Multi-target virtual machine integration method

Abstract

The invention discloses a multi-target virtual machine integration method, belongs to the field of Internet cloud computing application, and solves the problem that an existing virtual machine integration method cannot optimize more than four targets at the same time. According to the method, a second generation non-dominant sorting genetic algorithm (Non-dominant sorting genetic algorithm II) is used, and meanwhile, the energy consumption is minimized, the performance is reduced, the migration cost is reduced, and the resource waste is reduced. According to the method, a decision is made for the integration of the virtual machine based on a multi-objective optimization result, so that the integration of the virtual machine has better performance in the aspects of energy consumption, SLA violation, migration cost and resource waste, and a good compromise is realized among all objectives, so that the method has higher practical application value.

Description

Technical Field

[0001] The present invention relates to the application in the field of Internet cloud computing, and in particular to a multi-target virtual machine integration method. Background Art

[0002] Virtual machine (VM) consolidation aims to use resources by allocating VMs on the minimum number of physical machines (PMs). Optimizing VM consolidation is important for improving energy efficiency, application quality of service (QoS), and resource utilization. Among the existing research methods, methods [1-4] transform multiple objectives into single-objective problems by weighted sum methods. Among them, [1] uses the adaptive beetle swarm optimization (ABSO) algorithm to solve the multi-objective VM consolidation problem. [2] considers three objectives: energy consumption, the combination of processor-intensive VMs and disk-intensive VMs, and the number of PMs with excessive processor utilization, and uses the simulated annealing (SA) method to solve the VM consolidation problem. [3] takes packing efficiency and reducing the number of active hosts as the goals, and proposes a VM placement method based on the monarch butterfly algorithm. [4] Taking migration cost, server energy consumption, VM creation cost and penalty cost of unsatisfied resource requirements during task execution as optimization objectives, discrete cuckoo optimization (DCO) algorithm is used to solve VM integration problem. Some researchers select one objective as optimization objective and optimize other objectives as constraints. [5] Minimize the number of required PMs under SLA (Service level agreement) constraints. [6] Taking minimizing energy consumption as the goal and CPU, memory and bandwidth requirements as constraints. [7] Propose a security-aware VM integration method with the goal of reducing the security risk of PM and the constraint that the risk increase brought by each VM should not exceed the threshold value. [8] Propose a failure-aware VM integration method with the number of failures as the constraint. Other researchers consider multiple objectives and multiple constraints at the same time. The measure of the quality of multi-objective optimization solutions is different from the measure of the quality of single-objective optimization solutions. The concept of Pareto dominance is widely used in these studies. [9] proposed a dual-objective optimization method and used the fuzzy analytic hierarchy process (AHP) to achieve a compromise between the two objectives.

[10] With the goal of reducing energy consumption and bandwidth consumption, the VM integration problem was solved under SLA constraints.

[11] proposed an ε-dominance-based multi-objective artificial bee colony (ε-MOABC) algorithm to effectively solve the trade-off problem between energy consumption, resource waste, system availability and QoS.

[12] A new method that combines the sine and cosine algorithms with the ant lion algorithm is used to solve the multi-objective VM allocation problem. Reducing energy consumption, reducing resource waste and reducing SLA levels are the three optimization goals.

[13] A VM integration method based on a multi-objective ant colony system algorithm is proposed.

[14] Aiming to achieve a compromise between energy consumption, system performance and SLA compliance, an energy-aware and SLA-aware multi-objective ant colony optimization method is proposed to solve the VM integration problem.

[0003] The above research methods only focus on two or three optimization objectives, and regard other objectives as constraints. For VM consolidation problems, optimization objectives may conflict with each other due to the different interests of the service provider (SP) and different users in the data center. In order to achieve true multi-objective optimization, each objective should be regarded as an objective rather than a constraint, and these objectives cannot be simply weighted and summed to reduce the difficulty of multi-objective optimization, because the importance of each objective is difficult to express with a weight value;

[0004] This paper proposes a multi-objective optimization (MOO) method for VM consolidation problem, which minimizes energy consumption, performance degradation, migration cost and resource waste simultaneously. Experimental results using Microsoft Azure tracking show that compared with existing methods, the proposed method has better performance in terms of energy consumption, SLA violation, migration cost and resource waste, and achieves a good trade-off between all the above objectives.

[0005] References

[0006] [1]Hariharan B,Siva R,Kaliraj S,Prakash,PNSenthil.ABSO:an energy-efficient multi-objective VM consolidation using adaptive beetle swarmoptimization on cloud environment[J].2021.DOI:10.1007 / s12652-021-03429-w.

[0007] [2]Sharifi M,Salimi H,Najafzadeh M.Power-efficient distributedscheduling of virtual machines using workload-aware consolidation techniques[J].Journal of Supercomputing,2012,61(1):46-66.

[0008] [3]Ghetas M.A multi-objective Monarch Butterfly Algorithm for virtualmachine placement in cloud computing[J].Neural Computing and Applications,2021,33(17):11011-11025.

[0009] [4]Tavana M,Shandi-Pashaki S,Teymourian E,Santos-Arteaga,Francisco J,M Komaki.A Discrete Cuckoo Optimization Algorithm for Consolidation in CloudComputing[J].Computers&Industrial Engineering,2018,115(JAN.):495-511.andAdvanced Applications,2013.

[0010] [5]Ferreto,T.C.,a.S.Netto,M.,Calheiros,R.N.,a.F.De Rose,C.ServerConsolidation with Migration Control for Virtualized Data Centers.Journal ofFuture Generation Computer Systems 27(8),1027-1034(Oct 2011).

[0011] [6]Somnath Mazumdar,Marco Pranzo.Power efficient server consolidationfor Cloud data center.Future Generation Computer Systems,Volume 70,2017,4-16.

[0012] [7]Mohamed A.Elshabka,Hanan A.Hassan,Walaa M.Sheta,Hany M.Harb,Security-aware dynamic VM consolidation,Egyptian Informatics Journal,Volume22,Issue 3,2021,Pages 277-284.

[0013] [8]Yogesh Sharma,Weisheng Si,Daniel Sun,Bahman Javadi,Failure-awareenergy-efficient VM consolidation in cloud computing systems,FutureGeneration Computer Systems,Volume 94,2019,Pages 620-633.

[0014] [9]N.Kord and H.Haghighi.An energy-efficient approach for virtualmachine placement in cloud based data centers.Information and KnowledgeTechnology(IKT),2013 5th Conference on.IEEE,2013,pp.44-49.

[0015]

[10] Qing Liao,Zun Wang.Energy Consumption Optimization Scheme ofCloud Data Center Based on SDN.Procedia Computer Science,Volume 131,2018,1318-1327.

[0016]

[11] Sayadnavard M H,Haghighat A T,Rahmani A M.A multi-objectiveapproach for energy-efficient and reliabledynamic VM consolidation in clouddata centers[J].Engineering Science and Technology an International Journal,2021(2).

[0017]

[12] Gharehpasha S,Masdari M.A discrete chaotic multi-objective SCA-ALO optimization algorithm for an optimal virtual machine placement in clouddata center[J].Journal of Ambient Intelligence and Humanized Computing,2020.DOI:10.1007 / s12652-020-02645-0.

[0018]

[13] Yongqiang Gao,Haibing Guan,Zhengwei Qi,Yang Hou,Liang Liu.Amulti-objective ant colony system algorithm for virtual machine placement incloud computing.Journal of Computer and System Sciences,Volume 79,Issue 8,2013,1230-1242.

[0019]

[14] Malekloo M H,Kara N,Barachi M E.An Energy Efficient and SLACompliant Approach for Resource Allocation and Consolidation in CloudComputing Environments[J].Sustainable Computing Informatics&Systems,2018,17(mar.):9-24。 Summary of the invention

[0020] The present invention proposes a multi-objective virtual machine integration method, which simultaneously considers four optimization objectives: energy consumption, resource waste, performance, and migration cost, and uses the second-generation non-dominated sorting genetic II algorithm (NSGA-II) to optimize the four objectives simultaneously. Based on the multi-objective optimization results, decisions are made on virtual machine integration, so that virtual machine integration can take into account the interests of different entities such as SPs and users, thus having more practical application value;

[0021] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0022] A multi-target virtual machine integration method comprises the following steps:

[0023] Step 1) Establish energy consumption model, performance model, migration model and resource waste model corresponding to the four optimization objectives; Step 2) Use NSGA-II to perform multi-objective optimization, including the following sub-steps:

[0024] 2.1. Initialize a population of size Pop_size as the parent generation;

[0025] 2.2. Generate a child using the selection operator, crossover operator and mutation operator;

[0026] 2.3. If the current generation is less than the maximum generation, execute step 2.4, otherwise execute step 2.7;

[0027] 2.4. Obtain Pop_size non-dominant optimal solutions as the new population, including the following sub-steps:

[0028] 2.4.1. Merge the parent generation and the offspring generation as a new population;

[0029] 2.4.2. According to the cost model established in step 1), the population obtained in step 2.4.1 is sorted in a non-dominant manner to obtain several non-dominant frontiers;

[0030] 2.4.3. Calculate the congestion degree of each frontier according to the cost model established in step 1);

[0031] 2.4.4. If the number of solutions in the first frontier is greater than Pop_size, use the crowding comparison operator to sort the frontier in descending order and select the first Pop_size solutions as the new population, otherwise execute step 2.4.5;

[0032] 2.4.5. Take all the solutions from the front several fronts to form part of the new population. The insufficient part is obtained from the next front. The front is sorted in descending order using the crowding comparison operator, and the front several solutions of the front are selected to complete the new population.

[0033] 2.5. Use the selection operator, crossover operator and mutation operator to generate a child for the population generated in step 2.4;

[0034] 2.6. Update the current algebra and go to step 2.3;

[0035] 2.7. Select an optimal solution from the current population;

[0036] Step 3) implementing VM integration using the optimal solution obtained in step 2);

[0037] Further, the power consumption of the PM in step 1) is expressed as: Among them, i represents the i-th host, C i (PS k ) indicates that the i-th host is in PS k Energy consumption in state, α i is the energy consumption coefficient of the host. i is the calibration parameter of the host, and the energy consumption of PM is expressed by the following formula: The energy consumption of all PMs in the data center is expressed as: M is the number of PMs; the energy consumption of state switching is expressed as: Among them, SS ki Indicates whether the kth state switch has occurred on the i-th host. It represents the energy consumption of the kth state switching of the i-th host; the energy consumption of virtual machine migration is expressed as: in, Indicates the energy consumption of migrating virtual machines from the source host. Indicates the energy consumption of the target host when migrating virtual machines. represents the energy consumption generated during the transmission process, α, β are the model parameters that need to be trained, V j is the size of virtual machine j, B is the network bandwidth; the total energy consumption of the data center is expressed as: E(s) = E p (s)+E swit (s)+E mig (s), where s is a solution to the virtual machine consolidation problem, i.e., a mapping from all VMs to PMs.

[0038] Furthermore, the performance model in step 1) is expressed as: SLAV = SLATAH·PDM, where: Here, N is the number of PMs, T si is the total time that the utilization rate of the ith PM reaches 100%, T ai is the total time that the ith PM is in active mode. M is the number of VMs, C dj is the estimated performance degradation of the jth VM caused by migration, C rj is the total CPU capacity requested by the jth VM during its lifetime;

[0039] Furthermore, the migration model in step 1) is expressed as: MC = a·T mig +b·T down , where a and b are weights, T mig and T down Use the following algorithm to calculate:

[0040]

[0041]

[0042] Among them, the input parameter n th Indicates the VM pre-copy iteration threshold, m th represents the VM memory transfer threshold, p th Indicates the threshold value of the number of dirty pages in the previous round of VM pre-copy, v mem represents the memory size of the VM when the migration is started, B represents the bandwidth, d represents the dirty page rate, l represents the page size, and γ represents the correlation coefficient; output parameter V mig Indicates the total amount of memory transferred by the VM during the pre-copy phase;

[0043] Furthermore, the resource waste model in step 1) is expressed as: Where M represents the number of PMs, here, in, represents the capability of resource r owned by PMj, represents the VM's demand for resource r, x ij Indicates whether VMi is assigned to PMj, and N represents the number of VMs.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The present invention considers four optimization goals, namely, energy consumption, performance degradation, VM migration cost, and resource waste, and optimizes them simultaneously, so that the virtual machine placement method provided by the present invention can not only reduce the energy consumption of the data center, but also improve the performance, and reduce the number of VM migrations and resource waste, and has more application value in practical applications;

[0046] 2. The present invention uses the NSGA-II algorithm to generate offspring by using the two operations of non-dominant sorting and crowding calculation based on the selection operator, crossover operator and mutation operator. Compared with the method using the traditional genetic algorithm, the present invention solves the problem of being unable to compare the quality of solutions on multiple objectives, and the diversity and convergence of the solutions obtained by the search are better.

[0047] 3. The present invention uses an elite strategy when selecting offspring. As iterations proceed, the quality of the solution becomes better and better until a non-dominant optimal solution is finally found. Compared with some algorithms that use random strategies, the quality of the solution does not fluctuate randomly or cannot be guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Virtual Machine Integration Flowchart

[0049] Figure 2 Performance comparison with other methods. DETAILED DESCRIPTION

[0050] The present invention is further described in detail below with reference to the accompanying drawings and implementation examples;

[0051] A multi-target virtual machine integration method comprises the following steps:

[0052] Step 1) Establish energy consumption model, performance model, migration model and resource waste model corresponding to the four optimization objectives; Step 2) Use NSGA-II to perform multi-objective optimization, including the following sub-steps:

[0053] 2.1. Initialize a population of size Pop_size as the parent generation;

[0054] 2.2. Generate a child using the selection operator, crossover operator and mutation operator;

[0055] 2.3. If the current generation is less than the maximum generation, execute step 2.4, otherwise execute step 2.7;

[0056] 2.4. Obtain Pop_size non-dominant optimal solutions as the new population, including the following sub-steps:

[0057] 2.4.1. Merge the parent generation and the offspring generation as a new population;

[0058] 2.4.2. According to the cost model established in step 1), the population obtained in step 2.4.1 is sorted in a non-dominant manner to obtain several non-dominant frontiers;

[0059] 2.4.3. Calculate the congestion degree of each frontier according to the cost model established in step 1);

[0060] 2.4.4. If the number of solutions in the first frontier is greater than Pop_size, use the crowding comparison operator to sort the frontier in descending order and select the first Pop_size solutions as the new population, otherwise execute step 2.4.5;

[0061] 2.4.5. Take all the solutions from the front several fronts to form part of the new population. The insufficient part is obtained from the next front. The front is sorted in descending order using the crowding comparison operator, and the front several solutions of the front are selected to complete the new population.

[0062] 2.5. Use the selection operator, crossover operator and mutation operator to generate a child for the population generated in step 2.4;

[0063] 2.6. Update the current algebra and go to step 2.3;

[0064] 2.7. Select an optimal solution from the current population;

[0065] Step 3) implements VM consolidation using the optimal solution obtained in step 2).

[0066] The present invention is run on a PC computer (Core i7, 3.33GHz, 6G RAM) with Windows 10 operating system and simulated using CloudSim environment. The data center consists of 500-2500 virtual machines and 800 physical machines, and is tracked using Microsoft Azur data. The types of virtual machines include four types of virtual machine instances of Amazon EC2. Table 1 shows the CPU capabilities and memory capacities of these four virtual machine instances. The configuration of the physical machine is shown in Table 2;

[0067] Table 1 Virtual machine types

[0068]

[0069] Table 2 Physical machine configuration.

[0070]

[0071] Energy consumption, resource waste, number of VM migrations and SLA violations are selected as performance indicators, and three optimization methods, LR (Local Regression), MAD (Median Absolute Deviation) and IQR (Inter-Quartile Range), are selected as benchmark methods. The comparison between the present invention and the benchmark method in terms of energy consumption, resource waste, etc. is shown in the attached figure of the specification. Figure 2As shown. The results show that the present invention is superior to the benchmark method in terms of energy consumption, performance, number of migrations, etc. This can be attributed to the fact that the NSGA-II algorithm used in the present invention uses the two operations of non-dominant sorting and crowding calculation to generate offspring based on the use of selection operators, crossover operators and mutation operators. This makes the diversity and convergence of the solutions of the present invention better. In addition, the present invention uses an elite strategy when selecting offspring. As the iteration proceeds, the quality of the solution becomes better and better, and there will be no problem of random fluctuations and unguaranteed quality of the solution.

[0072] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A multi-target virtual machine integration method, comprising the following steps: Step 1) Establish energy consumption model, performance model, migration model and resource waste model corresponding to the four optimization objectives; Step 2) Use NSGA-II to perform multi-objective optimization, including the following sub-steps: 2.

1. Initialize a population of size Pop_size as the parent generation; 2.

2. Generate a child using the selection operator, crossover operator and mutation operator; 2.

3. If the current generation is less than the maximum generation, execute step 2.4, otherwise execute step 2.7; 2.

4. Obtain Pop_size non-dominant optimal solutions as the new population, including the following sub-steps: 2.4.

1. Merge the parent generation and the offspring generation as a new population; 2.4.

2. According to the cost model established in step 1), the population obtained in step 2.4.1 is sorted in a non-dominant manner to obtain several non-dominant frontiers; 2.4.

3. Calculate the congestion degree of each frontier according to the cost model established in step 1); 2.4.

4. If the number of solutions in the first frontier is greater than Pop_size, use the crowding comparison operator to sort the frontier in descending order and select the first Pop_size solutions as the new population, otherwise execute step 2.4.5; 2.4.

5. Take all the solutions from the front several fronts to form part of the new population. The insufficient part is obtained from the next front. The front is sorted in descending order using the crowding comparison operator, and the front several solutions of the front are selected to complete the new population. 2.

5. Use the selection operator, crossover operator and mutation operator to generate a child for the population generated in step 2.4; 2.

6. Update the current algebra and go to step 2.3; 2.

7. Select an optimal solution from the current population; Step 3) implements VM consolidation using the optimal solution obtained in step 2).

2. The multi-target virtual machine integration method according to claim 1, characterized in that: The power consumption of PM in step 1) is expressed as: Among them, i represents the i-th host, C i (PS k ) indicates that the i-th host is in PS k Energy consumption in state, α i is the energy consumption coefficient of the host. i is the calibration parameter of the host, and the energy consumption of PM is expressed by the following formula: The energy consumption of all PMs in the data center is expressed as: M is the number of PMs; the energy consumption of state switching is expressed as: Among them, SS ki Indicates whether the kth state switch has occurred on the i-th host. It represents the energy consumption of the kth state switching of the i-th host; the energy consumption of virtual machine migration is expressed as: in, Indicates the energy consumption of migrating virtual machines from the source host. Indicates the energy consumption of the target host when migrating virtual machines. represents the energy consumption generated during the transmission process, α, β are the model parameters that need to be trained, V j is the size of virtual machine j, B is the network bandwidth; the total energy consumption of the data center is expressed as: E(s) = E p (s)+E swit (s)+E mig (s), where s is a solution to the virtual machine consolidation problem, i.e., a mapping from all VMs to PMs.

3. The multi-target virtual machine integration method according to claim 1, characterized in that: The performance model in step 1) is expressed as: SLAV = SLATAH·PDM, where: Here, N is the number of PMs, T si is the total time that the utilization rate of the ith PM reaches 100%, T ai is the total time that the ith PM is in active mode. M is the number of VMs, C dj is the estimated performance degradation of the jth VM caused by migration, C rj is the total CPU capacity requested by the jth VM during its lifetime.

4. The multi-target virtual machine integration method according to claim 1, characterized in that: The migration model in step 1) is expressed as: MC = a·T mig +b·T down , where a and b are weights, T mig and T down Use the following algorithm to calculate: Among them, the input parameter n th Indicates the VM pre-copy iteration threshold, m th represents the VM memory transfer threshold, p th Indicates the threshold value of the number of dirty pages in the previous round of VM pre-copy, v mem represents the memory size of the VM when the migration is started, B represents the bandwidth, d represents the dirty page rate, l represents the page size, and γ represents the correlation coefficient; the output parameter V mig Indicates the total amount of memory transferred by the VM during the pre-copy phase.

5. The multi-target virtual machine integration method according to claim 1, characterized in that: The resource waste model in step 1) is expressed as: Where M represents the number of PMs, here, in, represents the capability of resource r owned by PM j, represents the VM's demand for resource r, x ij Indicates whether VM i is assigned to PM j, and N represents the number of VMs.

Images