CN109800059B - A cloud computing virtual machine migration method based on load curve similarity - Google Patents

Abstract

The invention proposes a cloud computing virtual machine migration method based on the similarity of the load curve, obtains the remaining resource utilization curves of all physical machines and the resource utilization curves of all virtual machines to be migrated, and obtains according to the Pearson correlation coefficient formula. The similarity vector is composed of the similarity of each type of resources, and finally the Euclidean distance is calculated according to the similarity vector and the average resource utilization to determine the virtual machine and the target physical machine to be migrated. The present invention takes the load of the overloaded physical machine for a period of time as the migration standard, instead of only paying attention to the load situation at the moment of overloading, effectively ensuring that the target physical machine will not be overloaded for a period of time after the migration, thereby reducing the number of migrations. Reduced energy consumption. At the same time, the similarity of the curves is used as one of the migration criteria, so that the virtual machine to be migrated and the target physical machine complement each other as much as possible, so that the load of the target physical machine is more balanced.

Description

A cloud computing virtual machine migration method based on load curve similarity

technical field

The invention belongs to the field of cloud computing and virtualization, and in particular relates to a cloud computing virtual machine migration method based on the similarity of load curves.

Background technique

The initial goal of cloud computing is to strengthen the management of resources, mainly in three aspects: computing resources, network resources and storage resources. The main way to strengthen management is to increase the flexibility of time and space. Time flexibility, that is, the service can be provided at any time when the user needs it. Flexibility of space, that is, providing services according to the amount of resources required by each user. The combination of flexibility of time and space is the elasticity of cloud computing.

To improve the elasticity of cloud computing, many solutions have been proposed. The initial solution was to provide the capabilities of physical devices, that is, increase the amount of hard disk data storage, server memory, increase device network bandwidth, and so on. But this approach has its own limitations, and it does not fully address the requirement of flexibility. Because servers, network equipment, etc. take a lot of time to purchase, it is impossible to obtain them at any time.

So hardware "virtualization" technology came into being. A key feature of virtualization technology is live migration: a running virtual machine is migrated from one physical host to another.

In real life, it is often necessary to migrate an overloaded physical host, but many live migration algorithms only consider the resource usage of the virtual machine on the physical machine at the moment of overload to select the target physical machine. However, the resource utilization of different virtual machines usually keeps changing over time, which may cause the virtual machine to migrate to the target host for a period of time and cause the target host to be overloaded again. It needs to be migrated again, which not only increases the number of migrations, but also Increased power consumption and reduced virtual machine performance.

SUMMARY OF THE INVENTION

In view of the above problems, the present invention proposes a cloud computing virtual machine migration method based on the similarity of load curves. The main idea is to obtain the remaining resource utilization curves of all physical machines and the resource utilization curves of all virtual machines to be migrated. According to the Pearson correlation coefficient formula, a similarity vector composed of the similarity of each type of resource is obtained. Finally, the Euclidean distance is calculated according to the similarity vector and the average resource utilization to determine the virtual machine and the target physical machine to be migrated.

A cloud computing virtual machine migration method based on the similarity of load curves, characterized in that it comprises the following steps:

Step 1, define the physical machine set and the virtual machine set on the overloaded physical machine, and define the resources owned by the physical machine;

Step 2, obtaining the resource utilization matrix of each virtual machine on the overloaded physical machine, and establishing a resource utilization matrix set, representing the set of resource utilization matrices of all virtual machines located on the overloaded physical machine;

Step 3: Calculate the mean and standard deviation of the resource utilization rate of each virtual machine in a period of time, and then normalize all virtual machine resource utilization matrices in the resource utilization matrix set to obtain a standardized matrix;

Step 4, obtain the resource utilization matrix of each physical machine in the same period of time;

Step 5, obtaining the remaining utilization matrix of resources on the physical machine;

Step 6: Calculate the mean and standard deviation of the remaining resource utilization ratio of each physical machine to the resource within a time period, and standardize the remaining resource utilization matrix of the physical machine to obtain a standardized matrix;

Step 7: Calculate the Pearson correlation coefficient for each row of the resource utilization matrix and each row of the remaining resource utilization matrix, respectively, to obtain a similarity vector about each type of resource;

Step 8, according to the similarity vector, obtain the maximum value of the similarity of resources between all pairs of virtual machines and physical machines;

Step 9: Calculate the Euclidean distance of each pair of virtual machines and physical machines to obtain a set of Euclidean distances;

Step 10: In the Euclidean distance set, obtain the overloaded virtual machine and the physical machine when the Euclidean distance takes the minimum value, assuming that the overloaded virtual machine is migrated to the physical machine, and judge whether it is satisfied after the migration. Whether the resource utilization is less than the resource utilization threshold on the physical machine, if so, migrate the virtual machine to the physical machine, and remove the corresponding matrix from the resource utilization matrix set; otherwise, the Euclidean distance is changed from the Euclidean distance Remove from the collection, and repeat step 10;

Step 11, if the overloaded virtual machine is no longer overloaded or the virtual machine resource utilization matrix set is empty, the migration is ended; if not, return to step 2, and reselect the virtual machine to be migrated and the target physical machine.

Further, in step 1, define a physical machine set P={p ₁ ,p ₂ ,...,p _j ,...,p _n }, there are n physical machines in the set, where j represents the jth physical machine, overloaded The virtual machine set V on the physical machine = {v ₁ , v ₂ ,...v _i ..., v _m }, there are m virtual machines in the set V, where i represents the ith virtual machine in the set V, and the physical machines have k Class resources, including CPU, memory, hard disk, etc.

Further, in the step 2, the resource utilization matrix A _i of each virtual machine on the overloaded physical machine is obtained;

where u _itk is defined as the utilization rate of resource _k by virtual machine vi at time t, where each row represents the utilization rate of resource _k by the same virtual machine vi from time 1 to time t, and each column represents a different resource;

The resource utilization matrix set S={A ₁ , . . . , A _i , . . . , _Am } represents the set of resource utilization matrices of all virtual machines located on the overloaded physical machine.

Further, in the step 3, according to formula (1), calculate the average value of the utilization rate of each virtual machine v _i to the resource k within the time t

Calculate the standard deviation σ _ik of the utilization rate of resource _k by each virtual machine vi within time t according to formula (2);

Standardize all virtual machine resource utilization matrices A _i in the resource utilization matrix set S according to formula (3) to obtain standardized matrix X _i ;

Thereby _{, a normalized matrix X i of} resource utilization of each virtual machine vi on the overloaded physical machine is obtained.

Further, in the step 4, obtain the resource utilization matrix B _j of each physical machine p _j in the same period of time t;

where U _jtk is defined as the utilization rate of resource k by physical machine p _j at time t, where each row represents the utilization rate of resource k by the same physical machine p _j from time 1 to time t, and each column represents a different Resources.

Further, in the step 5, the remaining utilization matrix B' _j of the resource k on the physical machine p _j is obtained according to formula (4), where T _jk represents the threshold value of the utilization rate of the resource k on the physical machine p _j :

Further, in the step 6, according to formula (5), calculate the average value of the remaining resource utilization rate of each physical machine p _j to the resource k within the time t

Calculate the standard deviation σ' _jk of the remaining resource utilization ratio of each physical machine p _j to resource k within time t according to formula (6);

Standardize the remaining resource utilization matrix B' _j of the physical machine p _j according to formula (7) to obtain the normalized matrix X' _j :

Thus, the normalized matrix X' _j of the remaining resource utilization ratio of each physical machine p _j is obtained respectively.

Further, in the step 7, according to formula (8), the Pearson correlation coefficient h _ijk is obtained from each row of the matrix X _i and each row of the matrix X'_j;

Obtain the similarity vector H _ij ={h _{ij1 ,h ij2 ,} ...,h _ijk } of the virtual machine v _i and the physical machine p _j with respect to all resources, where h _ijk represents the utilization curve of the resource k of the virtual machine v _i and the physical similarity of the residual utilization curve of resource k of machine p _j .

Further, in the step 8, an ideal similarity vector H ^* is obtained according to formula (9), and each r _k in H ^* represents the maximum value of the similarity of resource k between all pairs of virtual machines and physical machines;

H ^* ={max(h ₁₁₁ ,h ₁₂₁ ,…,h _ij1 ),…,max(h _11k ,h _12k ,…,h _ijk )}={r ₁ ,r ₂ ,…,r _k } (9) .

Further, in the step 9, calculate the Euclidean distance l _ij of each pair of virtual machine v _i and physical machine p _j according to formula (10);

The Euclidean distance set L represents the set of Euclidean distances l _ij of all virtual machine physical machine pairs, L={l ₁₁ ,...,l _1j ,...,l _ij }.

The invention proposes a cloud computing virtual machine migration strategy based on the similarity of the load curve. Compared with the current mainstream virtual machine migration strategy, the main advantage of this strategy is that the load of the overloaded physical machine for a period of time is used as the migration standard. , instead of only focusing on the load situation at the moment of overload, which effectively ensures that the target physical machine will not be overloaded for a period of time after the migration, thereby reducing the number of migrations and energy consumption. At the same time, the similarity of the curves is used as one of the migration criteria, so that the virtual machine to be migrated and the target physical machine complement each other as much as possible, so that the load of the target physical machine is more balanced.

Description of drawings

FIG. 1 is a flowchart of a cloud computing virtual machine migration method based on load curve similarity according to the present invention.

Detailed ways

The technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings.

A cloud computing virtual machine migration method based on load curve similarity, comprising the following steps:

Step 1, define a physical machine set P={p ₁ ,p ₂ ,...,p _j ,...,p _n }, there are n physical machines in the set, where j represents the jth physical machine, the virtual machine on the overloaded physical machine Set V={v ₁ ,v ₂ ,...v _i ...,vm }, there are _m virtual machines in set V, where i represents the ith virtual machine in set V, and physical machines have k types of resources, including CPU , memory, hard disk, etc.

Step 2: Obtain the resource utilization matrix A _i of each virtual machine on the overloaded physical machine.

where u _itk is defined as the utilization rate of resource _k by virtual machine vi at time t, where each row represents the utilization rate of resource _k by the same virtual machine vi from time 1 to time t, and each column represents a different resource.

The resource utilization matrix set S={A ₁ , . . . , A _i , . . . , _Am } represents the set of resource utilization matrices of all virtual machines located on the overloaded physical machine.

Step 3: Calculate the average utilization rate of resource k by each virtual machine v _i in time t according to formula (1).

According to formula (2), the standard deviation σ _ik of the utilization rate of resource k by each virtual machine v _i in time t is calculated.

Standardize all virtual machine resource utilization matrices A _i in the resource utilization matrix set S according to formula (3) to obtain a standardized matrix X _i .

Thereby _{, a normalized matrix X i of} resource utilization of each virtual machine vi on the overloaded physical machine is obtained.

Step 4: Obtain the resource utilization matrix B _j of each physical machine p _j within the same time period t.

where U _jtk is defined as the utilization rate of resource k by physical machine p _j at time t, where each row represents the utilization rate of resource k by the same physical machine p _j from time 1 to time t, and each column represents a different Resources.

Step 5: Obtain the remaining utilization matrix B' _j of the resource k on the physical machine p _j according to formula (4), where T _jk represents the threshold of the utilization rate of the resource k on the physical machine p _j :

Step 6: Calculate the average value of the remaining resource utilization of each physical machine p _j for resource k within time t according to formula (5).

Calculate the standard deviation σ' _jk of the remaining resource utilization ratio of each physical machine p _j to resource k within time t according to formula (6).

The remaining resource utilization matrix B' _j of the physical machine p _j is normalized according to formula (7) to obtain a normalized matrix X' _j .

Thus, the normalized matrix X' _j of the remaining resource utilization ratio of each physical machine p _j is obtained respectively.

Step 7, according to formula (8), obtain the Pearson correlation coefficient h _ijk from each row of the matrix X _i and each row of the matrix X' _j .

Obtain the similarity vector H _ij ={h _{ij1 ,h ij2 ,} ...,h _ijk } of the virtual machine v _i and the physical machine p _j with respect to all resources, where h _ijk represents the utilization curve of the resource k of the virtual machine v _i and the physical similarity of the residual utilization curve of resource k of machine p _j .

Step 8: Obtain an ideal similarity vector H ^* according to formula (9). Each r _k in H ^* represents the maximum value of the similarity of resource k between all pairs of virtual machines and physical machines.

H ^* ={max(h ₁₁₁ ,h ₁₂₁ ,…,h _ij1 ),…,max(h _11k ,h _12k ,…,h _ijk )}={r ₁ ,r ₂ ,…,r _k } (9) .

Step 9: Calculate the Euclidean distance l _{ij of each pair of virtual machine vi and physical machine p j according to} formula (10).

The Euclidean distance set L represents the set of Euclidean distances l _ij of all virtual machine physical machine pairs, L={l ₁₁ ,...,l _1j ,...,l _ij }.

如果满足，则将虚拟机v_i迁移到物理机p_j之上，并从资源利用率矩阵集合S中去除A_i；否则将l_ij从欧氏距离集合L中去除，重新进行步骤10。Step 10: In the Euclidean distance set L, obtain the subscripts i and j when l _ij takes the minimum value; assuming that the virtual machine v _i is migrated to the physical machine p _j , it is judged whether U _jtk <T _jk is satisfied after the migration ,

If satisfied, migrate the virtual machine v _i to the physical machine p _j , and remove A _i from the resource utilization matrix set S; otherwise, remove l _ij from the Euclidean distance set L, and perform step 10 again.

Step 11, if the overloaded virtual machine _Pj is no longer overloaded or the virtual machine resource utilization matrix set S is empty, end the migration; if not, return to step 2 to reselect the virtual machine to be migrated and the target physical machine.

The above descriptions are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above-mentioned embodiments, but any equivalent modifications or changes made by those of ordinary skill in the art based on the contents disclosed in the present invention should be included in the within the scope of protection described in the claims.

Claims (9)

1. a cloud computing virtual machine migration method based on load curve similarity is characterized in that: comprise the steps: Step 1, define the physical machine set and the virtual machine set on the overloaded physical machine, and define the resources owned by the physical machine; Step 2, obtaining the resource utilization matrix of each virtual machine on the overloaded physical machine, and establishing a resource utilization matrix set, representing the set of resource utilization matrices of all virtual machines located on the overloaded physical machine; Step 3: Calculate the mean and standard deviation of the resource utilization rate of each virtual machine in a period of time, and then normalize all virtual machine resource utilization matrices in the resource utilization matrix set to obtain a standardized matrix; Step 4, obtain the resource utilization matrix of each physical machine in the same period of time; Step 5, obtaining the remaining utilization matrix of resources on the physical machine; Step 6: Calculate the mean and standard deviation of the remaining resource utilization ratio of each physical machine to the resource within a time period, and standardize the remaining resource utilization matrix of the physical machine to obtain a standardized matrix; Step 7: Calculate the Pearson correlation coefficient for each row of the resource utilization matrix and each row of the remaining resource utilization matrix, respectively, to obtain a similarity vector about each type of resource; In the step 7, according to formula (8), the Pearson correlation coefficient h _ijk is obtained from each row of the matrix X _i and each row of the matrix X'_j;

Obtain the similarity vector H _ij ={h _{ij1 ,h ij2 ,} ...,h _ijk } of the virtual machine v _i and the physical machine p _j with respect to all resources, where h _ijk represents the utilization curve of the resource k of the virtual machine v _i and the physical similarity of the remaining utilization curve of resource k of machine p _j ; Step 8, according to the similarity vector, obtain the maximum value of the similarity of resources between all pairs of virtual machines and physical machines; Step 9: Calculate the Euclidean distance of each pair of virtual machines and physical machines to obtain a set of Euclidean distances; Step 10: In the Euclidean distance set, obtain the overloaded virtual machine and the physical machine when the Euclidean distance takes the minimum value, assuming that the overloaded virtual machine is migrated to the physical machine, and judge whether it is satisfied after the migration. Whether the resource utilization is less than the resource utilization threshold on the physical machine, if so, migrate the virtual machine to the physical machine, and remove the corresponding matrix from the resource utilization matrix set; otherwise, the Euclidean distance is changed from the Euclidean distance Remove from the collection, and repeat step 10; Step 11, if the overloaded virtual machine is no longer overloaded or the virtual machine resource utilization matrix set is empty, the migration is ended; if not, return to step 2, and reselect the virtual machine to be migrated and the target physical machine. 2. The cloud computing virtual machine migration method based on load curve similarity according to claim 1, characterized in that: in said step 1, a physical machine set P={p ₁ , p ₂ ,...,p _j is defined ,...,p _n }, there are n physical machines in the set, where j represents the jth physical machine, the set of virtual machines on the overloaded physical machine V={v ₁ ,v ₂ ,...v _i ...,v _m }, the set There are m virtual machines in V, where i represents the ith virtual machine in the set V, and the physical machines have k types of resources. 3. a kind of cloud computing virtual machine migration method based on load curve similarity according to claim 1, is characterized in that: described step 2, obtains the resource utilization matrix A _i of each virtual machine on overloaded physical machine;

where u _itk is defined as the utilization rate of resource _k by virtual machine vi at time t, where each row represents the utilization rate of resource _k by the same virtual machine vi from time 1 to time t, and each column represents a different resource; The resource utilization matrix set S={A ₁ , . . . , A _i , . . . , _Am } represents the set of resource utilization matrices of all virtual machines located on the overloaded physical machine. 4. a kind of cloud computing virtual machine migration method based on load curve similarity according to claim 1, is characterized in that: described step 3, according to formula (1), calculate each virtual machine v _i in time t to The mean utilization of resource k

Calculate the standard deviation σ _ik of the utilization rate of resource _k by each virtual machine vi within time t according to formula (2);

Standardize all virtual machine resource utilization matrices A _i in the resource utilization matrix set S according to formula (3) to obtain standardized matrix X _i ;

Thereby _{, a normalized matrix X i of} resource utilization of each virtual machine vi on the overloaded physical machine is obtained. 5. The cloud computing virtual machine migration method based on the similarity of the load curve according to claim 1, wherein in the step 4, the resource utilization matrix of each physical machine p _j within the same time period t is obtained B _j ;

where U _jtk is defined as the utilization rate of resource k by physical machine p _j at time t, where each row represents the utilization rate of resource k by the same physical machine p _j from time 1 to time t, and each column represents a different Resources. 6. The cloud computing virtual machine migration method based on load curve similarity according to claim 1, wherein in said step 5, the remaining utilization rate of resource k on physical machine p _j is obtained according to formula (4) Matrix B' _j , where T _jk represents the threshold for resource k utilization on physical machine p _j :

7. a kind of cloud computing virtual machine migration method based on load curve similarity according to claim 1, is characterized in that: described step 6, according to formula (5), calculate each physical machine p _j within time t Mean remaining resource utilization of resource k

Calculate the standard deviation σ' _jk of the remaining resource utilization ratio of each physical machine p _j to resource k within time t according to formula (6);

Standardize the remaining resource utilization matrix B' _j of the physical machine p _j according to formula (7) to obtain the normalized matrix X' _j :

Thus, the normalized matrix X' _j of the remaining resource utilization ratio of each physical machine p _j is obtained respectively. 8. a kind of cloud computing virtual machine migration method based on load curve similarity according to claim 1, is characterized in that: described step 8, obtains ideal similarity vector H ^* according to formula (9), in H ^* Each r _k represents the maximum similarity of resource k between all virtual machines and physical machine pairs; H ^* ={max(h ₁₁₁ , h ₁₂₁ ,...,h _ij1 ),...,max(h _11k ,h _12k ,...,h _ijk )}={r ₁ ,r ₂ ,...,r _k }(9) . 9. a kind of cloud computing virtual machine migration method based on load curve similarity according to claim 1, is characterized in that: described step 9, calculates each pair of virtual machine v _i and physical machine p according to formula (10) The Euclidean distance l _ij of _j ;

The Euclidean distance set L represents the set of Euclidean distances l _ij of all virtual machine physical machine pairs, L={l ₁₁ ,...,l _1j ,...,l _ij }.

Images