KR20210083465A - Multi-Server Management Method to Enhance Power Efficiency Using Workload Prediction of Virtual Machine Clustering - Google Patents

Abstract

The present embodiments provide a multi-server management method and device for efficiently and stably operating multi-server resources by clustering virtual machines and predicting workload using artificial intelligence learned for each cluster. The method includes the steps of: clustering a plurality of virtual machines mixedly located in a plurality of servers into a plurality of clusters according to a pattern of the workload of the virtual machines; predicting a next workload of the virtual machine through a workload prediction model from a current workload of the virtual machine belonging to the cluster; combining the plurality of virtual machines into a virtual machine set based on the predicted next workload of the virtual machine; and deploying the combined virtual machine set to the plurality of servers.

Description

Multi-Server Management Method to Enhance Power Efficiency Using Workload Prediction of Virtual Machine Clustering}

The technical field to which the present invention belongs relates to a method and apparatus for managing multiple servers using virtual machine workload clustering prediction.

The content described in this section merely provides background information for the present embodiment and does not constitute the prior art.

Virtualization technology is a technology that can install and use a computer operating system (OS) without being affected by system structure or hardware. By dividing one physical machine (eg, a server) into several virtual pieces, the physical machine can be used efficiently. Dozens of virtual machines are allocated the server's processor and memory resources.

Numerous servers are set up in a virtualized environment in a data center. In a data center with tens to hundreds, thousands to tens of thousands of servers, the number of virtual machines is greater than the number of servers. Data centers are inherently characterized by high computational complexity, which is far beyond the intuition of an administrator. A data center is an energy-dense building, where operating costs are much higher than facility costs, and a single building uses the most power.

Due to the condition that data centers must be operated continuously 24 hours a day, 365 days a year, managers are adopting conservative operation methods. The longer the server is down, the more costly the damage will be. Although the size of the data center is growing, it is still the case that the data center is operated by the intuition of the manager.

Korean Patent Publication No. 10-2018-0105874 (2018.10.01)

Embodiments of the present invention have a main purpose of the invention to efficiently and stably operate multiple server resources by clustering virtual machines and predicting workloads using artificial intelligence learned for each cluster.

Other objects not specified in the present invention may be additionally considered within the scope that can be easily inferred from the following detailed description and effects thereof.

According to one aspect of this embodiment, clustering a plurality of virtual machines located in a mixture in a plurality of servers into a plurality of clusters according to the pattern of the workload of the virtual machines, from the current workload of the virtual machines belonging to the cluster Predicting the next workload of the virtual machine through a workload prediction model, combining the plurality of virtual machines into a virtual machine set based on the predicted next workload of the virtual machine, and the combined virtual machine It provides a multi-server management method comprising the step of deploying a set to the plurality of servers.

The workload of the virtual machine may be time series data.

The step of disposing the combined virtual machine set to the plurality of servers minimizes the distribution of the workload of the combined virtual machine set in a manner that offsets the pattern of the workload of the plurality of virtual machines, and Workload distribution can be minimized.

In the multi-server management method, after the step of arranging the combined virtual machine set to the plurality of servers, if the predicted change in the next workload of the virtual machine exceeds a reference range, the virtual machine located in one server is transferred to another server. The method may further include migrating to and relocating the plurality of virtual machines.

The workload prediction model may be independently and parallelly trained for each cluster.

The workload prediction model may predict the next workload based on a Recurrent Neural Network (RNN) of a Long Short Term Memory (LSTM) structure.

The clustering may include classifying the clusters based on metadata.

The metadata may include a name of the virtual machine, an application list, a type of workload, and a ratio of resources used.

In the clustering step, a keyword is extracted from the name of the virtual machine, and when the keyword corresponds to a stopword, weights of a plurality of clusters to correspond to the virtual machine may be uniformly set.

The clustering may include extracting information on an application belonging to a preset usage frequency range from the application list and classifying the cluster based on the information on the application.

The type of workload is (i) a first type that gradually increases in a first time interval, (ii) a second type that increases rapidly in a second time interval that is narrower than the first time interval, and (iii) ) can be classified into a third type, which is a constant form regardless of time.

The ratio of the used resources may be divided into a processor usage ratio and a memory usage ratio.

As described above, according to the embodiments of the present invention, there is an effect of efficiently and stably operating multiple server resources by clustering a plurality of virtual machines and predicting workload using artificial intelligence learned for each cluster. have.

Even if it is an effect not explicitly mentioned herein, the effects described in the following specification expected by the technical features of the present invention and their potential effects are treated as if they were described in the specification of the present invention.

1 is a diagram illustrating a virtual machine installed in a server.
2 and 3 are flowcharts illustrating the overall operation of a multi-server management method according to embodiments of the present invention.
4 is a flowchart illustrating an operation of clustering a new virtual machine in a multi-server management method according to an embodiment of the present invention.
5 is a diagram illustrating meta data processed by a multi-server management method according to an embodiment of the present invention.
6 is a diagram illustrating cluster weights processed by the multi-server management method according to an embodiment of the present invention.
7 is a diagram illustrating an operation of learning a workload prediction model for each cluster in a multi-server management method according to an embodiment of the present invention.
8 is a diagram illustrating a state in which a plurality of virtual machines are disposed on a plurality of servers in a method for managing multiple servers according to an embodiment of the present invention.
9 is a diagram illustrating a multi-server management apparatus according to another embodiment of the present invention.

Hereinafter, in the description of the present invention, if it is determined that the subject matter of the present invention may be unnecessarily obscured as it is obvious to those skilled in the art with respect to related known functions, the detailed description thereof will be omitted, and some embodiments of the present invention will be described. It will be described in detail with reference to exemplary drawings.

1 is a diagram illustrating a virtual machine.

A virtual machine (VM) is a software implementation of a computing environment. Virtual machines are designed to provide a hardware-independent program execution environment and abstraction. You can install and run an operating system or application on a virtual machine.

Many servers are set up in a virtualized environment. The virtual machine installed on the server operates 24 hours a day, 365 days a year and uses resources such as CPU, Memory, Disk I/O, and Network I/O. The use of these resources is called a workload. The data center manager monitors whether numerous data center resources are operating normally by using the manager tool that can monitor the workload occurring per minute. As a result of monitoring, if an abnormality occurs in the server, the administrator takes action.

Virtualization technology supports the migration function. Migration is a function of moving a virtual machine located on one server to another server. If all virtual machines assigned to a server are moved to another server, the server enters the idle state and can be switched to power saving mode or turned off. Using virtual machine relocation to save server power is called Virtual Machine Consolidation. Virtual machine consolidation is a matter of deciding which virtual machine located on which server to migrate to which other server.

Migrating a virtual machine takes time, and the migration itself is an overloaded operation. During migration, service is provided without interruption, but the quality of service may be degraded due to the existence of time intervals. The administrator should be able to understand to some extent which virtual machine is located on which server, and if migration occurs frequently, the number of management points will increase from an operational point of view. For virtual machine consolidation, the condition that virtual machine migration should be minimized should be considered.

Existing virtual machine integration studies monitor the server environment 24 hours a day, and migrate the virtual machine when an overload occurs or a server with a low load is detected. The existing method sets the threshold to a certain extent, but when an overload occurs, it is necessary to calculate where to migrate the virtual machine, and to perform the virtual machine migration task that causes an overload on the overloaded server. must be done forcefully.

The multi-server management method according to the embodiments of the present invention applies artificial intelligence (deep learning) technology to predict time-series workload data used by server resources, so that there is a big problem affecting the server operation of the data center Take precautions before they happen. The multi-server management method can efficiently and stably operate multi-server resources by clustering multiple virtual machines and predicting workloads using artificial intelligence learned for each cluster.

2 and 3 are flowcharts illustrating the overall operation of a multi-server management method according to embodiments of the present invention.

The multi-server management method may be performed by the multi-server management apparatus. The configuration of the multi-server management apparatus is exemplarily shown in FIG. 9 .

In step S210, the processor clusters a plurality of virtual machines coexistingly located in a plurality of servers into a plurality of clusters according to a pattern of a workload of the virtual machine. The workload of the virtual machine is time series data, and resource usage information of the virtual machine.

In the step of clustering ( S210 ), clusters may be classified based on metadata. The metadata may include a name of a virtual machine, an application list, a type of workload, and a ratio of resources used.

In the clustering step ( S210 ), a keyword is extracted from the name of the virtual machine, and when the keyword corresponds to a stopword, weights of a plurality of clusters to correspond to the virtual machine may be uniformly set.

In the clustering step (S210), information on an application belonging to a preset usage frequency range may be extracted from the application list, and the cluster may be classified based on the information of the application.

In step S220, the processor predicts the next workload of the virtual machine from the current workload of the virtual machine belonging to the cluster through the workload prediction model. The workload prediction model can be trained independently and in parallel for each cluster. The workload prediction model can predict the next workload based on a Recurrent Neural Network (RNN) of a Long Short Term Memory (LSTM) structure.

In step S230, the processor combines a plurality of virtual machines into a virtual machine set based on the predicted next workload of the virtual machine.

In step S240, the processor places the combined virtual machine set on the plurality of servers. Placing the combined virtual machine set on the plurality of servers (S240) minimizes the distribution of the workload of the combined virtual machine set in a way that offsets the pattern of the workload of the plurality of virtual machines, and the workload of the plurality of servers The load distribution can be minimized.

In step S250, after the step of arranging the combined virtual machine set to the plurality of servers, if the predicted change in the next workload of the virtual machine exceeds the reference range, the virtual machine located in one server is migrated to another server, Multiple virtual machines can be relocated.

In step S310, the multi-server management device reads metadata from the virtual machine metadata database. Metadata is analyzed and defined according to the data type.

In step S320, the multi-server management device clusters the virtual machine. The multi-server management device builds a workload data learning database based on the virtual machine cluster information.

In step S330, the multi-server management device predicts virtual machine workload clustering using the workload data learning database. It identifies which cluster the virtual machine is classified into, and time-series predictions of workloads such as CPU and memory based on the cluster's training file.

In step S340, the multi-server management apparatus searches for a virtual machine combination according to the workload prediction value for each virtual machine. Based on the virtual machine workload forecast, the workload of the server is kept stable, and a combination of virtual machines that balances the workload for each server is explored.

In step S350, the multi-server management device executes virtual machine migration. If the derived virtual machine combination is different from the previous combination, the virtual machine is migrated according to the newly derived virtual machine combination.

4 is a flowchart illustrating an operation of clustering a new virtual machine in a multi-server management method according to an embodiment of the present invention.

In step S410, the processor extracts the name keyword of the new virtual machine.

In step S420, the processor determines whether the keyword corresponds to a stopword. Check whether the name of the virtual machine is not written intuitively. For example, there may be names such as test123, aaa123, and abc.

In step S430, if the keyword corresponds to the stopword, the processor evenly distributes the cluster weight. Since virtual machines cannot be classified, the cluster weight is distributed evenly. For example, the new virtual machine #100 can be set to have a weight of 0.25 for Cluster #1, 0.25 for Cluster #2, 0.25 for Cluster #3, and 0.25 for Cluster #4.

If the keyword does not correspond to the stopword in step S440, the processor classifies the virtual machine clustering (S440).

5 is a diagram illustrating meta data processed by a multi-server management method according to an embodiment of the present invention.

Clustering of time series data is carried out by dividing by time period. Since the workload of a virtual machine is time-series data, it shows a unique pattern over time. Patterns can be classified by time period, such as whether a high workload is seen on weekdays rather than on weekends, or whether a high workload is seen during work hours rather than dawn. Since the purpose of virtual machine consolidation itself is to explore the combination of virtual machines that make the workload stable, it is possible to classify them by time period.

It is simple to classify the patterns of virtual machines only by time period, but additional information is considered in order to increase the accuracy. Referring to FIG. 5 , information to be considered for clustering a virtual machine is exemplified.

The metadata may include a name of a virtual machine, an application list, a type of workload, and a ratio of resources used. Metadata can be extracted through a programming interface associated with the server.

When an administrator creates a virtual machine for the first time, it intuitively writes the name of the virtual machine, so it provides a clue to what kind of virtual machine it is.

If you understand the application, that is, which process is used, you can know the characteristics of the virtual machine.

The type of workload is (i) a first type in a progressively increasing form in a first time interval, (ii) a second type in a rapidly increasing form in a second time interval narrower than the first time interval, and (iii) time It can be classified into a third type, which is a constant form regardless of For example, the first type is a normal type in which the utilization rate of the server gradually increases based on the business time period, and the second type is a ramp type in which the utilization rate increases rapidly during a specific time period. Type 3 can be set as a steady type, which is a constant type regardless of time.

The ratio of used resources may be divided into a processor usage ratio and a memory usage ratio. For example, the sum of the items of CPU and Memory should be 1, and if both items are 0.5, it means that they are used uniformly.

6 is a diagram illustrating cluster weights processed by the multi-server management method according to an embodiment of the present invention.

The clustering technique applies fuzzy clustering, so that one virtual machine can be included in multiple clusters, and how close it is to each cluster is derived in the form of a weight. In fuzzy clustering, each observation can belong to several clusters rather than to only one cluster, and it presents the probability or probability of belonging to each cluster.

Referring to FIG. 6 , virtual machine #1 has a weight of 0.82 in cluster #1, and weights of 0.14 and 0.04 in cluster #2 and cluster #5, respectively. This means that virtual machine #1 corresponds to clusters #1, #2, and #5, and when making predictions, using the learning results of clusters #1, #2, and #5, prediction is made based on weights.

7 is a diagram illustrating an operation of learning a workload prediction model for each cluster in a multi-server management method according to an embodiment of the present invention.

It considers the workload pattern for virtual machine integration, predicts the workload pattern for each virtual machine, and decides on virtual machine relocation.

A typical prediction model based on artificial intelligence collects various patterns and learns through an artificial intelligence algorithm. When the data to be predicted is input to the corresponding learning model, the prediction data is output.

VMs #1 to VM #10 are collected data to be used for training. When learning through the LSTM algorithm, a learning model is created. When the existing value of the object to be predicted (Target VM) is input into the created learning model, the value of the object is predicted (Predicted VM) according to the pattern stored in the existing learning model.

RNN is a neural network useful for expressing and learning patterns that have continuity and whose back and forth order is important, and the LSTM model is used to encode temporally ordered data. LSTM analyzes the data and creates a vector.

RNN has a loop connected so that the result of the hidden layer goes back to the input of the same hidden layer, and has a structure in which past data influences the future. That is, in RNN, hidden nodes are connected to edges with directions to form a cyclic structure.

The LSTM model is a structure in which cells with multiple gates are added to the hidden layer. The hidden layer has a cell including an input gate, an output gate, and a forget gate, that is, a memory block. The forget gate is a gate for forgetting past information, and the input gate is a gate for remembering present information. The gates each have an intensity and a direction. Cells act as conveyor belts, and gradients can relatively sustain propagation over long periods of time. Forgetting gate is a step in determining how much information in the previous state to remember. A value between 0 and 1 is output. If it is 0, it completely forgets the information in the previous state, and if it is 1, it can be that the previous information is fully remembered.

The multi-server management method according to the present embodiment analyzes patterns of data to be learned, classifies them into N (natural number) clusters, and performs learning among data belonging to the corresponding cluster. That is, it proceeds N times. For example, virtual machine data is divided into four patterns, and the learning model is managed by collecting virtual machine data belonging to each pattern and conducting learning. The number of patterns may be set to various numerical values according to required design requirements.

Compared to a typical predictive model, this split clustering learning can maintain the specificity of the pattern, and if the learning is performed in parallel, the learning time can be minimized.

8 is a diagram illustrating a state in which a plurality of virtual machines are disposed on a plurality of servers in a method for managing multiple servers according to an embodiment of the present invention.

8, virtual machine #1 is allocated to server #1, virtual machine #2, virtual machine #3, virtual machine #4 are allocated to server #2, virtual machine #5 is allocated to server #3, Assuming that virtual machine #6 is allocated, server #1 may be interpreted as a low load state, and server #2 may be interpreted as being close to an overload state.

The multi-server management method according to the present embodiment monitors time-series workload data of the server so that the workload of the server is continuously operated stably. In other words, virtual machine migration is minimized.

Referring to the right side of FIG. 8 , server #1, server #2, and server #3 are maintained at relatively similar levels rather than under low load and overload, and it can be seen that they are stable over time. That is, it is possible to minimize the distribution of the workload of the combined virtual machine set and minimize the distribution of the workload of a plurality of servers.

Minimize migration by keeping the server's workload as stable as possible. The closer the standard deviation of the workload of one server to 0, the more stable it can be defined, and the standard deviation of the workload of each server is minimized.

The server temperature rises as the load increases, affecting the lifespan of the equipment. If the load level is high because virtual machines are clustered on one server, the lifespan of the server will be shortened and the failure rate will increase. Balance the workload on each server. Minimize the standard deviation of the workload for each server.

The virtual machine integration according to this embodiment combines virtual machines showing opposite patterns, cancels the patterns to bring the standard deviation of the workload close to zero, and the workload deviation for each server is also close to zero.

9 is a diagram illustrating a multi-server management apparatus according to another embodiment of the present invention.

The multiple server management device 12 includes at least one processor 14 , a computer readable storage medium 16 , and a communication bus 18 .

The processor 14 may be controlled to operate as the multi-server management device 12 . For example, the processor 14 may execute one or more programs stored in the computer-readable storage medium 16 . The one or more programs may include one or more computer-executable instructions, which when executed by the processor 14 configure the multi-server management device 12 to perform operations in accordance with the exemplary embodiment. can be

Computer-readable storage medium 16 is configured to store computer-executable instructions or program code, program data, and/or other suitable form of information. The program 20 stored in the computer readable storage medium 16 includes a set of instructions executable by the processor 14 . In one embodiment, computer-readable storage medium 16 includes memory (volatile memory, such as random access memory, non-volatile memory, or a suitable combination thereof), one or more magnetic disk storage devices, optical disk storage devices, It may be flash memory devices, other types of storage media accessed by the multi-server management apparatus 12 and capable of storing desired information, or a suitable combination thereof.

A communication bus 18 interconnects the various other components of the multi-server management device 12, including a processor 14 and a computer-readable storage medium 16.

The multi-server management device 12 may also include one or more input/output interfaces 22 and one or more communication interfaces 26 that provide interfaces for one or more input/output devices. The input/output interface 22 and the communication interface 26 are coupled to the communication bus 18 . The input/output device may be connected to other components of the multi-server management device 12 via the input/output interface 22 .

The server may be implemented in a logic circuit by hardware, firmware, software, or a combination thereof, or may be implemented using a general-purpose or special-purpose computer. The server may be implemented using a hardwired device, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), or the like. In addition, the server may be implemented as a system on chip (SoC) including one or more processors and controllers.

The server may be mounted in the form of software, hardware, or a combination thereof on a computing device provided with hardware elements. A server is a variety of devices including, in whole or in part, a communication device such as a communication modem for communicating with various devices or wired and wireless communication networks, a memory for storing data for executing a program, and a microprocessor for executing operations and commands by executing the program. can mean

Although it is described that each process is sequentially executed in FIGS. 2 to 4 , this is only an exemplary description, and those skilled in the art are shown in FIGS. 2 to 4 within the range that does not depart from the essential characteristics of the embodiment of the present invention. Various modifications and variations may be applied by changing the order described, executing one or more processes in parallel, or adding other processes.

The operations according to the present embodiments may be implemented in the form of program instructions that can be performed through various computer means and recorded in a computer-readable medium. Computer-readable media refers to any medium that participates in providing instructions to a processor for execution. Computer-readable media may include program instructions, data files, data structures, or a combination thereof. For example, there may be a magnetic medium, an optical recording medium, a memory, and the like. A computer program may be distributed over a networked computer system so that computer readable code is stored and executed in a distributed manner. Functional programs, codes, and code segments for implementing the present embodiment may be easily inferred by programmers in the technical field to which the present embodiment pertains.

The present embodiments are for explaining the technical idea of the present embodiment, and the scope of the technical idea of the present embodiment is not limited by these embodiments. The protection scope of the present embodiment should be interpreted by the following claims, and all technical ideas within the equivalent range should be construed as being included in the scope of the present embodiment.

Claims (5)

clustering a plurality of virtual machines mixedly located in a plurality of servers into a plurality of clusters according to a pattern of a workload of the virtual machines;
predicting the next workload of the virtual machine through a workload prediction model from the current workload of the virtual machine belonging to the cluster;
combining the plurality of virtual machines into a virtual machine set based on the predicted next workload of the virtual machine; and
Deploying the combined virtual machine set to the plurality of servers
Multiple server management methods including. According to claim 1,
The workload of the virtual machine is time series data,
The step of disposing the combined virtual machine set on the plurality of servers minimizes the distribution of the workload of the combined virtual machine set in a manner that offsets the pattern of the workload of the plurality of virtual machines, and A multi-server management method characterized in that the distribution of the workload is minimized. According to claim 1,
After disposing the combined virtual machine set to the plurality of servers,
Multi-server management, characterized in that when the predicted change in the next workload of the virtual machine exceeds a reference range, migrating a virtual machine located in one server to another server and relocating the plurality of virtual machines Way. According to claim 1,
The workload prediction model is independently trained in parallel for each cluster,
The workload prediction model is a multi-server management method, characterized in that it predicts the next workload based on a Recurrent Neural Network (RNN) of a Long Short Term Memory (LSTM) structure. According to claim 1,
In the clustering step, the cluster is classified based on metadata,
The metadata includes a name of the virtual machine, an application list, a type of workload, and a ratio of resources used,
In the clustering step, a keyword is extracted from the name of the virtual machine, and if the keyword corresponds to a stopword, uniformly set weights of a plurality of clusters to correspond to the virtual machine,
The clustering includes extracting information on applications belonging to a preset usage frequency range from the application list and classifying the clusters based on the information on the applications,
The type of workload is (i) a first type that gradually increases in a first time interval, (ii) a second type that increases rapidly in a second time interval that is narrower than the first time interval, and (iii) ) is divided into the third type, which is a constant form regardless of time,
The multi-server management method, characterized in that the ratio of the resource used is divided into a processor usage rate and a memory usage rate.

Images