CN110995863B - Data center load distribution method and system based on load demand characteristics - Google Patents
Data center load distribution method and system based on load demand characteristics Download PDFInfo
- Publication number
- CN110995863B CN110995863B CN201911317923.9A CN201911317923A CN110995863B CN 110995863 B CN110995863 B CN 110995863B CN 201911317923 A CN201911317923 A CN 201911317923A CN 110995863 B CN110995863 B CN 110995863B
- Authority
- CN
- China
- Prior art keywords
- resource
- interference
- load
- module
- matrix
- 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
Classifications
-
- 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
- H04L67/10—Protocols in which an application is distributed across nodes in the network
- H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
- H04L67/1004—Server selection for load balancing
- H04L67/1008—Server selection for load balancing based on parameters of servers, e.g. available memory or workload
-
- 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
- H04L67/10—Protocols in which an application is distributed across nodes in the network
- H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
- H04L67/1004—Server selection for load balancing
- H04L67/1023—Server selection for load balancing based on a hash applied to IP addresses or costs
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Debugging And Monitoring (AREA)
Abstract
A data center load distribution method based on load demand characteristics is characterized in that interference classification is carried out on incoming loads through time sequence resource demand characteristics of the loads which are measured in advance, corresponding interference degree indexes are obtained according to resource use conditions of different computing nodes, the loads are distributed to the computing nodes enabling the sum of the interference degree indexes to be minimum through scheduling nodes, and therefore scheduling of interference among the loads is achieved to be minimum. The invention carries out the scheduling of the load by a mode of minimizing the interference between the loads. The method avoids the whole resource idleness of the server caused by the overload of the single resource, and the resource waste and the efficiency reduction caused by the idleness. And simultaneously, the service quality reduction caused by single resource overload is reduced. By calculating the interference generated by the load and avoiding the interference scheduling mode as much as possible, the utilization rates of different resource types are improved in a balanced manner, and the overall resource utilization rate of the data center is improved.
Description
Technical Field
The invention relates to a technology in the field of computers, in particular to a data center load distribution method and system based on load demand characteristics.
Background
In a typical data center environment, servers may distribute workloads to computing nodes in a network to perform related services. The administrator of the data center needs to utilize the resources in each computing node to the maximum extent, so that the utilization rate of the resources is maximized. In practice, however, when running a load that requires the same computational resources, resource contention occurs, resulting in a severe degradation of the performance of the program, a phenomenon also referred to as interference. Interference occurs because different loads at the same time request a large amount of the same resources, exceeding the supply of computers, resulting in resource contention. For a particular resource, such as a processor, memory, network bandwidth, etc., if the server allocates a number of workloads that are heavily dependent on the resource, the workloads available in the compute nodes may be overloaded, creating interference. As a result, the overall performance of the computing node is negatively impacted even if other resources in the computing node are nearly idle. In order to alleviate interference and meet the requirements of users in the prior art, a data center usually maintains a low resource utilization rate, which results in resource waste and increased calculation cost.
Disclosure of Invention
The invention provides a data center load distribution method and system based on load demand characteristics, aiming at the defects in the prior art, the load requiring different resources is distributed to the same computer, so that the interference is avoided, the program performance is improved under the condition of ensuring the same resource utilization rate, or the resource utilization rate is improved under the condition of maintaining the program performance without reduction.
The invention is realized by the following technical scheme:
the invention relates to a data center load distribution method based on load demand characteristics, which is characterized in that the interference classification is carried out on the coming loads through the time sequence resource demand characteristics of the measured loads in advance, the corresponding interference degree indexes are obtained according to the resource use conditions of different computing nodes, and the loads are distributed to the computing nodes with the minimum sum of the interference degree indexes by a scheduling node, so that the scheduling of the interference among the loads is minimized.
The time sequence resource demand characteristics are represented by a demand characteristic matrix, the demand characteristic matrix is a 4 x 3 matrix, and the time sequence resource demand characteristics specifically comprise the following components: dividing the life cycle of the load into three time intervals according to the time sequence, namely a loading time interval, an execution time interval and an ending time interval; the demand of the load for the resource is divided into four dimensions, namely: computing resources, memory resources, hard disk resources, and network resources.
And the values of the four dimensions are recorded in the matrix according to the respective demand of the load.
The interference classification is as follows: the method comprises the following steps of recording the resource use condition of future loads to obtain a demand characteristic matrix, namely the interference classification when the loads really arrive, wherein the specific operation process comprises the following steps: the load is independently operated on one physical server, other interferences are eliminated, an operating system instruction is called to read the resource use conditions of the four dimensions of the physical server, and a demand characteristic matrix of the type of load is recorded.
The future load refers to: the data center is working with all possible assigned workloads.
The resource use condition is represented by a maximum resource vector, an allocable resource vector and a used resource matrix, and specifically, the computing node is recorded in the four dimensions according to the actual number of physical computing resources to obtain a four-dimensional maximum resource vector to represent the maximum resource of the computing node; a 4 x N used resource matrix, wherein: n represents the number of working time periods, all elements in the matrix are initialized to zero, and the number of the used resources is obtained by adding the corresponding number of the resources used by the scheduled load, wherein the number of the used resources of the computing node is represented by the used resource matrix; the allocable resource vector at a certain moment is obtained by subtracting the used resource vector at a certain moment from the maximum resource vector.
The allocable resource vector at a certain time is: a column of four-dimensional vectors representing a time instant in the 4 x N used resource matrix.
The interference degree index is as follows: calculating the allocable resource vectors of the node at the time of arrival and two times after the time of arrival according to the load by the method, forming a 4 x 3 matrix by the three allocable resource vectors, and subtracting the matrix by using the demand characteristic matrix of the load, wherein an element larger than 0 represents the interference generated by the resource of the dimension when the load is scheduled on the physical server, and summing the values larger than 0 in the matrix to obtain the interference degree index. This is performed for all the compute nodes, the compute node with the smallest interference level index is selected, and the load is scheduled on it.
The arrival time of the load is as follows: the moment the load is distributed to the data center at which the corresponding computing resources are to be acquired and begin working.
Technical effects
The invention integrally solves the problems of resource idling and performance reduction caused by interference in the data center.
Compared with the prior art, the resource demand of the load is divided into three sections according to the life cycle of the load, and a 4 x 3 matrix is used for representing the resource demand of the life cycle of the load. And the interference calculation of the calculation node is carried out according to the interference calculation, and the load distribution is carried out according to the interference degree index.
The invention carries out the scheduling of the load by a mode of minimizing the interference between the loads. The method avoids the whole resource idleness of the server caused by the overload of the single resource, and the resource waste and the efficiency reduction caused by the idleness. And simultaneously, the service quality reduction caused by single resource overload is reduced. By calculating the interference generated by the load and avoiding the interference scheduling mode as much as possible, the utilization rates of different resource types are improved in a balanced manner, the condition that one resource is already distributed but other resources have a large amount of surplus can not occur, and the overall resource utilization rate of the data center is improved.
Drawings
FIG. 1 is a schematic diagram of a load distribution system of the present invention;
FIG. 2 is a timing diagram illustrating a load distribution method according to the present invention;
fig. 3 and 4 are schematic diagrams of embodiments of the present invention.
Detailed Description
As shown in fig. 1, the present embodiment relates to a data center load distribution system based on load demand characteristics, including: the system comprises a load demand characteristic analysis module, an interference classification module, a computing node resource analysis module, an interference calculation module, a scheduling module and resource monitoring modules located in all nodes, wherein the load demand characteristic analysis module, the interference classification module, the computing node resource analysis module, the interference calculation module and the scheduling module are located in a data center, and the resource monitoring modules are located in all nodes: the load demand characteristic analysis module is connected with the interference classification module and transmits the user load information acquired in real time to the interference classification module after analyzing the acquired demand characteristic information; the interference classification module reads the analysis result of the load demand characteristic analysis module and records the resource demand of each load; the computing node resource analysis module is connected with the resource monitoring module and records the resource use condition and the resource residual condition of the physical server in real time; the interference calculation module is respectively connected with the interference classification module and the calculation node resource analysis module and carries out interference degree index calculation on results obtained by the interference classification module and the calculation node resource analysis module; the scheduling module is connected with the interference calculation module and receives the calculation result, and the load is distributed to the physical server with the minimum interference degree index according to the calculation result.
The load demand characteristic analysis module divides the load into a loading time interval, an execution time interval and an ending time interval according to the life cycle, and each time interval corresponds to four dimensions of computing resources, memory resources, hard disk resources and network resources to form a 4 x 3 demand characteristic matrix.
And the interference classification module reads a demand characteristic matrix corresponding to the load demand characteristic analysis module according to the load.
The computing node resource analysis module commonly represents the resource use condition of each physical server by using a vector and a matrix, and specifically obtains a four-dimensional maximum resource vector to represent the maximum resource of the computing node according to the computing resource, namely the number of physical computing resources with four dimensions of processor frequency, memory size, hard disk speed and network bandwidth; representing the used resource matrix by a 4 x N matrix, wherein: n represents the number of working time periods, the used resource use condition is obtained by monitoring the operating system parameters of the physical server, the future resource use condition is obtained according to the used resources plus the corresponding resources used by the scheduled load, and the allocable resource vector at any moment is obtained by subtracting the corresponding column vector representing the moment in the used resource matrix from the maximum resource vector.
The interference calculation module calculates the allocable resource vectors of the node at the time of arrival and two subsequent times by the method according to the time of arrival of the load, forms a 4 x 3 matrix by the three allocable resource vectors, subtracts the matrix by using the demand characteristic matrix of the load, wherein an element larger than 0 represents the size of interference generated by the resource of the dimension when the load is scheduled on the physical server, and sums the values larger than 0 in the matrix to obtain the interference degree index.
And the scheduling module performs scheduling distribution according to the interference degree index of the load in each physical server, and optimizes the load in the scheduling process by using a greedy algorithm.
The greedy algorithm is as follows: and in each scheduling, the subsequent load is not considered, the current load is scheduled to the server with the minimum generated interference degree index, and when the interference degree indexes of a plurality of physical servers to the load to be scheduled are the same, one physical server is randomly selected for scheduling. This is performed for each incoming load.
As shown in fig. 2, the physical server of the data center obtains the resource usage after analyzing the resource usage: consisting of a maximum resource vector and a used resource matrix. And after the loads to be distributed are subjected to interference classification, a demand characteristic matrix is obtained, the interference degree index of each server is calculated according to the interference degree index calculation method, and then the load is dispatched to the server with the minimum interference degree index by the dispatcher.
Compared with the prior art, the method considers the resource demand change of the load, records according to different stages of the life cycle of the load, expresses the resource demand of the load by a matrix, and calculates the interference according to the matrix.
Through concrete actual experiment, at the concrete environment of data center simulation system, set up 10 ~ 30 physical servers, under the setting of 100 ~ 300 loads of distribution, the experimental data that can obtain are: as shown in fig. 3 and 4, the four technical solutions in the drawings are respectively expressed as (a) Best-Fit scheduling (BFS) preferentially schedules the load on the physical server closest to full load; (b) first time adaptive scheduling (FFS) is used for preferentially scheduling the load on a First physical server capable of meeting the load resource requirement, and if all the servers cannot completely meet the load requirement, random scheduling is performed for one time; (c) random Scheduling (RS) performs completely Random scheduling on all loads; (d) the interference minimization scheduling (MIS) is a scheduling algorithm for minimizing interference between loads, which is proposed in this example. The rate of Service Level Agreement (SLA) violation in fig. 4 represents the proportion of the load that is not completed within a specified time according to the customer requirements, which in practice is most likely caused by a performance degradation due to interference. According to experimental data, the load distribution method for minimizing the interference among the loads can effectively reduce the interference among the loads of the data center, and meanwhile, the SLA violation rate of the data center is effectively reduced.
Compared with the prior art, the method can reduce the inter-load interference by 34% on average and reduce the SLA violation rate by 17% on average.
The foregoing embodiments may be modified in many different ways by those skilled in the art without departing from the spirit and scope of the invention, which is defined by the appended claims and all changes that come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
Claims (7)
1. A data center load distribution method based on load demand characteristics is characterized in that interference classification is carried out on future loads through measuring time sequence resource demand characteristics of the loads in advance, corresponding interference degree indexes are obtained according to resource use conditions of different computing nodes, and the loads are distributed to the computing nodes enabling the sum of the interference degree indexes to be minimum through scheduling nodes, so that scheduling of interference among the loads is minimized;
the time sequence resource demand characteristics are represented by a demand characteristic matrix, the demand characteristic matrix is a 4 x 3 matrix, and the time sequence resource demand characteristics specifically comprise the following components: dividing the life cycle of the load into three time intervals according to the time sequence, namely a loading time interval, an execution time interval and an ending time interval; the demand of the load for the resource is divided into four dimensions, namely: computing resources, memory resources, hard disk resources and network resources;
the interference classification is as follows: the method comprises the following steps of recording the resource use condition of future loads to obtain a demand characteristic matrix, namely the interference classification when the loads really arrive, specifically comprising the following steps: the resource use condition of each physical server is represented by a vector and a matrix together, and specifically, a four-dimensional maximum resource vector is obtained by a computing node according to the number of computing resources, namely the physical computing resources with four dimensions of processor frequency, memory size, hard disk speed and network bandwidth, so as to represent the maximum resource of the computing node; representing the used resource matrix by a 4 x N matrix, wherein: n represents the number of working time periods, the used resource use condition is obtained by monitoring the operating system parameters of the physical server, the future resource use condition is obtained according to the used resources plus the corresponding resources used by the scheduled load, and the allocable resource vector at any moment is obtained by subtracting the corresponding column vector representing the moment in the used resource matrix from the maximum resource vector.
2. A system for implementing the method of any preceding claim, comprising: the system comprises a load demand characteristic analysis module, an interference classification module, a computing node resource analysis module, an interference calculation module, a scheduling module and resource monitoring modules located in all nodes, wherein the load demand characteristic analysis module, the interference classification module, the computing node resource analysis module, the interference calculation module and the scheduling module are located in a data center, and the resource monitoring modules are located in all nodes: the load demand characteristic analysis module is connected with the interference classification module and transmits the user load information acquired in real time to the interference classification module after analyzing the acquired demand characteristic information; the interference classification module reads the analysis result of the load demand characteristic analysis module and calculates and classifies the interference type of each load; the computing node resource analysis module is connected with the resource monitoring module and records the resource use condition and the resource residual condition of the physical server in real time; the interference calculation module is respectively connected with the interference classification module and the calculation node resource analysis module and carries out interference degree index calculation on results obtained by the interference classification module and the calculation node resource analysis module; the scheduling module is connected with the interference calculation module and receives the calculation result of the interference calculation module, and the load is distributed to the physical server with the minimum interference degree index according to the calculation result;
the interference degree index is as follows: and adding the load demand characteristic matrix to the used resource matrix of the corresponding calculation node, and calculating the difference value between each column in the new matrix and the maximum resource vector of the node, wherein the sum of the values of the difference value larger than 0 is the interference degree index.
3. The system of claim 2, wherein the load demand characteristic analysis module divides the load into a loading period, an execution period and an end period according to the life cycle, and each period corresponds to four dimensions of computing resources, memory resources, hard disk resources and network resources to form a 4 x 3 demand characteristic matrix;
the four dimensions are quantized into eleven grades from 0-10 according to respective demand degrees.
4. The system of claim 3, wherein the interference classification module reads the demand signature matrix corresponding to the load demand signature analysis module according to the load.
5. The system according to claim 2, wherein the compute node resource analysis module uses vectors and matrices to jointly represent the resource usage of each physical server, specifically, the compute node obtains a four-dimensional maximum resource vector according to the quantization level of the compute resource to represent the maximum resource of the compute node; and quantifying the distributed computing resources of the three time intervals to obtain a used resource matrix of 4 × N, wherein: n represents the number of working time periods, the used resource usage is obtained by monitoring operating system parameters of the physical server, the future resource usage is obtained according to the used resources plus the corresponding resources used by the scheduled load, and the allocable resource vector is obtained by subtracting the used resource vector at a certain moment from the maximum resource vector.
6. The system of claim 2, wherein the interference calculation module adds the requirement characteristic matrix of each load to the used resource matrix of the corresponding calculation node, and calculates the difference between each column in the new matrix and the maximum resource vector of the node, wherein the difference is greater than 0, and the sum of the values is the interference degree index.
7. The system of claim 2, wherein the scheduling module performs optimal scheduling according to the interference degree index of each load, a greedy algorithm is used, during each scheduling, the current load is scheduled to the server with the smallest generated interference degree index without considering the subsequent load, and when the interference degree indexes of a plurality of physical servers to the load to be scheduled are the same, one physical server is randomly selected for scheduling.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911317923.9A CN110995863B (en) | 2019-12-19 | 2019-12-19 | Data center load distribution method and system based on load demand characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911317923.9A CN110995863B (en) | 2019-12-19 | 2019-12-19 | Data center load distribution method and system based on load demand characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110995863A CN110995863A (en) | 2020-04-10 |
| CN110995863B true CN110995863B (en) | 2021-07-30 |
Family
ID=70063152
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911317923.9A Active CN110995863B (en) | 2019-12-19 | 2019-12-19 | Data center load distribution method and system based on load demand characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110995863B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112231314A (en) * | 2020-11-05 | 2021-01-15 | 深圳市丽湖软件有限公司 | Quality data evaluation method based on ETL |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102958166A (en) * | 2011-08-29 | 2013-03-06 | 华为技术有限公司 | Resource allocation method and resource management platform |
| CN103516744A (en) * | 2012-06-20 | 2014-01-15 | 阿里巴巴集团控股有限公司 | A data processing method, an application server and an application server cluster |
| CN107040475A (en) * | 2016-11-14 | 2017-08-11 | 平安科技（深圳）有限公司 | Resource regulating method and device |
| US10104165B1 (en) * | 2012-08-30 | 2018-10-16 | Amazon Technologies, Inc. | Sharing network connections to content sources |
| CN110399200A (en) * | 2019-06-26 | 2019-11-01 | 苏州浪潮智能科技有限公司 | A kind of cloud platform resource regulating method and device |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102346460B (en) * | 2011-05-27 | 2013-11-13 | 运软网络科技(上海)有限公司 | Transaction-based service control system and method |
| CN102780759B (en) * | 2012-06-13 | 2016-05-18 | 合肥工业大学 | Based on the cloud computing resource scheduling method in regulation goal space |
| JP6597823B2 (en) * | 2018-03-23 | 2019-10-30 | 日本電気株式会社 | Load balancing device, communication system, control method and program |
| CN108429815A (en) * | 2018-03-23 | 2018-08-21 | 重庆邮电大学 | Dynamic resource scheduling method based on OpenStack |
| CN110311965A (en) * | 2019-06-21 | 2019-10-08 | 长沙学院 | Method for scheduling task and system under a kind of cloud computing environment |
-
2019
- 2019-12-19 CN CN201911317923.9A patent/CN110995863B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102958166A (en) * | 2011-08-29 | 2013-03-06 | 华为技术有限公司 | Resource allocation method and resource management platform |
| CN103516744A (en) * | 2012-06-20 | 2014-01-15 | 阿里巴巴集团控股有限公司 | A data processing method, an application server and an application server cluster |
| US10104165B1 (en) * | 2012-08-30 | 2018-10-16 | Amazon Technologies, Inc. | Sharing network connections to content sources |
| CN107040475A (en) * | 2016-11-14 | 2017-08-11 | 平安科技（深圳）有限公司 | Resource regulating method and device |
| CN110399200A (en) * | 2019-06-26 | 2019-11-01 | 苏州浪潮智能科技有限公司 | A kind of cloud platform resource regulating method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110995863A (en) | 2020-04-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109218355B (en) | Load balancing engine, client, distributed computing system and load balancing method | |
| Mazumdar et al. | Power efficient server consolidation for cloud data center | |
| CN107592345B (en) | Transaction current limiting device, method and transaction system | |
| US8417499B2 (en) | Enabling real-time testing of on-demand infrastructure to predict service level agreement compliance | |
| CN103699445A (en) | Task scheduling method, device and system | |
| CN103401947A (en) | Method and device for allocating tasks to multiple servers | |
| CN108123998B (en) | Heuristic request scheduling method for delay sensitive application in multi-cloud data center | |
| Huang et al. | Migration-based elastic consolidation scheduling in cloud data center | |
| Tang et al. | Data-aware resource scheduling for multicloud workflows: A fine-grained simulation approach | |
| CN110149377A (en) | A kind of video service node resource allocation methods, system, device and storage medium | |
| CN108491255B (en) | Self-service MapReduce data optimal distribution method and system | |
| Ashraf | Cost-efficient virtual machine provisioning for multi-tier web applications and video transcoding | |
| Cheng et al. | Proscale: Proactive autoscaling for microservice with time-varying workload at the edge | |
| Meskar et al. | Fair multi-resource allocation in mobile edge computing with multiple access points | |
| CN110995863B (en) | Data center load distribution method and system based on load demand characteristics | |
| Karthikeyan et al. | Preserving resource handiness and exigency-based migration algorithm (PRH-EM) for energy efficient federated cloud management systems | |
| Shi et al. | Characterizing and orchestrating VM reservation in geo-distributed clouds to improve the resource efficiency | |
| CN115794421B (en) | Resource allocation method and device and electronic equipment | |
| Farooq et al. | Adaptive and resilient revenue maximizing dynamic resource allocation and pricing for cloud-enabled IoT systems | |
| CN115421930B (en) | Task processing method, system, device, equipment and computer readable storage medium | |
| Seracini et al. | A comprehensive resource management solution for web-based systems | |
| Ambati et al. | Modeling and analyzing waiting policies for cloud-enabled schedulers | |
| Wei et al. | Towards multi-resource physical machine provisioning for IaaS clouds | |
| Zhou et al. | Fine-grained scheduling in multi-resource clusters | |
| Mehta | Designing an effective dynamic load balancing algorithm considering imperative design issues in distributed systems |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |