KR101557747B1 - System and method for allocating virtual machine for effective use of multi resource in cloud - Google Patents

Abstract

The present invention relates to a virtual machine allocation system and method for efficient utilization of multiple resources in a cloud, the system comprising: a node performance database; a data center broker having a node list retrieved from the node performance database and managing a data center; A service provider's hypervisor, which consists of a virtual machine creation controller that creates a virtual machine on the optimal node to allocate a virtual machine and a virtual machine allocation control system that includes a virtual machine migration controller that controls the migration of the virtual machine. ); And a node performance analyzer for analyzing and evaluating the performance of the nodes and the nodes, and storing the nodes in the node performance database, thereby increasing the utilization rate of the high performance nodes to efficiently use the entire computer resources, Memory management not only reduces fragmentation, but also allows rapid allocation of virtual machines.

Description

TECHNICAL FIELD The present invention relates to a virtual machine allocation system and method for effectively utilizing multiple resources in a cloud,

The present invention relates to a virtual machine allocation system and method, and more particularly, to a virtual machine allocation system and a virtual machine allocation method. More particularly, the present invention relates to a virtual machine allocation system and method, And more particularly, to a system and method for allocating virtual machines efficiently.

Social network service, etc., the problem of excess traffic caused by the increase in the amount of information generated and transmitted by individuals is becoming serious enough to be felt by consumers. To cope with this, many companies are using cloud computing to solve load balancing and cost problems. In addition, cloud computing is no longer the exclusive category of IT majors, providing individuals with high-quality resources in the form of services.

In cloud computing, virtualization of computing resources creates a virtual machine that handles the user's needs and assigns it to the nodes registered in the data center. In this process, FCFS, round-robin, and other virtual machine allocation techniques are used to satisfy the node's fair use and user requirements.

Specifically, virtualization, which is an important technology for creating a virtual machine in a virtual machine allocation system, represents a resource having a logical form in a physical resource used by a user.

In cloud computing, virtualized computer resources are provided through virtual machines through technologies that abstract resources to provide services to meet user needs. The goal of virtualization is to simplify access and infrastructure management of computer resources through layers between services and computer resources.

To provide services in cloud computing, a provider creates a logical operating system (OS) through a virtual machine using virtualization technology. Choosing a host OS to create a guest OS is largely reflected in the quality of service do. If the host OS selected by the computing resources required by the cloud infrastructure system is not capable of generating the guest OS due to lack of performance, the service can not be provided and the efficiency and economical burden of the cloud service can not be reduced.

In order to allocate the virtualized resources in the existing cloud computing computer, there are FCFS (First Come First Serve), Round-Robin, and Match-Making.

First, First Come First Serve (FCFS) is a simple scheduling that operates as a FIFO (First In First Out) that processes the first incoming service with non-linear scheduling. As a representative example, Queue among data structures is implemented in the form of FCFS.

Fig. 5 is a structural view of FCFS. FCFS is not suitable for an interactive system requiring fast response, and is very efficient for a batch processing system. The disadvantage of FCFS is that there is no process of distinguishing or analyzing nodes, so that a virtual machine waiting for processing can wait for the virtual machine being processed by the node to end, resulting in a convoy effect. This means that the processor has an unbalanced imbalance.

FCFS, which is used to assign a virtual machine to a node in the cloud computing, requests a virtual machine allocation to an unused node in the order registered in the data center .

Although FCFS is simple to implement, it allocates virtual machines that process user requests in the order in which the nodes are registered in the data center. Therefore, even if the requests of users with high throughput are allocated to the low-level nodes, As a result, the overall processing speed of the operation increases.

The representative virtual machine allocation system studied based on FCFS has virtual machine allocation scheduling provided in the cloud sim. In the cloud core, when a virtual machine allocation request is received, the user sequentially assigns the virtual machine to the node having the largest number of unused cores among the registered nodes in the data center.

FIG. 6 is a diagram illustrating a virtual machine allocation system of a virtual machine prioritization scheme considering a scheduling delay in a Xen environment. The virtual machine allocation system includes a SEDF (Simple Earliest Deadline First), we propose a virtual machine allocation scheme in which a node that has returned a processed virtual machine fetches an unprocessed virtual machine from a node that has an unprocessed virtual machine and processes it instead. In this technique, a node that returns a terminated virtual machine fetches an unprocessed virtual machine and processes the virtual machine with FCFS.

Second, round-robin (scheduling) is a scheduling in which a virtual machine is allocated in a time-by-time manner without assigning a priority to a task requested by preemption-type scheduling. All nodes use alternate nodes without being restricted by priority or the like.

(

: 프로세스 개수)로 실행되는 것처럼 보이는 것을 말한다.In order for the round robin to allocate the virtual machine to another node, the amount of time that the virtual machine stays in the node is set and the virtual machine is assigned to the next node when the execution of the virtual machine at the node passes the specified time. The performance of the round robin is greatly influenced by the specified amount of time. In Linux, which is a typical OS, the size of the specified time amount varies according to the priority and operation, and it is based on 100 ms. In the case of Windows XP, the size varies according to the front or rear operation, and is based on 20 ms. If the defined amount of time is 1μsec, the round robin is called Processor Sharing,

(

: Number of processes).

FIG. 7 is a schematic diagram of a round robin. In Eucalyptus Systems, Eucalyptus is a representative system in which round robin is used for virtual machine allocation in cloud computing. In Eucalyptus, we use alternate nodes that are able to allocate virtual machines, and continue to work on one node until another assignable node is found.

In addition to round robin, eucalyptus includes Greedy to select the first node found to be able to allocate and execute a virtual machine, and Power-Save to select a node to be woken up first, .

Third, Match-Making is one of the projects initiated at Wisconsin University in the United States to support large-scale calculations in distributed computing environments. Matchmaking finds the best node to connect to a node that meets the requirements of the virtual machine and allocates the virtual machine.

FIG. 8 is a structure diagram of matchmaking. Referring to FIG. 8, a provider transmits information (process 1) to nodes of a data center to a match maker. When a requester requests a resource, it requests a matchmaker to select a node to allocate resources to the virtual machine (process ②). In the match maker, when the requester requests a resource, the requestor requests the provider a virtual machine allocation service (process (4)) when providing the information of the node suitable for the resource requested by the virtual machine among the information of the node received through the provider, , And the provider allocates a virtual machine to the node that meets the request of the requester (step 5) to provide the service. Here, the matchmaker provides only the node information to resolve directly between the provider and the requester, and does not act as a relay.

In cloud computing, matchmaking classifies information of nodes satisfying the requirements of a virtual machine, and sequentially assigns virtual machines to nodes that meet the user's request. OpenNebula is a typical cloud service system that uses matchmaking as a virtual machine allocation system. Open Nebula allocates virtual machines by providing information about the nodes whose CPU usage can satisfy the requirements of the virtual machine.

As we have seen, the existing virtual machine allocation system aims at fair use of the equipment by minimizing the difference in the usage rate of all nodes. However, since the cloud system is not considered for long-term use, when the equipment is added, the performance difference of the nodes is not reflected, and high-performance nodes are frequently used due to the fair use standard.

Korean Patent Publication No. 10-2010-0092850 (published on Aug. 23, 2010) Korean Patent Laid-Open Publication No. 10-2012-0071979 (published July 23, 2012)

SUMMARY OF THE INVENTION The present invention has been made in order to solve the problems of the prior art described above, and it is an object of the present invention to provide a method and apparatus for analyzing nodes in a data center to prioritize high- And to provide a virtual machine allocation system and method for efficient utilization of multiple resources in the cloud.

According to another aspect of the present invention, there is provided a virtual machine allocation system for efficiently utilizing multiple resources in a cloud, including a node performance database, a data center broker having a node list loaded from the node performance database and managing a data center, And a virtual machine assignment control system including a virtual machine creation controller for creating a virtual machine in a node optimal for allocating a virtual machine in the node list and a virtual machine migration controller for controlling migration of the virtual machine, Service Provider's Hypervisor; And a data center including nodes, and a node performance analyzer for analyzing and evaluating the performance of the nodes and storing them in the node performance database.

Meanwhile, a virtual machine allocation method for efficient utilization of multiple resources in a cloud of the present invention includes: searching nodes in a data center for performance analysis of nodes when a cloud service is started; Assigning weights to the performance analysis results and storing them in a node performance analysis database; Querying node information from a node list loaded from the node performance analysis database when a virtual machine allocation request is received from a user; Selecting an optimal node for allocating a virtual machine in the node list; And allocating a virtual machine to the optimal node.

As described above, according to the virtual machine allocation system and method for efficient utilization of multiple resources in the cloud according to the present invention, when the new node is added, the performance of the nodes managed in the data center is analyzed, , It is possible to increase the utilization rate of high performance nodes by efficiently allocating the entire computing resources and to reduce the fragmentation phenomenon through the memory management of the nodes, Virtual machines can be allocated.

1 is an overall configuration diagram of a virtual machine allocation system for efficiently utilizing multiple resources according to an embodiment of the present invention.
2 is a flowchart illustrating a process of allocating a virtual machine according to an embodiment of the present invention.
3 is an algorithm for analyzing performance of a node according to an embodiment of the present invention.
4 is an algorithm for finding an optimal node for allocating a virtual machine in a node list according to an embodiment of the present invention.
5 is a structural diagram of FCFS which is a conventional virtual machine allocation system.
FIG. 6 is a structural diagram of a virtual machine allocation system using a virtual machine priority allocation scheme considering a scheduling delay in a Xen environment.
7 is a structural diagram of a round robin which is a conventional virtual machine allocation system.
8 is a structure diagram of match making, which is a conventional virtual machine assignment system.

Hereinafter, a virtual machine allocation system and method for efficient utilization of multiple resources in a cloud according to the present invention will be described in detail with reference to the accompanying drawings.

1 is an overall configuration diagram of a virtual machine allocation system for efficiently utilizing multiple resources according to an embodiment of the present invention.

1, a virtual machine allocation system 1 according to an embodiment of the present invention includes a data center 10 including a node performance analyzer 11 and nodes 12, ; A service provider's hypervisor 20 configured with a node performance database 21, a data center broker 22 and a virtual machine allocation control system 23.

First, the system 1 searches for nodes 12 in all data centers connected to a network computer (NC) when a cloud service is started. Then, the system 1 retrieves information of the nodes from the node performance database 21 when the user requests resources.

Next, the data center 10 includes a node performance analyzer 11 and nodes 12. [

The node performance analyzer 11 analyzes the performance of the nodes 12 using the LU decomposition or the like while the system 1 searches for the nodes and analyzes the analyzed results in a high performance order And stores it in the node performance database 21. The detailed operation of the node performance analyzer 11 will be described in detail below.

Finally, the service provider hypervisor 20 includes a node performance database 21, a data center broker 22, a virtual machine allocate control system 23 ).

The data center broker 22 is responsible for managing and arbitrating the data center in the cloud system. The data center broker 22 has a node list retrieved from the node performance database 21 and this node list is composed of node information of all data centers in the node performance database 21. [

The virtual machine allocation control system 23 includes a virtual machine creation controller 25 and a virtual machine migration controller 24 and is responsible for all of the creation and migration of virtual machine allocations . When the user requests a resource, the virtual machine creation controller 25 creates a virtual machine by virtualizing the resource requested by the user, and when the virtual machine migration controller 24 migrates to another node in a special situation during system operation Control migration of virtual machines. At this time, the virtual machine creation controller 25 and the virtual machine migration controller 24 refer to the node information from the node list in the data center broker 22 to obtain necessary information when controlling the change in the system. In addition, in the virtual machine assignment control system 23, a node list satisfying the requirement among the node lists sorted in high-performance order is separately generated to select the node to which the virtual machine is to be allocated, A virtual machine is created in the appropriate node.

Here, the node performance analyzer 11 of the data center 10 will be described in detail. The node performance analyzer 11 analyzes and evaluates the performance of the nodes 12 managed by the data center 10.

분해(Decomposition)와 가우시안 소거법(Gaussian Elimination)을 이용한 역행렬을 구하는 과정을 이용하여 노드를 평가하기 위한 지표를 정한다.

분해는 행렬

를 하 삼각행렬

과 상 삼각행렬

로 분해하는 방법이다. 즉,First, the node performance analyzer 11 according to an embodiment of the present invention

The index for evaluating the node is determined by using the decomposition and Gaussian elimination inverse matrix calculation.

The decomposition is a matrix

Lower triangular matrix

Over-triangular matrix

. In other words,

를Matrix with the shape of

To

분해는 역행렬을 구하는 것 뿐아니라 연립방정식의 해를 구할 때도 사용된다.To solve the problem.

Decomposition is used not only to find the inverse matrix but also to solve the simultaneous equations.

An upper triangular matrix U is composed of a row of the matrix A trapezoidal (RREF, Reduced Row Echelon Form) and a lower triangular matrix L is configured to obtain the value to be used as component of the upper triangular matrix U. The row trapezoid satisfies the properties shown in Table 1.

The LU decomposition is divided into the Doolittle algorithm and the Crout algorithm according to the method of decomposing the matrix A into A = LU . The Doolittle algorithm decomposes the diagonal elements of the lower triangular matrix L to 1. When decomposing a matrix using the Doolittle algorithm, the upper triangular matrix U becomes a trapezoidal row by forward elimination of the Gaussian elimination method, and a lower triangular matrix L is obtained by using A = LU in reverse.

The Crout algorithm decomposes the diagonal elements of the upper triangular matrix U to 1. The components of the upper triangular matrix U can be obtained by dividing the components of the matrix A by the lower triangular matrix L and the first component of the row. A lower triangular matrix L is the first column is the value obtained by subtracting the process value copied from the matrix A and obtained by multiplying the first two components of the line in the U component and the front row of the second column starting from the L component of the matrix A. The formula is as follows.

은 하 삼각행렬 L의 각 행의 첫 번째 요소, 즉 첫 번째 열로 행렬 A의 각 행의 첫 번째 요소

을 가진다. 삼 삼각행렬 U의 첫 번째 행

는 하 삼각행렬 L의 (1,1) 요소를 행렬 A의 첫 번째 행렬의 각 요소에 나눠준 값이다. 여기서 j는 상 삼각행렬 U의 각 열의 번호이다.

The first element of each row of the lower triangular matrix L , that is, the first column, the first element of each row of matrix A

. First row of trilinear matrix U

Is the value obtained by dividing the (1,1) element of the lower triangular matrix L by each element of the first matrix of the matrix A. Where j is the number of each column of the upper triangular matrix U.

는 행렬 A의 요소에 하 삼각행렬 L의 i번째 행의 j-1번째까지의 요소와 상 삼각행렬 U의 j열의 j-1번째까지의 요소를 곱한 것의 합을 뺀 값이다.The remaining values except for the first column of the lower triangular matrix L

Is a value obtained by subtracting the sum of the elements of the matrix A up to the j- 1th element of the i- th row of the lower triangular matrix L and the elements up to the j- 1th element of the j- th column of the upper triangular matrix U.

는 먼저 상 삼각행렬 L의 i번째 행의 i-1까지, 하 삼각행렬 j번째 열의 U의 i-1까지의 각 요소를 곱한 것들의 합에서 행렬

값을 빼고, 하 삼각행렬 L의 대각요소로 나눈 값이다.

The first upper triangular matrix L i to i-1 of the second row, the lower triangular matrix in the matrix, j the sum of those multiplied by a respective element to the i-1-th column of U

And subtracting the value and dividing it by the diagonal elements of the lower triangular matrix L.

In the present invention, by adopting a commonly used Doolittle algorithm, a matrix A is defined as A = LU And the inverse matrix can be obtained. LU Table 2 shows the process of obtaining the inverse matrix using decomposition.

Matrix to LU Since the determinant is much easier to obtain than the method of Gauss elimination, which is a basic technique for obtaining the inverse matrix after decomposition, the increase in the calculation time is not noticeable even if the matrix size is large.

In addition, since the determinant of the triangular matrix is a product of all diagonal elements, the computation time of the Gaussian elimination method can be greatly reduced. Table 3 shows the Gaussian elimination method and the LU And the calculation speed of the Gaussian elimination method after decomposition is compared. As the matrix size increases, the LU It can be confirmed that the difference in the operation speed of the Gaussian elimination method after decomposition becomes large.

Next, the node performance analyzer 11 evaluates how quickly and accurately each node processes a given operation to analyze the performance of the node. The resources referred to in the node performance evaluation are CPU, memory, and storage, and the criteria for evaluating them are shown in Table 4

LU described above to evaluate the arithmetic processing speed of the CPU Find the inverse matrix using decomposition and extract the time required to process this computation. In this specification LU The process of obtaining the inverse matrix using decomposition is functionalized and has the following equation.

는 하 삼각행렬 L의 역행렬을,

는 상 삼각행렬 U의 역행렬을 의미한다. 행렬의 역행렬은 inverseMat() 함수를 이용하여 구할 수 있다.The above equation is a process of decomposing a matrix A into a lower triangular matrix L through a lowerMat () function and an upper triangular matrix U through an upperMat () function and obtaining an inverse matrix

Represents the inverse matrix of the lower triangular matrix L ,

Denotes an inverse matrix of an upper triangular matrix U. The inverse matrix of a matrix can be obtained by using the inverseMat () function.

을 구하는 식은 다음과 같다. When the inverse of L and U is obtained, the inverse of matrix A can be obtained through the product of the two matrices. The inverse of the matrix A using the inverse of the lower triangular matrix L and the upper triangular matrix U

Is obtained as follows.

와

을 기준으로 CPU의 성능

을 정의한다.In order to evaluate the operation processing of the CPU, the time required to process and process the above operation directly at the node

Wow

CPU performance based on

.

는 리소스가 가지는 최대 메모리 크기,

는 사용 중인 메모리 크기로

은 현재 사용 가능한 메모리 크기로 정의된다.In this way, the performance of the memory can also be evaluated. The dynamic memory allocation is used to obtain the matrix space needed to obtain the inverse matrix and define the maximum size that can be allocated for the dynamic memory allocation to be the maximum size that the node memory can allocate. This process can determine the actual available memory space, excluding the memory that the OS is using, and it can know the processing time of the memory through the time required to allocate the dynamic memory.

Is the maximum memory size of the resource,

Is the amount of memory in use.

Is defined as the current available memory size.

Finally, the node performance analyzer 11 analyzes the performance of each node through the performance analysis as described above, and relays the nodes to reflect the performance of the nodes. In the present invention, it is possible to select a node optimal for assigning and assigning a predetermined weight to CPU, memory, and storage determined as the performance of the node. The weighted attributes are the number of cores of the CPU, the processing speed of the CPU, and the remaining amount of memory.

)와 현재 노드의 성능을 비교한 후 각 속성별로 가중치를 부여한다.For the relative evaluation of the nodes, the highest high performance value of all the nodes under management in the data center 10 (

) Is compared with the performance of the current node, and weight is assigned to each attribute.

는 데이터센터에서 관리 중인 노드들의 코어 수 중 가장 큰 값을 가지는

에 상대적인 현재 노드의 코어 수를 수치화하여 가진다. 예를 들어,

Has the largest number of cores in the nodes being managed in the data center

The number of cores of the current node relative to the number of cores. E.g,

과

의 4 이므로, The node with the largest number of cores

and

Of 4

의 코어의 수를 수치화한 값은. According to the proposed formula

The numerical value of the number of cores of

And the numerical values of the remaining nodes are obtained as follows.

The same algorithm applies to the number of cores as well as the CPU's processing speed and memory footprint.

When each node obtains a numerical value for each attribute, it assigns a weight to finally quantify the performance of the node.

The following experiments were conducted to determine the optimal weight for the analysis of the correct node.

Table 5 shows the case in which the difference in weight ratio by attribute is at least 0.5. It is assumed that there is a difference in the performance of all nodes, that Case 1 has memory, Case 2 has processing speed, and Case 3 has a large number of cores.

Host # 1 has relatively fast CPU processing speed but fewer cores. Memory is also relatively small, making it suitable for handling large amounts of virtual machines to process. Host # 3 has 8 cores, though CPU processing speed is slow as opposed to Host # 1. This allows you to allocate multiple lightweight virtual machines, making them ideal for entry-level virtual machines. Table 6 shows the node performance table of Table 7 which shows the results of case-by-case node evaluation when the weight ratio difference is 0.5 or more.

In Case 2, Host # 4 has a high percentage of CPU processing speed, but has not received a high score because its processing speed is relatively slow compared to other nodes. However, if you apply Case 1, you will get a high score of 0.925 because you have 8 cores. The graph on the right in Table 7 shows how the node was evaluated per case.

As a result of the experiment, it is confirmed that the fair performance evaluation is difficult to proceed because the weight difference ratio is too large. In Case 2, which has a high rate of CPU processing speed, the memory and core of Host # 1 are significantly shorter than Host # 4, but they are recognized as a better node and can not efficiently use a large amount of core and memory. .

In order to reduce the problems caused by excessive ratio difference, in Table 8, the experiment was performed in which the weight ratio difference was reduced to at least 0.3. The experimental results are shown in Table 10.

It can be seen that the difference of the node evaluation result is reduced compared to the previous experiment. By reducing the difference in the evaluation results, the weight can be applied stably to changes in the environment. In Table 7, since the difference in the ratio of the weights is large, when there is a change in the resource, the value to be changed is large, and therefore, the result is unstable. In Table 10, in which the ratio difference is reduced, since the difference in the ratio of the weights is small, the value to be changed is not large even if the resource is changed, so that the stable evaluation result can be maintained. This can reduce node fragmentation and provide stable service.

Hereinafter, a virtual machine allocation method for efficiently utilizing multiple resources of the present invention using the system configured as described above will be described.

2 is a flowchart illustrating a process of allocating a virtual machine according to an embodiment of the present invention.

When the cloud service platform is started, the system 1 searches all the nodes 12 in the data center 10 connected to the NC for performance analysis of the nodes 12 managed in the data center 10 (S201) .

와

의 차이를 이용하여 CPU의 성능을 평가한

에 저장한다.

는 i번째 노드의 성능을 분석결과에 가중치를 부여하여 수치화한 값으로, 이는 노드 성능 분석 데이터베이스(21)에 저장된다.Here, an algorithm for analyzing the performance of a node is shown in FIG. In the algorithm of FIG. 3, the performance of a node is analyzed through an InverMatrix () method having a process of obtaining an inverse matrix of a matrix. InvertMatrix () checks memory availability to store the matrix and inverse matrix in order to evaluate the memory capability, and creates an arbitrary matrix in the allocated memory. Through this process, the capacity of the memory and the loading speed can be evaluated. A node loaded with an arbitrary matrix uses the computational processing of the CPU to obtain the inverse matrix of the matrix. When the operation is completed, the computation speed and accuracy can be known, and using Times (), which returns the current time,

Wow

To evaluate the performance of the CPU

.

Is a value obtained by weighting the performance of the i < th > node by weighting the analysis result, and this value is stored in the node performance analysis database 21. [

As described above, in the first execution of the platform, the evaluation of the memory capability and the performance of the CPU are applied to all the nodes in step S201, and the returned CPU performance and memory performance are loaded (S202). The stored result is used to select a node with high performance among the lists of nodes that can be allocated to a virtual machine when a resource request is received in the future, and reflects efficiency by allocating a virtual machine to the selected node.

After the system is started and the performance analysis of the node is completed, the user's request is ready to be processed. When a request for a resource is received from a user (S203), a virtual machine assignment control system 23 for creating a virtual machine and managing allocation, migration, and the like generates a virtual machine through virtualization of resources, S204) A node suitable for allocating a virtual machine in the node list is selected (S205) and a virtual machine is allocated to the optimal node (S206).

At this time, an algorithm for finding an optimal node in the node list is shown in FIG. In the algorithm of FIG. 4, the first for statement prepares availableNodeList, which is a list of nodes that can satisfy the user's request and reflect the current status and whether the request is satisfied or not. N is all the nodes managed by the data center 10. After all the nodes are searched and the assignable list is determined, it is sorted for high performance and fast node selection.

Then, when the virtual machine is created, it assigns the virtual machine to the first node among the registered nodes in the availableNodeList, and stores the virtual machine assigned to the selected node in the vmtable having the virtual machine allocation information. If the virtual machine is not assigned to a node, assign the virtual machine to the next node registered in availableNodeList.

Hereinafter, the validity of the virtual machine allocation system 1 proposed by the present invention is verified through a comparative experiment with an existing allocation system.

In the experimental example of the present invention, the cloud computing emulator developed by the Laboratory of Computer Science and Software Engineering of Melbourne University and the Could Computing and Distributed System (CLOUDS) laboratory of Australia for comparison with various virtual machine allocation systems and a large amount of virtual machine environment In-cloud sim (CloudSim). CloudSim aims to enable modeling, simulation and experimentation of cloud computing infrastructure and application services and to provide a scalable simulation framework. Table 11 describes the cloud simulation functions.

CloudSim supports modeling and simulation of large cloud computing data centers and supports customization policies and virtual server node modeling and simulation for provisioning node resources to virtual machines. In addition, since it supports a virtual machine for assigning node resources to a virtual machine and a user-defined policy for node allocation, it is suitable for experimenting with the virtual machine allocation system proposed in this paper. Table 12 shows the cloud seam layer architecture.

In the experimental example of the present specification, nodes require different experiments in the environment of the resources of the nodes or the constraints of the performance of the resources. Experiments were carried out by dividing all nodes into the same performance and different performance.

Another system that is experimentally compared with the virtual machine allocation system according to an embodiment of the present invention is the round robin used, which is basic scheduling provided in the cloud core and eucalyptus basic scheduling.

The basic scheduling provided in the cloud shim assigns virtual machines to the nodes that have a large number of unused hosts in the node when the virtual machine resource request is received.

Round robin assigns virtual machines to nodes that can resolve resource requests of virtual machines in the order of the nodes set in the system.

Experimental Example  1: All nodes have different performance

Experimental Example 1 is a case where all nodes have different performance. In order to prove how efficiently the system proposed by the multiple resources uses resources, in the case where the performance of the node is different, it is compared with the cloud core basic scheduling, round robin I have experimented.

In the first experiment, the performances of the virtual machines requesting allocation are set to the same although the performance of the node 5 is different. Table 13 shows the node performance of the first experiment.

Host # 1 and Host # 4, Host # 2 and Host # 5 have the same number of cores, but the processing power of the CPU is about twice that of the core. 3 times or more. In order to prove the effect of the performance analysis method of the node described above, the ambiguous case for evaluating the eyes is set.

Table 14 shows the performance of 20 virtual machines with the same performance that request allocation to nodes in the first experiment.

In the second experiment, not only the performance of the node 5 but also the amount of resources required by the virtual machine are different. The performance of the five nodes is the same as the first experiment, and the resource amounts requested by the 20 virtual machines are shown in Table 15.

In the third experiment, the number of nodes and the number of virtual machines requesting resources are increased in the second experiment. Table 16 shows the performance of the added nodes, and Table 17 shows the resource amounts requested by the virtual machines.

Experimental Example  2: All nodes have the same performance

Experimental Example 2 assumes that all nodes have the same performance.

Many scheduling studies for assigning virtual machines to nodes use a small number of nodes and virtual machines, so we construct an experimental environment based on the performance of most nodes. In the present invention, the purpose of maximizing the advantage according to the performance difference of the resources is maximized. However, in order to compare with other researches, all the nodes have the same performance. Node performance of the fourth experiment is shown in Table 18, and the virtual machine request resource is the same as the virtual machine request resource of the second experiment

In the present invention, it is proved that a virtual machine allocation success rate and a fragmentation minimization in a cloud computing environment are compared with a virtual machine allocation system used in an actual cloud platform system when all the nodes have the same performance as those having different performance do. The virtual machine allocation system compared with the virtual machine allocation system proposed in the present invention is a round robin which is a virtual machine allocation system based on FCFS and a virtual machine allocation system of Eucalyptus provided in the cloud core.

Table 19 shows the distribution of virtual machines that have been successfully allocated to nodes by each scheduling result.

Since the basic scheduler of the cloud core places importance on the number of unused cores in the node, it can be confirmed that Host # 5 and Host # 2 are assigned preferentially. On the other hand, round robin is allocated in the order of Host # 5 → # 4 → # 3 → # 2 → # 1, which is the order in which nodes are registered, and can not be allocated to nodes that can not satisfy the virtual machine requirements since VM # 13 . On the contrary, the proposed method analyzes the performance of the nodes based on the performance analysis method of the node of the present invention, and confirms that the virtual machine is preferentially allocated to Host # 2 which is the highest evaluated node among the nodes .

Table 20 shows the ratio of successfully processed virtual machine allocation requests for each scheduling.

In the first and fourth experiments, the virtual machine allocation success rate of the three virtual machine allocation systems was the same because there was little difference in the performance of the virtual machines requiring resources. However, in the case of the second experiment in which the performance of the virtual machines is different, it is confirmed that the virtual machine allocation system proposed by the present invention has a higher success rate than the round robin. This explains that the processing of more requirements has increased the amount of requirements that can be efficiently used to process the nodes.

Table 21 and Table 22 show the number of virtual machine allocations per node. The nodes are the nodes used in the first experiment. Host # 4 has one core, but with a CPU performance score of 2000, it can quickly handle more than two cores of Host # 3 and a CPU performance score of 300. In the case of the cloud virtual machine allocation system and round robin, it is impossible to process requests quickly because the virtual machines are assigned to Host # 3 with a higher number of cores, regardless of the fast processing speed. However, since the proposed method determines the node to allocate the virtual machine to reflect not only the number of cores of the node but also the processing speed and memory performance of the CPU of the node, it is confirmed that the number of virtual machine allocation of Host # 4 is larger . This allows for faster processing of requirements than the virtual machine allocation system and round robin of the cloud core, and even more virtual machine requirements for the same amount of time. It also shows that the node is used in a balanced manner and the load of the node is reduced.

Table 23 compares the amount of memory remaining in the node after processing the resource requests of the virtual machines of the proposed scheme and the virtual machine allocation scheme of the cloud sim that have almost the same success rate of virtual machines.

As with the other results, the first experiment and the fourth experiment, which have no difference in the performance of the virtual machines, are the same as those of the virtual machine allocation system and the proposed technique. In the case of the second experiment with the performance difference of the virtual machine and the third experiment, the difference between the two allocation systems is shown. In the second experiment, compared with the cloud sim virtual machine allocation system, the proposed technique not only leaves more memory but also collects the remaining memory amount so that more virtual machine requests can be processed. In this way, the virtual machine allocation scheme proposed in the present invention minimizes the fragmentation of the memory of the node, thereby making it possible to efficiently utilize the performance of the node.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the invention.

1: Virtual Machine Allocation System
10: Data center
11: Node Performance Analyzer
12: Node
20: Service Provider Hypervisor
21: Node Performance Database
22: Data center broker
23: Virtual Machine Allocation Control System
24: Virtual Machine Migration Controller
25: Virtual Machine Creation Controller
30: Cloud service provider

Claims (4)

In a virtual machine allocation system,
Wherein the virtual machine assignment system comprises a service provider hypervisor and a data center,
The service provider hypervisor,
Node performance database,
A data center broker having a node list retrieved from the node performance database and
And a virtual machine allocation control system including a virtual machine creation controller for creating a virtual machine in a node optimal for allocating a virtual machine in the node list and a virtual machine migration controller for controlling migration of the virtual machine,
The data center comprises:
Nodes and
And a node performance analyzer for analyzing and evaluating the performance of the nodes and storing the nodes in the node performance database,
Wherein the node performance analyzer randomly generates a matrix for analyzing the performance of the nodes and extracts an operation time spent by the nodes to obtain an inverse matrix of the matrix, Respectively, so that the virtual machines are first allocated in the order of the nodes having the short computation time,
Wherein the data center broker comprises:
Virtual machine allocation system. The method of claim 1,
Wherein the node performance analyzer comprises:
Allocating a predetermined weight to a CPU, memory, and storage determined as the performance of the nodes to select the optimal node,
The weight is given in accordance with the number of cores of the CPU, the computation processing speed of the CPU, and the remaining amount of memory,
Virtual machine allocation system. Searching for nodes in the data center by the service provider hypervisor for performance analysis of the nodes;
Assigning weights to the results of the performance analysis and storing the weights in a node performance analysis database;
When a virtual machine allocation request is received from a user, the data center broker inquires node information from a node list loaded from the node performance analysis database;
The service provider hypervisor selecting an optimal node for allocating a virtual machine in the node list; And
And allocating a virtual machine to the optimal node by the virtual machine allocation control system,
Wherein the node performance analyzer randomly generates a matrix for analyzing the performance of the nodes and extracts an operation time spent by the nodes to obtain an inverse matrix of the matrix, And assigning the virtual machines to the nodes in the order of the nodes having the shortest computation time,
Virtual machine allocation method. 4. The method of claim 3,
The step of searching for the nodes comprises:
The inverse matrix of the matrix is obtained, and the time required for each node to process a given operation is extracted to analyze the performance of the nodes.
Virtual machine allocation method.

Images