption upper limit;
B, for each working group specifies a group policy, comprise maximum performance, minimum power consumption, adjusts automatically; According to monitor data, be that working group distributes power consumption;
C, each node obtain the power consumption upper limit separately according to cluster power consumption control strategy, and by adjusting cpu power, the power supply status of kernel switch and parts regulates power consumption;
Working group's power consumption assigning process is:
B1, from historical data, obtain the peak value of each working group's power consumption in a upper adjustment cycle;
If the power peak sum of each working group of B2 is less than the power consumption upper limit of cluster;
B3, according to the power consumption peaks of each working group, carry out static first sub-distribution;
If the power consumption sum of each working group of B4 is greater than or equal to the power consumption upper limit of cluster;
B5, according to average power consumption, carry out just sub-distribution;
B6, according to working group's priority, carry out sub-distribution again.
2. a kind of cluster power consumption distributes and control method as claimed in claim 1, it is characterized in that: in steps A, by rule, obtaining cluster power consumption upper limit process is, according to historical power consumption load value, calculate and obtain the cluster power consumption upper limit, if without historical basis use Cluster Theory maximum power dissipation 80% as the cluster power consumption upper limit.
3. a kind of cluster power consumption distributes and control method as claimed in claim 1, it is characterized in that: working group's interior nodes power consumption assigning process is:
C1, obtain the peak value that each node power consumes in an adjustment cycle;
If the power peak sum of each node of C2 is less than the power consumption upper limit of working group, according to the power consumption peaks of each node, carry out static first sub-distribution, in whole adjustment cycle according to the dynamic power consumption request of actual node loading condition responsive node;
If the power consumption sum of each node of C3 is greater than or equal to the power consumption upper limit of working group, according to average power consumption, carry out just sub-distribution, according to node priority, carry out sub-distribution again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110068692.XA CN102208986B (en) | 2011-03-22 | 2011-03-22 | Cluster power consumption distribution and control method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110068692.XA CN102208986B (en) | 2011-03-22 | 2011-03-22 | Cluster power consumption distribution and control method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102208986A CN102208986A (en) | 2011-10-05 |
| CN102208986B true CN102208986B (en) | 2014-01-15 |
Family
ID=44697659
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110068692.XA Active CN102208986B (en) | 2011-03-22 | 2011-03-22 | Cluster power consumption distribution and control method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102208986B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102497275A (en) * | 2011-12-02 | 2012-06-13 | 曙光信息产业(北京)有限公司 | Cluster power consumption distribution method based on power consumption pool |
| US9158351B2 (en) * | 2012-03-29 | 2015-10-13 | Intel Corporation | Dynamic power limit sharing in a platform |
| CN104572272B (en) * | 2013-10-12 | 2018-02-09 | 杭州华为数字技术有限公司 | A kind of method for scheduling task, apparatus and system |
| CN104679215B (en) * | 2013-11-28 | 2017-10-17 | 杭州华为数字技术有限公司 | Energy consumption binds method of adjustment and device |
| CN107329811A (en) * | 2017-06-09 | 2017-11-07 | 北京云集智造科技有限公司 | A kind of power consumption of data center adjusting method and device |
| CN110022246A (en) * | 2019-04-15 | 2019-07-16 | 苏州浪潮智能科技有限公司 | Distributed type assemblies equipment power dissipation monitoring method, device, system and associated component |
| CN110399216B (en) | 2019-06-27 | 2021-10-15 | 苏州浪潮智能科技有限公司 | Method, system and device for distributing power consumption of whole machine box and readable storage medium |
| CN114326609A (en) * | 2021-12-02 | 2022-04-12 | 成都工业学院 | Multi-energy-flow SCADA energy real-time monitoring and collecting system based on Internet of things technology |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100550680C (en) * | 2006-01-06 | 2009-10-14 | 华为技术有限公司 | A kind of method for controlling forward shared channel total power of cluster system |
| US8082454B2 (en) * | 2007-11-07 | 2011-12-20 | International Business Machines Corporation | Managing power consumption based on historical average |
| US8069359B2 (en) * | 2007-12-28 | 2011-11-29 | Intel Corporation | System and method to establish and dynamically control energy consumption in large-scale datacenters or IT infrastructures |
-
2011
- 2011-03-22 CN CN201110068692.XA patent/CN102208986B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN102208986A (en) | 2011-10-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102208986B (en) | Cluster power consumption distribution and control method | |
| Zhu et al. | A three-dimensional virtual resource scheduling method for energy saving in cloud computing | |
| Femal et al. | Boosting data center performance through non-uniform power allocation | |
| EP2399183B1 (en) | Energy-aware server management | |
| US8301742B2 (en) | Systems and methods for coordinated management of power usage and runtime performance in performance-managed computing environments | |
| Zhang et al. | Dynamic energy-aware capacity provisioning for cloud computing environments | |
| US8583945B2 (en) | Minimizing power consumption in computers | |
| US9128766B1 (en) | Computer workload redistribution schedule | |
| US20130104136A1 (en) | Optimizing energy use in a data center by workload scheduling and management | |
| US20090235097A1 (en) | Data Center Power Management | |
| US20090187776A1 (en) | Server power consumption controller, and method and computer program for controlling server power consumption | |
| CA2716646C (en) | Coordinated approach between middleware application and sub-systems | |
| CN103856512A (en) | Cloud computing management server, working host, idle host and resource scheduling method | |
| EP2524277A2 (en) | Methods and apparatus for coordinated energy management in virtualized data centers | |
| CN102508714A (en) | Green-computer-based virtual machine scheduling method for cloud computing | |
| CN102759984A (en) | Power supply and performance management system for virtualization server cluster | |
| Adhikari et al. | Double threshold energy aware load balancing in cloud computing | |
| CN103179048A (en) | Method and system for changing main machine quality of service (QoS) strategies of cloud data center | |
| Guo et al. | JET: Electricity cost-aware dynamic workload management in geographically distributed datacenters | |
| WO2011076486A1 (en) | A method and system for dynamic workload allocation in a computing center which optimizes the overall energy consumption | |
| Song et al. | Server consolidation energy-saving algorithm based on resource reservation and resource allocation strategy | |
| CN109857574A (en) | Under a kind of low energy consumption cluster environment can overloaded load perception Service Promotion method | |
| CN102043676B (en) | Visualized data centre dispatching method and system | |
| Li et al. | CSL-driven and energy-efficient resource scheduling in cloud data center | |
| Kaur et al. | Load balancing techniques of cloud computing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C41 | Transfer of patent application or patent right or utility model | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20170124 Address after: 211106 Jiangning economic and Technological Development Zone, Jiangsu, general Avenue, No. 37, No. Patentee after: JIANGSU DAWN INFORMATION TECHNOLOGY CO., LTD. Address before: 100084 Beijing Haidian District City Mill Street No. 64 Patentee before: Dawning Information Industry (Beijing) Co., Ltd. |