Snake optimization budget perception scheduling algorithm based on negative migration mechanism in heterogeneous cloud

Abstract

The invention discloses a snake optimization budget perception scheduling algorithm based on a negative offset mechanism in a heterogeneous cloud, which comprises the following steps: simulating foraging and breeding behaviors of snakes through a mathematical method, designing different updating mechanisms of an exploration mode, a development mode, a fighting mode and a mating mode to search in a solution space, and then combining with the negative offset mechanism to obtain a snake optimization budget perception scheduling algorithm based on the negative offset mechanism. On the basis that certain randomness is reserved, it is ensured that updating of the positions of the individuals is not purposeless random exploration any more, but based on the cost characteristics that the individuals exceed the budget, the individuals are guided to carry out directed position offset in the direction of cost reduction, and therefore the positioning accuracy of the individuals is improved. The problems that in the heterogeneous cloud environment, an existing snake optimization algorithm is prone to falling into a local optimal solution, complexity and uncertainty of a search path are aggravated, and calculation difficulty is improved are solved.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing technology, and more specifically, to a snake-optimized budget-aware scheduling algorithm based on a negative offset mechanism in a heterogeneous cloud. Background Art

[0002] In recent years, cloud computing technology has made rapid progress. Its computing method is cost-effective and easy to use. It has not only revolutionized the way global users share computing resources through the Internet, but also, as a key component of the country's "new infrastructure" strategy, cloud computing technology has injected strong impetus into the development of cutting-edge science and technology fields such as 5G, Internet of Things, and Industrial Internet, becoming an indispensable and important support. With its powerful computing power, cloud centers have been widely used to solve massive data and complex workflow problems in various fields such as biology, physics, and astronomy. Virtualization technology is a key technology in cloud computing, which can realize dynamic scheduling and expansion of resources. Through virtualization technology, a large number of computing nodes are integrated together to form a huge resource pool, and users use the resources in the resource pool in a "pay as you go" manner through the Internet. In heterogeneous cloud computing centers, computing resources have different computing capabilities and characteristics, and users can rent them at any time and in any quantity.

[0003] Based on the above elastic and heterogeneous computing environment, the workflow scheduling algorithm of the cloud center needs to consider how to find the most appropriate mapping relationship from a series of interdependent tasks to a set of heterogeneous virtual resources, so as to minimize or maximize the specified goals that users care about. This is crucial for a sustainable and efficient cloud data center. However, workflow scheduling is an NP-hard problem, and it is currently difficult to achieve the optimal scheduling of workflows. Therefore, people usually explore the workflow scheduling problem under the constraints of user-defined Quality of Service (QoS).

[0004] At present, the mainstream solutions to workflow scheduling problems are mainly heuristics and metaheuristics. With the development of deep reinforcement learning (DRL), DRL-based methods are gradually being applied to workflow scheduling problems. Among them, heuristic algorithms can quickly formulate workflow scheduling strategies based on expert experience, but such algorithms are mainly applied to small-scale workflows, adopt simple rules, are prone to fall into local optimal solutions, and are difficult to find the optimal solution in a dynamic cloud environment. The DRL-based method is still in its infancy. When used to solve large-scale complex application scheduling problems, it still faces challenges such as sparse rewards, hyperparameter sensitivity, and low exploration efficiency. In contrast, metaheuristic algorithms use a combination of random search and local utilization techniques to find approximate optimal solutions. They have good optimization performance and are widely used in large-scale workflows to find approximate optimal scheduling solutions. The Snake Optimization Algorithm (SO) is a new metaheuristic algorithm that simulates the foraging and reproduction behavior of snakes through mathematical methods, designs different update mechanisms to search in the solution space, and solves a series of continuous optimization problems that imitate the special mating behavior of snakes. It is significantly superior to existing technologies in the field of metaheuristics and shows extremely high performance advantages.

[0005] In CN202311463211.4, a cloud resource planning and scheduling method with a global search mechanism and fast convergence is disclosed, which includes the following steps: S1, obtaining the user's initialization parameter data, the parameter data including the task byte size and the virtual machine bandwidth; then establishing the objective function; S2, using the improved snake optimization algorithm to obtain the optimal population; S3, selecting the individual with the smallest objective function value from the optimal population as the optimal resource scheduling scheme. This patent can further improve the rationality of cloud resource scheduling and maximize resource utilization. However, it introduces randomness and enhances diversity to avoid the dilemma of the snake optimization algorithm easily falling into the local optimal solution. Although this promotes the global search capability of the algorithm, it also aggravates the complexity and uncertainty of the search path, thereby increasing the computational difficulty of approaching the global optimal solution. Summary of the invention

[0006] The main technical problem to be solved by the present invention is that the introduction of randomness and enhanced diversity in the prior art to avoid the snake optimization algorithm makes it easy to fall into the local optimal solution, aggravates the complexity and uncertainty of the search path, and increases the difficulty of calculation. When using the snake optimization algorithm to accurately solve the complex workflow scheduling problem with budget cost as the constraint condition and scheduling length as the optimization target, the present invention considers balancing randomness and directionality, and provides a snake optimization budget-aware scheduling algorithm based on a negative offset mechanism in a heterogeneous cloud to achieve dual optimization of search efficiency and solution quality.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A snake optimization budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud, the steps include:

[0009] S1. Generate a random initial population Divide the population individuals into two groups of males and females;

[0010] S2. Start iterative calculation to find the best male individual X in the male group and the female group respectively. m,best and the optimal female individual X f,best ;

[0011] S3. Calculate the iterative food quantity Q and temperature Temp, and calculate the food quantity Q and temperature Temp according to the set food quantity threshold Q ts and temperature threshold T ts Select mode to update position;

[0012] S4. After completing the position update, the offspring population is obtained For offspring population The dimension value x of the individual in i Make adjustments;

[0013] S5. Comparison and The corresponding individuals in the population are selected to form a new population with better individuals.

[0014] S6. Update the population using negative offset mechanism Get the new parent population to enter the next iteration

[0015] S7. Complete the iterative calculation and search for the optimal individual X best .

[0016] Furthermore, the food quantity Q is expressed as:

[0017]

[0018] Temperature Temp is expressed as:

[0019]

[0020] Among them, d is the current iteration number, D is the maximum iteration number, c 1 is a constant equal to 0.5, exp(·) represents the exponential function, Q and Temp represent the amount of food and temperature at the dth iteration, respectively.

[0021] Furthermore, the mode selection for updating the position is:

[0022] If Q<Q ts , then the male individual Xm,r The update mechanism can be expressed as:

[0023]

[0024] in, represents the updated position of the rth male individual, X m,rand represents the position of a random male individual, c 2 is a constant equal to 0.05, X max and X min represent the upper and lower bounds of the solution space respectively, Indicates the exploration ability of male individuals;

[0025] Female individual X f,r The update mechanism can be expressed as:

[0026]

[0027] in, represents the updated position of the rth female individual, X f,rand represents the position of a random female individual, Indicates the exploration ability of female individuals;

[0028] If Q>Q ts ,Temp>T ts , then the male individual X m,r The update mechanism is expressed as:

[0029]

[0030] Female individual X f,r The update mechanism is expressed as:

[0031]

[0032] Among them, X best The parent population The best individual in 3 is a constant equal to 2;

[0033] If Q>Q ts ,Temp<T ts ,rand>0.6, then the male individual X m,r The update mechanism is:

[0034]

[0035] Among them, X f,best is the best female individual among the parent female individuals, Indicates the fighting ability of male individuals;

[0036] Female individual X f,r The update mechanism is:

[0037]

[0038] Among them, X m,best is the best male individual among the parent male individuals. Indicates the fighting ability of female individuals;

[0039] If Q>Q ts ,Temp<T ts , rand<0.6, then the male individual X m,r The update mechanism is:

[0040]

[0041] in, Indicates the mating ability of male individuals;

[0042] Female individual X f,r The update mechanism is:

[0043]

[0044] in, Indicates the mating ability of female individuals.

[0045] Further, The expression is:

[0046]

[0047] The expression is:

[0048]

[0049] Further, The calculation expression is:

[0050]

[0051] The expression is:

[0052]

[0053] Further, The expression is:

[0054]

[0055] The expression is:

[0056]

[0057] Furthermore, the adjustment is: if x i <1, then adjust x i =1; if x i >|V|, then adjust x i =|V|, so that x i Satisfies [1,|V|].

[0058] Further, the negative offset mechanism step includes:

[0059] S51. Calculate the number of random male individuals ξ according to the random probability γ m,rand and the number of female individuals ξ f,rand ;

[0060] S52. From the population The number of choices in the male and female groups is ξ m,rand The number of male individuals is ξ f,rand Female individuals;

[0061] S53. traverse all randomly selected male individuals and female individuals respectively, and shift their positions;

[0062] S531. If the current traversal individual does not meet the budget constraint, the negative relaxation cost of the individual is calculated as

[0063]

[0064] And calculate the minimum cost deviation MCD (n i ) and budget cost bgt (n i );

[0065] S532. Randomly select a dimension value x of an individual h , and then find the offset queue Q of the hth dimension O (h), from Q O In (h), a random VM index is selected to replace x h .

[0066] Furthermore, the minimum cost deviation MCD(n i ) is expressed as:

[0067] MCD(n i )=Cost(n i ,vm k )-Cost min (n i ).

[0068] Cost(ni ,vm k )=w i,k ×price k .Cost min (n i ) is the minimum execution cost.

[0069] Furthermore, the budget cost bgt (n i ) is expressed as:

[0070]

[0071] Furthermore, the offset queue Q O (h) is expressed as:

[0072]

[0073] Compared with the prior art, the beneficial effects are:

[0074] The present invention combines the snake optimization algorithm with the negative offset mechanism. On the basis of retaining a certain degree of randomness, it ensures that the update of individual positions is no longer a random exploration without purpose, but is based on the cost characteristics of the individual exceeding the budget, guiding the individual to perform a directional position offset in the direction of reducing costs. The negative offset mechanism reduces the individual total cost from the perspective of the individual dimension based on the negative relaxation cost of the individual and the minimum cost deviation of the task, speeds up the search speed for feasible solutions that meet the budget constraint in the early stage of iteration, and enables the iteration process to enter the stage of minimizing the scheduling length earlier. At the same time, the negative offset mechanism has randomness, gets rid of the local optimum in the solution update stage to a certain extent, and enhances the search diversification. The negative offset mechanism is to balance randomness and directionality, improve the success rate of searching for feasible solutions that meet the budget constraint while preventing falling into the local optimum, and efficiently approach the global optimal solution. The method (NO-SO) described in the present invention is used to schedule workflows with budget constraints in heterogeneous cloud computing, which not only performs well in the success rate of searching for feasible solutions, but also has obvious effects in reducing the scheduling length. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 It is the flow chart of NO-SO algorithm in Example 1.

[0076] Figure 2 This is a flow chart of the negative offset mechanism in Example 2.

[0077] Figure 3 This is the convergence effect of the four algorithms in Example 3 in four types of workflows. DETAILED DESCRIPTION

[0078] The present invention will be further explained and illustrated below in conjunction with the embodiments, but the specific embodiments do not limit the present invention in any form.

[0079] Example 1

[0080] The resources of a cloud computing center are usually modeled as a set of heterogeneous virtual machine instances, represented by VM = {vm 1 ,vm 2 ,...,vm |V|}, where |V| is the number of virtual machine instances. All virtual machine instances are heterogeneous and have different types and resource configurations. The data communication cost between virtual machine instances can be ignored. For the convenience of description, this embodiment stipulates that the bandwidth of the virtual machine is regarded as the ideal bandwidth, and there will be no communication competition between virtual machine instances.

[0081] The workflow application in the cloud computing system represents a set of tasks with data dependencies in the form of a directed acyclic graph (DAG). Tasks are represented by nodes, and the dependencies between tasks are represented by edges. It can be represented by a four-dimensional vector G = {N, E, C, W}. 1 ,n 2 ,...,n |N|} represents a set of tasks in a workflow, and |N| represents the number of tasks in the workflow. i,j |n i ,n j ∈N} represents the set of communication edges with data dependency in the DAG graph. Each edge represents a task dependency constraint, that is, task n i Should be in task n j Complete its execution before starting. c i,j ∈C represents task n i The relevant data is transferred to task n j The communication time required. If task n i With task n j Assigned to the same virtual machine for execution, the communication time c i,j is zero. i,k ∈W represents task n i In the virtual machine vm k The execution time on pred(n i ) represents task n i The set of direct predecessor tasks, succ(n i ) represents task n i The set of direct successor tasks. In a given workflow graph, a task without a predecessor task is called an entry task n. entry , and the task without a successor task is called an exit task n exit .

[0082] Based on the existing workflow scheduling model, the following definitions of related concepts are given.

[0083] Task i In the allocated virtual machine vm k The start time and completion time of execution are represented as ST(n i ) and FT(n i ), the calculation expressions are:

[0084]

[0085] FT(n i )=ST(n i )+w i,k ,

[0086] Among them, avail(vm k ) indicates a virtual machine vm k Available time. For entry task n entry , then:

[0087] ST(n entry )=avail(vm k ).

[0088] The scheduling length Makespan of workflow G refers to the time required from the start of workflow scheduling to the completion of the workflow execution. Its calculation expression is:

[0089]

[0090] This embodiment provides a snake optimization budget-aware scheduling algorithm based on a negative offset mechanism in a heterogeneous cloud. Figure 1 , the steps include:

[0091] S1. Build cost model;

[0092] Define Task n i In the virtual machine vm k The execution cost required to run on is:

[0093] Cost(n i ,vm k )=w i,k ×price k .

[0094] The total cost Cost(G) of workflow G is the sum of the execution costs of all tasks, which can be expressed as:

[0095]

[0096] Task iThere is a minimum execution cost when executing on different virtual machines min (n i ) and the maximum execution cost Cost max (n i ). Then the minimum cost and maximum cost of workflow G can be expressed as Cost min (G) and Cost max (G), the calculation expressions are:

[0097]

[0098] Assume that the budget cost of workflow G is Cost budget (G) The budget cost must be greater than or equal to the minimum cost of the workflow. min (G) and is less than or equal to the maximum cost of the workflow Cost max (G). Otherwise, the budget beyond the specified range is invalid. Therefore, the model needs to meet the following requirements:

[0099] Cost min (G)≤Cost budget (G)≤Cost max (G).

[0100] S2. Generate a uniformly distributed random initial population in the solution space The number of individuals is ξ, and the initial population Divided into two groups of males and females, the number of male individuals is ξ m , the number of female individuals is ξ f , the calculation expressions are:

[0101]

[0102] ξ f =ξ-ξ m ..

[0103] S3. Start iterative calculation to find the best male individual X in the male group and the female group respectively. m,best and the optimal female individual X f,best , calculate the iterative food quantity Q and temperature Temp:

[0104]

[0105] Among them, d is the current iteration number, D is the maximum iteration number, c 1 is a constant equal to 0.5, exp(·) represents an exponential function, Q and Temp represent the amount of food and temperature at the dth iteration, respectively, that is, as the iteration proceeds, the amount of food gradually increases and the temperature gradually decreases.

[0106] S4. Based on the food quantity threshold Q ts and temperature threshold T ts Determine the abundance of the current food quantity Q and the temperature Temp. ts =0.25, temperature threshold T ts =0.6.

[0107] If Q<Q ts , the snake individuals enter the exploration mode to update their positions, specifically:

[0108] Due to the lack of food in the environment, the snake individuals randomly choose a location to look for more food. m,r The update mechanism can be expressed as:

[0109]

[0110] in, represents the updated position of the rth male individual, X m,rand represents the position of a random male individual, c 2 is a constant equal to 0.05, X max and X min represent the upper and lower bounds of the solution space respectively. Represents the exploration ability of male individuals, and the calculation expression is:

[0111]

[0112] Where Makespan(·) represents the scheduling length generated by the workflow scheduling scheme represented by the corresponding snake individual.

[0113] Female individual X f,r The update mechanism can be expressed as:

[0114]

[0115] in, represents the updated position of the rth female individual, X f,rand represents the position of a random female individual. It represents the exploration ability of female individuals, and the calculation expression is:

[0116]

[0117] If Q>Q ts ,Temp>T ts , the snake individuals enter the development mode to update their positions, specifically:

[0118] Although the amount of food has reached sufficient level, the environmental temperature is still high, and the snakes only move to the optimal position. m,r and female individual X f,r The update mechanism can be expressed as:

[0119]

[0120] Among them, X best The parent population The best individual in 3 is a constant equal to 2.

[0121] If Q>Q ts ,Temp<T ts ,rand>0.6, the snake individuals enter the fighting mode and update their positions, specifically:

[0122] When food resources are sufficient and the ambient temperature is low, snakes show a specific behavioral tendency, that is, there is a 40% probability of entering fighting mode and a 60% probability of entering mating mode. In fighting mode, males will compete fiercely for the right to mate with the best females. At the same time, females will also make careful choices and strive to mate with the most ideal males. Male X m,r The update mechanism is:

[0123]

[0124] Among them, X f,best It is the best female individual among the parent female individuals. Represents the fighting ability of male individuals, and the calculation expression is:

[0125]

[0126] Female individual X f,r The update mechanism is:

[0127]

[0128] Among them, X m,best It is the best male individual among the parent male individuals. Represents the fighting ability of female individuals, and the calculation expression is:

[0129]

[0130] If Q>Q ts ,Temp<T ts ,rand<0.6, the snake individuals enter the mating mode and update their positions, specifically:

[0131] The mating between each pair of snakes is closely related to the amount of food. In the mating mode, the female individual may lay eggs and hatch new snake individuals. m,r The update mechanism is:

[0132]

[0133] in, It represents the mating ability of male individuals, and the calculation expression is:

[0134]

[0135] Female individual X f,r The update mechanism is:

[0136]

[0137] in, It represents the mating ability of female individuals, and the calculation expression is:

[0138]

[0139] If the egg hatches, replace the worst male X m,worst and female individual X f,worst , the expressions are:

[0140] X m,worst =X min +θ×(X max -X min ),

[0141] X f,worst =X min +θ×(X max -X min ).

[0142] S5. When the population After all snake individuals in the update mode complete their position update, the offspring population is obtained. For offspring population The dimension value x of the individual in i Adjust so that x i Satisfies [1,|V|], if x i <1, then adjust x i =1; if x i >|V|, then adjust x i =|V|.

[0143] S6. Comparison and The corresponding individuals in the population are selected to form a new population with better individuals.

[0144] S7. Update the population using negative offset mechanism Get the new parent population to enter the next iteration

[0145] S8. The iteration ends and the optimal individual X is obtained best .

[0146] Example 2

[0147] This embodiment provides an algorithm for the negative offset mechanism, such as Figure 2 , the steps include:

[0148] S1. Calculate the number of random male individuals ξ according to the random probability γ m,rand and the number of female individuals ξ f,rand ;

[0149] S2. From the population The number of choices in the male and female groups is ξ m,rand The number of male individuals is ξ f,rand Female individuals;

[0150] S3. traverse all randomly selected male and female individuals respectively and shift their positions;

[0151] Determine whether this male individual meets the budget constraint. If the individual cost is greater than the budget cost, calculate the individual's negative relaxation cost as:

[0152]

[0153] Otherwise, determine the next male individual.

[0154] S4. Calculate the minimum cost deviation MCD (n i ):

[0155] MCD(n i )=Cost(n i ,vm k )-Cost min (n i ).

[0156] Budget Cost bgt (n i ):

[0157]

[0158] S5. Randomly select a dimension value x of an individual h , and then find the offset queue Q of the hth dimensionO (h);

[0159]

[0160] Among them, for the non-existent vm k Make Cost(n h ,vm k )≤Cost bgt (n h ) situation, individual The offset queue of the hth dimension is:

[0161]

[0162] S6. From Q O In (h), a random VM index is selected to replace x h .

[0163] S5. After traversing all randomly selected male individuals, traverse all randomly selected female individuals, and the position offset is the same as steps S3 to 6.

[0164] Example 3

[0165] This embodiment uses four types of real-world typical scientific workflow applications widely used in the field of high-performance computing as test task sets: the first type is the Epigenomics workflow application (EP) in the field of biological genetics; the second type is the LIGO workflow application in the field of gravitational physics; the third type is the Gaussian elimination parallel application (GE); and the fourth type is the Montage workflow application in the field of astronomy.

[0166] The resource pool of the heterogeneous cloud computing system simulated in the experiment consists of 128 heterogeneous virtual machine instances. The type and price of the virtual machine instances are set with reference to Amazon Elastic Compute Cloud (EC2), and the algorithm program is implemented in Python. This embodiment uses OSAM, MG-PRO and SO as comparison algorithms, aiming to comprehensively verify and demonstrate the effectiveness of the NO-SO proposed in the present invention in solving the problem of minimizing the length of workflow scheduling under budget constraints through performance comparison with these three algorithms. Since the four algorithms are all population-based metaheuristic algorithms, a consistent population size and maximum number of iterations are set in the experiment. In addition, the specific parameters of OSAM, MG-PRO and SO are referenced from the original work, and the parameter settings of the NO-SO algorithm proposed in this article are shown in Table 1:

[0167] parameter value describe ξ 50 Population size D 1000 Maximum number of iterations γ 10％ Random probability of negative offset phase <![CDATA[c 1 ,c 2 ,c 3 ]]> 0.5,0.05,2 Step size parameter

[0168] Compared with OSAM, MG-PRO and SO, NO-SO's success rate in searching for feasible solutions under budget constraints increased by an average of 49.44%, 34.44% and 18.89% respectively. In the scheduling length comparison experiment under different budget constraints, different task scales and different numbers of virtual machines, NO-SO reduced the scheduling length by an average of 42.88%, 33.90% and 34.42% respectively compared with OSAM, MG-PRO and SO. The entire optimization process evolved and converged from "reducing costs to meet budget constraints" to "reducing time to minimize the workflow scheduling length". The convergence effect is shown in Figure 2. Figure 3 As shown in the figure, NO-SO achieves more efficient scheduling performance and obtains a smaller scheduling length.

[0169] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A snake optimization budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud, characterized in that the steps include: S1. Generate a random initial population Divide the individuals of the population into two groups of males and females; S2. Start iterative calculation to find the best male individual X in the male group and the female group respectively. m,best and the optimal female individual X f,best ; S3. Calculate the iterative food quantity Q and temperature Temp, and calculate the food quantity Q and temperature Temp according to the set food quantity threshold Q ts and temperature threshold T ts Select mode to update position; S4. After completing the position update, the offspring population is obtained For offspring population The dimension value x of the individual in i Make adjustments; S5. Comparison and The corresponding individuals in the population are selected to form a new population with better individuals. S6. Update the population using negative offset mechanism Get the new parent population to enter the next iteration S7. Complete the iterative calculation and search for the optimal individual X best .

2. The snake-optimized budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud according to claim 1, characterized in that: The mode selection for updating the position is: If Q<Q ts , then the male individual X m,r The update mechanism can be expressed as: in, represents the updated position of the rth male individual, X m,rand represents the position of a random male individual, c2 is a constant, X max and X min represent the upper and lower bounds of the solution space respectively, Indicates the exploration ability of male individuals; Female individual X f,r The update mechanism can be expressed as: in, represents the updated position of the rth female individual, X f,rand represents the position of a random female individual, Indicates the exploration ability of female individuals; If Q>Q ts ,Temp>T ts , then the male individual X m,r The update mechanism is expressed as: Female individual X f,r The update mechanism is expressed as: Among them, X best The parent population The optimal individual in , c3 is a constant equal to 2; If Q>Q ts ,Temp<T ts ,rand>0.6, then the male individual X m,r The update mechanism is: Among them, X f,best is the best female individual among the parent female individuals, Indicates the fighting ability of male individuals; Female individual X f,r The update mechanism is: Among them, X m,best is the best male individual among the parent male individuals. Indicates the fighting ability of female individuals; If Q>Q ts ,Temp<T ts , rand<0.6, then the male individual X m,r The update mechanism is: in, Indicates the mating ability of male individuals; Female individual X f,r The update mechanism is: in, Indicates the mating ability of female individuals.

3. The snake-optimized budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud according to claim 2 is characterized in that: The expression is: The expression is:

4. The snake-optimized budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud according to claim 2, characterized in that: The calculation expression is: The expression is:

5. The snake-optimized budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud according to claim 2, characterized in that: The expression is: The expression is:

6. The snake-optimized budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud according to claim 1, characterized in that: The adjustment described in S4 is: if x i <1, then adjust x i =1; if x i >|V|, then adjust x i =|V|, so that x i Satisfies [1,|V|].

7. The snake-optimized budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud according to claim 1, characterized in that: The negative offset mechanism steps include: S51. Calculate the number of random male individuals ξ according to the random probability γ m,rand and the number of female individuals ξ f,rand ; S52. From the population The number of choices in the male and female groups is ξ m,rand The number of male individuals is ξ f,rand Female individuals; S53. traverse all randomly selected male individuals and female individuals respectively, and shift their positions; S531. If the current traversal individual does not meet the budget constraint, the negative relaxation cost of the individual is calculated as: And calculate the minimum cost deviation MCD (n i ) and budget cost bgt (n i ); S532. Randomly select a dimension value x of an individual h , and then find the offset queue Q of the hth dimension O (h), from Q O In (h), a random VM index is selected to replace x h .

8. The snake-optimized budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud according to claim 7, characterized in that: Minimum cost deviation MCD(n i ) is expressed as: MCD(n i )=Cost(n i ,vm k )-Cost min (n i ). Cost(n i ,vm k )=w i,k ×price k .Cost min (n i ) is the minimum execution cost.

9. The snake-optimized budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud according to claim 7, characterized in that: Budget Cost bgt (n i ) is expressed as:

10. The snake-optimized budget-aware scheduling algorithm based on negative offset mechanism in heterogeneous cloud according to claim 7, characterized in that: Offset Queue Q O (h) is expressed as:

Images