CN113778630B - A Virtual Machine Migration Method Based on Genetic Algorithm - Google Patents

Abstract

Virtualization has become a support technology for cloud computing that enables a large number of third party applications to run in the form of loading virtual machines. To reduce the running cost, the virtual machine may be migrated to run on other servers. Virtual machine migration introduces additional data transmission overhead, and in addition, different virtual machines originally having a dependency relationship, therefore, reducing network cost becomes an important target considered in virtual machine migration. In the existing research, for migration cost, only migration data amount and transmission bandwidth are considered, dependency relationship among VMs and topology logic of a data center are not considered, so that two VMs with the dependency relationship may be migrated to a host computer far away from each other during VM migration, thereby generating larger communication cost. The invention provides a VM migration cost model by considering migration cost and communication cost during VM migration, in particular to a virtual machine migration method based on a genetic algorithm, which solves the VM migration problem by using the genetic algorithm. By using a wheel disc selection operator, a double-point crossover operator and a bit trigger mutation operator and determining parameters such as crossover rate, mutation rate and the like through repeated experiments, the method can greatly reduce the cost during VM migration.

Description

A Virtual Machine Migration Method Based on Genetic Algorithm

technical field

The invention relates to the field of Internet cloud computing, in particular to a genetic algorithm-based virtual machine migration method.

Background technique

Virtualization has become the supporting technology of cloud computing, which enables a large number of third-party applications to run in the form of virtual machines. In order to reduce operating costs, virtual machines can be migrated to other servers to run. Virtual machine migration will bring additional data transmission overhead, plus the network communication traffic of different virtual machines that originally have dependencies (for example, a multi-tier application composed of multiple virtual machines), so reducing network costs becomes virtual machine Important goals to consider when migrating.

Virtual machine migration is an NP-hard problem. So far, researchers have done a lot of research on the cost model of VM migration from the perspectives of migration cost and communication cost. Verma et al. [1] found that the migration cost is independent of load and only related to VM characteristics. Migration cost is estimated as the drop in throughput due to migration. Liu et al. [2] studied the performance overhead and energy overhead of VM migration, and verified that the migration energy consumption is proportional to the size of the network traffic caused by the migration through experiments. Mann et al. [3] proposed a VM migration framework, which considered the underlying network topology logic of the data center and the network communication traffic requirements between VMs. Meng et al. [4] proposed a two-layer approximation algorithm to solve the traffic-aware VM mapping problem. Jayasinghe et al. proposed a VM migration algorithm that satisfies availability constraints and communication constraints. Shrivastava et al. [5] formalized the data center as a model considering the dependencies between VMs and the underlying network topology, and proposed a greedy algorithm to minimize communication costs. Chen et al. [6] proposed an algorithm that considered network links and node loads to improve traffic load and resource utilization during migration. In the current research, the migration cost and communication cost are studied separately. When studying the migration cost, only the migration data volume and transmission bandwidth are considered, and the dependencies between VMs and the topology logic of the data center are not considered. It is possible to migrate two VMs with dependencies to distant hosts, resulting in large communication costs.

The invention proposes a VM migration cost model including migration cost and communication cost, and uses a genetic algorithm to solve the VM migration problem.

references:

[1] A.Verma, P.Ahuja, A.Neogi, PMapper: Power and migration cost aware application placement in virtualized systems, in: Proc. of the 9th ACM/IFIP/USENIX International Conference on Middleware, Middleware'08, 2008, pp. 243–264.

[2] H.Liu, C.Xu, H.Jin, J.Gong, X.Liao, Performance and energy modeling for live migration of virtual machines, in: Proc. of the 20th International Symposium on High Performance Distributed Computing, HPDC'11 , 2011, pp.171–182.

[3] V.Mann, A.Kumar, P.Dutta, S.Kalyanaraman, VMFlow: Leveraging VMmobility to reduce network power costs in data centers, in: Proc. of the 10th International IFIP TC 6Conference on Networking—Volume Part I, Networking '11, 2011, pp.198–211.

[4] D. Jayasinghe, C. Pu, T. Eilam, M. Steinder, I. Whalley, E. C. Snible, Improving performance and availability of services hosted on IaaS clouds with structural constraint-aware virtual machine placement, in: Proc. the 8th IEEE International Conference on Services Computing, SCC'11, 2011, pp.72–79.

[5] D. Jayasinghe, C. Pu, T. Eilam, M. Steinder, I. Whalley, E. C. Snible, Improving performance and availability of services hosted on IaaS clouds with structural constraint-aware virtual machine placement, in: Proc. the 8thIEEE .International Conference on Services Computing, SCC'11, 2011, pp.72–79.

[6] J. Chen, W. Liu, J. Song, Network performance-aware virtual machine migration in data centers, in: Proc. of the Third International Conference on Cloud Computing, GRIDs, Cloud Computing 2012, 2012, pp.65–71 .

Contents of the invention

In order to overcome the deficiencies of the prior art, the present invention proposes a method for virtual machine migration based on genetic algorithm, which more comprehensively models the cost of virtual machine migration; uses genetic algorithm to solve the problem of virtual machine migration to ensure the diversity of understanding and coverage. The present invention solves its technical problem through the following technical solutions: a virtual machine migration method based on genetic algorithm, comprising the following steps:

(1) Initialize host capacity, virtual machine load requirements, distance between hosts and dependencies between virtual machines;

(2) Initialize the population. Each solution in the population is a randomly generated solution that satisfies capacity constraints; calculate the cost of each solution and the total cost of the population, the cost includes migration cost and communication cost, and select the optimal solution of the population; the migration cost of the solution It can be expressed as: Cost_Mig=∑ _i∈V ∑ _k∈S Size _i ×D _lk ×X _ik , where Size _i represents the size of virtual machine i, D _lk represents the distance of virtual machine migration from host l to host k, and X _ik means that virtual machine i is migrated to physical machine k; the communication cost of the solution can be expressed as: Among them, Cost(V _i ,S _k ,V _j ,S _l )=Distance(S _k ,S _l )×W(V _i ,V _j ), and Distance(S _k ,S _l ) represents the host S _k and S _l The distance between, W(V _i , V _j ) represents the transmission traffic between virtual machines V _i and V _j ;

(3) Perform iterations, and if the number of iterations reaches the specified maximum number of iterations or the optimal solution cannot be improved within several generations, the iteration ends. Contains the following substeps:

(3.1) Use the roulette selection operation to select two solutions from the population as the parent solutions;

(3.2) Execute the two-point crossover operator on the parent solution with a certain crossover probability to generate two child solutions;

(3.3) Verify the validity of the cross operation, including the following sub-steps:

(3.3.1) Whether the child solution satisfies the capacity constraints, if yes, return the child solution; otherwise, execute step (3.3.2); (3.3.2) repeat steps (3.2)-(3.3) until the child solution is effective solution;

(3.4) Execute the bit-triggered mutation operator on the child solution with a certain mutation probability;

(3.5) Verify the validity of the mutation operation, including the following sub-steps:

(3.5.1) Whether the child solution satisfies the capacity constraints, if yes, return the child solution; otherwise, perform step (3.5.2); (3.5.2) repeat steps (3.4)-(3.5) until the child solution is effective solution;

(3.6) Repeat (3.1)-(3.5) until the number of new generation populations reaches the specified requirements;

(4) Calculate the cost of each solution in the new population and the total cost of the population, and find the solution with the lowest cost as the optimal solution for this iteration;

(5) If the optimal solution of this iteration is better than the global optimal solution, update the global optimal solution;

(6) Repeat (3)-(5) until the end condition is met;

(7) The global optimal solution obtained in step (5) is used as the final virtual machine migration scheme;

(8) Implement virtual machine migration. Migrate the virtual machine to the target host according to the optimal virtual machine migration scheme obtained in step (7);

In step (2), the solution in the population does not map all virtual machines to hosts, but reallocates the target hosts for those virtual machines that need to be migrated, thus shortening the length of each chromosome (solution);

In step (3.1), generate a uniformly distributed pseudo-random number r in the interval [0, 1], calculate the cumulative probability q _i of each solution, select the first individual i whose q _i >r, and repeat the steps Step (3.1) is performed twice, and a total of two child individuals are selected;

In step (3.2), use the two-point crossover operator to randomly set two crossover points in the two paired individual code strings, and exchange the partial chromosomes of the two individuals between the two set crossover points;

In step (3.4), the bit-triggered mutation operator is used to mutate, a gene to be mutated is randomly selected in the chromosome, a host number is randomly generated, and the gene is mutated into a new host number.

Description of drawings

Figure 1 Virtual machine migration algorithm flow chart

Detailed ways

(1) Initialize host capacity, virtual machine load requirements, distance between hosts and dependencies between virtual machines;

(2) Initialize the population. Each solution in the population is a solution that satisfies the capacity constraints and is randomly generated. Calculate the cost of each solution (including migration cost and communication cost) and the total cost of the population, and select the optimal solution of the population. Solution; the migration cost of the solution can be expressed as: Cost_Mig=∑ _i∈V ∑ _k∈S Size _i ×D _lk ×X _ik , where Size _i represents the size of virtual machine i, and D _lk represents the migration of virtual machine from host l to The distance of host k, Xi _ik represents the migration of virtual machine i to physical machine k; the communication cost of the solution can be expressed as: Among them, Cost(V _i ,S _k ,V _j ,S _l )=Distance(S _k ,S _l )×W(V _i ,V _j ), and Distance(S _k ,S _l ) represents the host S _k and S _l The distance between, W(V _i , V _j ) represents the transmission traffic between virtual machines V _i and V _j ;

(3) Perform iterations, and if the number of iterations reaches the specified maximum number of iterations or the optimal solution cannot be improved within several generations, the iteration ends. Contains the following substeps:

(3.1) Use the roulette selection operation to select two solutions from the population as the parent solutions;

(3.2) Execute the two-point crossover operator on the parent solution with a certain crossover probability to generate two child solutions;

(3.3) Verify the validity of the cross operation, including the following sub-steps:

(3.3.1) Whether the child solution satisfies the capacity constraints, if yes, return the child solution; otherwise, execute step (3.3.2);

(3.3.2) Repeat steps (3.2)-(3.3) until the child solution is a valid solution;

(3.4) Execute the bit-triggered mutation operator on the child solution with a certain mutation probability;

(3.5) Verify the validity of the mutation operation, including the following sub-steps:

(3.5.1) Whether the child solution satisfies the capacity constraint condition, if yes, return the child solution; otherwise, execute step (3.5.2);

(3.5.2) Repeat steps (3.4)-(3.5) until the child solution is a valid solution;

(3.6) Repeat (3.1)-(3.5) until the number of new generation populations reaches the specified requirements;

(4) Calculate the cost of each solution in the new population and the total cost of the population, and find the solution with the lowest cost as the optimal solution for this iteration;

(5) If the optimal solution of this iteration is better than the global optimal solution, update the global optimal solution;

(6) Repeat (3)-(5) until the end condition is met;

(7) The global optimal solution obtained in step (5) is used as the final virtual machine migration scheme;

(8) Implement virtual machine migration. Migrate the virtual machine to the target host according to the optimal virtual machine migration scheme obtained in step (7).

Claims (5)

1. The virtual machine migration method based on the genetic algorithm is characterized by comprising the following steps of:

(1) Initializing host capacity, virtual machine load requirements, inter-host distance and inter-virtual machine dependence;

(2) Initializing a population, wherein each solution in the population is a randomly generated solution meeting the capacity constraint condition; calculating the cost of each solution and the total cost of the population, wherein the cost comprises migration cost and communication cost, and selecting the optimal solution of the population; the migration cost of a solution can be expressed as: cost_mig= Σ _i∈V ∑ _k∈S Size _i ×D _lk ×X _ik Wherein Size is _i Representing the size of virtual machine i, D _lk Representing the distance, X, that a virtual machine migrates from host l to host k _ik Representing that the virtual machine i is migrated to the physical machine k; the communication cost of the solution can be expressed as:wherein Cost (V _i ,S _k ,V _j ,S _l )＝Distance(S _k ,S _l )×W(V _i ,V _j )，Distance(S _k ,S _l ) Representing a host S _k And S is equal to _l Distance between each other, W (V _i ,V _j ) Representing virtual machine V _i And V is equal to _j Transmission flow between them;

(3) And (3) iterating, wherein if the iteration times reach the specified maximum iteration times or the optimal solution cannot be improved in a plurality of generations, the iteration is ended, and the method comprises the following substeps:

(3.1) selecting two solutions from the population as parent solutions using a roulette selection operator;

(3.2) executing a double-point crossover operator on the parent solution with a certain crossover probability to generate two child solutions;

(3.3) verifying the validity of the interleaving operation, comprising the sub-steps of:

(3.3.1) if the child solution meets the capacity constraint condition, if so, returning the child solution; otherwise, executing the step (3.3.2);

(3.3.2) repeatedly performing the steps (3.2) - (3.3) until the child solution is a valid solution;

(3.4) performing a position trigger mutation operator on the child solution with a certain mutation probability;

(3.5) verifying the effectiveness of the mutation operation, comprising the sub-steps of:

(3.5.1) if the child solution meets the capacity constraint condition, if so, returning the child solution; otherwise, executing the step (3.5.2);

(3.5.2) repeatedly performing the steps (3.4) - (3.5) until the child solution is a valid solution;

(3.6) repeatedly performing (3.1) - (3.5) until the number of new generation populations generated reaches the specified requirement;

(4) Calculating the cost of each solution in the new population and the total cost of the population, and finding out the solution with the lowest cost as the optimal solution of the iteration;

(5) If the optimal solution of the iteration is better than the global optimal solution, updating the global optimal solution;

(6) Repeating (3) - (5) until the end condition is met;

(7) Taking the global optimal solution obtained in the step (5) as a final virtual machine migration scheme;

(8) And (3) implementing virtual machine migration, and migrating the virtual machine to the target host according to the optimal virtual machine migration scheme obtained in the step (7).

2. The method of claim 1, wherein in step (2), the solutions in the population do not map all virtual machines to hosts, but reassign target hosts to those virtual machines that need to be migrated, thus shortening the length of each chromosome (solution).

3. The method of claim 1, wherein in step (3.1), at [0,1]Generating a uniformly distributed pseudo-random number r in the interval, and calculating the cumulative probability q of each solution _i The first is selected to make q _i >r, repeating the step (3.1) twice, and selecting two child individuals altogether.

4. The virtual machine migration method based on genetic algorithm according to claim 1, wherein in step (3.2), two crossover points are randomly set in two individual code strings paired with each other using a double crossover operator, and partial chromosomes of the two individuals between the set two crossover points are swapped.

5. The method according to claim 1, wherein in step (3.4), mutation is performed by using a bit-triggered mutation operator, a gene to be mutated is randomly selected from the chromosome, a host number is randomly generated, and the gene is mutated into a new host number.