Research on Optimization of Cloud Computing Resource Allocation Strategy in Scenic Area Operation and Management under Smart Tourism

Under the wave of informationization, smart tourism has gradually become a key force in promoting the management and industrial upgrading of tourist attractions. The core of smart tourism lies in its thought leadership, which advocates data-driven decision-making, with the help of big data analysis, cloud computing, Internet of Things and other advanced technological means, to improve management efficiency and optimize the operation mode, so as to significantly improve the service quality and attractiveness of tourist attractions [1-2]. In today’s highly developed era of information technology, smart tourism, as an emerging tourism model, is quietly changing the way people travel [3]. Smart tourism, also known as intelligent tourism, its core concept lies in the deep integration of cutting-edge technologies such as cloud computing, Internet of Things, mobile Internet, etc., to achieve instant perception, processing and sharing of tourism information, so as to provide tourists with a more personalized and intelligent tourism experience. The introduction of this concept not only promotes the tourism industry in the direction of digitalization and intelligence, but also promotes the innovative development of the tourism industry [4-5].

The core of smart tourism lies in “wisdom”, i.e., through advanced information technology means, to realize the active perception and intelligent processing of multi-dimensional information on tourism resources, tourism activities, tourism processes, etc. [6]. Specifically, the smart tourism system can collect and analyze the data of tourists’ behaviors, environmental changes, and facility status in real time through sensors, cameras, and other Internet of Things (IoT) devices inside and outside the scenic spots, and provide tourists with accurate navigation, booking, payment, evaluation, and other services [7-8]. Meanwhile, through big data analysis, the smart tourism platform is able to gain insight into tourists’ preferences, push customized tourism information, help tourists plan their itineraries, and enhance tourists’ satisfaction [9].

The rise of smart tourism is due to the deep integration of information technology with the tourism industry. With the popularization of the Internet and the maturity of mobile communication technology, the acquisition and sharing of tourism information has become more convenient than ever. Tourists only need a smartphone to access rich tourism information anytime and anywhere, and complete various operations from pre-trip preparation to the itinerary [10-11]. Through the establishment of intelligent management system, the scenic spot can realize all-round monitoring and scheduling of staff, facilities, safety and other aspects, and improve management efficiency. At the same time, through big data analysis technology, scenic area managers can accurately grasp the needs of tourists, optimize the allocation of resources, improve the quality of service, so as to attract more tourists and drive the growth of tourism economy [12-14]. The realization of intelligent tourism cannot be separated from the strong support of information technology. Taking cloud computing as an example, it has the ability of mass data storage and processing, which makes all kinds of complex data analysis and intelligent applications possible. The Internet of Things (IoT) technology, on the other hand, realizes a comprehensive perception of the tourism environment by connecting the physical world with the digital world. The popularization of mobile Internet, on the other hand, enables tourists to access the smart tourism system anytime and anywhere to enjoy seamless tourism services [15-17].

Currently, the upstream and downstream supply chain of the tourism industry has a very low degree of informatization, and is in the “Stone Age” of the Internet, almost all the data release, collection, statistics and distribution are done manually, and in the face of such a large amount of data, the duplication of work consumes huge resources, and the efficiency of information dissemination is extremely low [18-19]. Eradicating the industry’s pain points will generate a huge new market space, thus opening a breakthrough in the deep integration of the entire tourism industry. The construction of the scenic mobile service system will build a digital scenic management model, integrate scenic information resources, and reduce the waste of scenic resources. Through the cloud data platform, it is convenient for tourists to access the attraction data information [20-21]. Realize the two-way interaction of tourists’ scenic spots and create a new management and service mode. Scenic mobile service system to meet the basic needs of tourists in the scenic area tour, to a certain extent, to enhance tourist travel satisfaction, intelligent navigation routes, to solve the problem of unclear routes attractions stranded, and indirectly accelerate the speed of personnel flow for the scenic area to generate income. Therefore, the construction of this system has an extremely wide range of application and promotion value [22-23].

In this paper, we design FOA-ACA to implement the ant colony algorithm coupled with Drosophila in order to optimize the cloud computing resource allocation in scenic area management. After the initialization of parameters, the position of the fruit fly population is initialized and this path is temporarily taken as the optimal solution. Performing the iterative process of the ACO algorithm, the ants calculate the state transfer probability to each point and select the next point by combining the roulette wheel algorithm specific to the Drosophila ACO fusion algorithm. Then the taboo table is modified, the fitness function of each fruit fly position (path traveled by each ant) is calculated, the optimal solution is found and the location of the optimal solution is recorded, and the iteration is ended when the end condition is satisfied. Simulation experiments are conducted with two baseline models, ACO and GA, to verify the task completion time and system load balancing degree of this paper’s algorithm.

2

Resource allocation strategy and optimization algorithm of scenic cloud computing platform

2.1

Modeling the Basic Problem of Resource Allocation for Cloud Computing Platforms in Scenic Areas

2.1.1

Definition of Resource Allocation for Landscape Cloud Computing Platforms

Scenic Cloud Computing Platform Resource Allocation Evaluation in checking whether or to what extent the scheduling results satisfy the QoS requirements of the users with a certain policy [24].

Assume that function $f$ is a scheduling relation over $m$ processing units and tasks $G = (T, E)$ , and that $f$ maps each task to a processing unit starting at a particular time. The process of allocating resources to processing units for a scenic cloud computing platform can be formally described as: 1 $f : T \to {1, 2, \dots, m} \times [0, \infty]$

A Gantt chart is a quaternion table that can be described as the following representation: 2 $G a n t t (P_{i}, t_{j}, S T (t_{j}, P_{i}), F T (t_{j}, P_{i}))$

where:

$P_{i} (i = 1, 2, \dots, m)$ is the processing unit;

$t_{j} (j = 1, 2, \dots, n)$ for tasks assigned to $P_{i}$ ;

$S T (t_{j}, P_{i})$ is the start time of $t_{j}$ on $P_{i}$ ;

$F T (t_{j}, P_{i})$ is the end time of $t_{j}$ on $P_{i}$ .

The maximum schedule length SL (Schedule Length) of a scenic cloud computing platform resource allocation system is defined as among all processing $P_{i}$ machines: 3 $\max_{i} \{\sum_{j} F T (t_{j}, P_{i})\}$

2.1.2

Classification of Resource Allocation Strategies for Landscape Cloud Computing Platforms

Considering from the scenic cloud computing platform resource allocation strategy itself, the scenic cloud computing platform resource allocation can be divided into two categories: static scheduling and dynamic scheduling. A hybrid scheduling approach that combines the dynamic and static scheduling strategies is adopted, while taking into account the superiority of both [25].

From the support of system state estimation, the resource allocation of Gridscape Cloud Computing Platform can be divided into two categories: stateful prediction and stateless prediction, and the prediction methods include heuristic methods, economic model-based methods and machine learning methods. The methods without prediction include heuristic methods and probability distribution methods.

2.2

QoS-based task categorization

Quality of Service QoS is a security mechanism for the performance of the interconnected network, in the smart tourism cloud computing environment, QoS is an important factor to measure the satisfaction of the user using cloud computing services, the evaluation of the function and performance of the cloud computing services will no longer use the evaluation criteria of the traditional services (e.g., speed, cost-effective, etc.), but with the goal of user satisfaction, measured in terms of quality of service. The QoS parameters considered are as follows: 1)

Network bandwidth: When customers have high bandwidth requirements for smart tourism communications, such as multimedia data transmission, they need to prioritize bandwidth requirements and provide higher bandwidth.

2)

Service completion time: for users with high real-time requirements, it is necessary to complete the task in as short a time as possible to respond to user-submitted jobs in a timely manner.

3)

System reliability: For users running a large number of complex tasks, the cloud computing center is required to provide stable and reliable performance support, such as massive data storage services.

4)

Required cost: Cloud computing is pay-as-you-go, and the cost of using it is a factor of concern for users, and for users who want inexpensive services, the cost is a measure.

Therefore, different QoS parameters are set for different user requirements, and the user’s satisfaction level of smart tourism is measured according to these parameters, so as to establish different quantitative evaluation standards.

2.3

Map Reduce level scheduling

In Map Reduce programming model, concurrent processing, fault-tolerant processing, load balancing and other issues are abstracted into a function library (lib), through the Map Reduce interface, the user can automatically concurrent and distributed execution of large-scale computation.The computation execution process in Map Reduce programming model can be abstracted into three roles]: Master, worker and Master is the central controller of the system, responsible for task allocation, load balancing, fault-tolerant processing, etc. Worker is responsible for receiving tasks from Master, performing data processing and computation, and responsible for data transmission and communication, and User is the user side, which inputs tasks to realize the Map and Reduce functions, and controls the whole computation process.

Master is the main control program that keeps some data structures, it stores for each Map and Reduce task their state (idle, working, completed) and the identification of the Worker machine (the machine of the non-idle task), the Map Worker parses the Key/Value pairs of the tasks, performs the Map operation, caches the resultant Key/Value on the local disk and returns the address to the Master; Reduce Worker obtains the address of the Key to Value pair of the intermediate result from the Master, reads the data, sorts and simplifies it by Key, and returns the result to the user program.

3

Drosophila ant colony fusion algorithm

3.1

Mathematical Modeling of Ant Colony Algorithm

Let $M$ ant be put into $N$ randomly selected cities, ant $k (k = 1, 2, \dots, m)$ in the process of searching for the target city, according to the concentration of pheromone on each path to determine the direction of its transfer, always moving in the direction of the concentration of the larger direction, the initial stage, due to the amount of pheromone on the paths do not differ much, the ants on the random selection of a [26]. Table $t a b u_{k} (k = 1, 2, \dots, m)$ is used to record the paths traveled by ant $k$ , and dynamically adjusted with the ants’ continuous movement change process, and $p_{i j}^{k} (t)$ is used to represent the state transfer probability of ant $k$ choosing city $j$ as the target city at $t$ moments: 4 $P_{i j}^{k} = \{\begin{matrix} \frac{τ_{i j}^{α} (t) η_{i j}^{β} (t)}{\sum_{m \in a l l o w e d_{k}} τ_{i n}^{α} (t) η_{i n}^{β} (t)} & j \in a l l o w e d_{k} \\ 0 & Otherwise \end{matrix}$

Which:

$a l l o w e d_{k}$ denotes the city that the ant $k$ is allowed to choose next; $τ_{i j} (t)$ denotes the pheromone concentration left on the paths of cities $i$ and $j$ at moment $t$ ;

$η_{i}$ denotes the initial information transferred from city $i$ to city $j$ , which is generally available from the problem itself.

$η_{i j} = \frac{1}{d_{i j}}$ is the a priori value from city $i$ to city $j$ , $d_{i j}$ indicates the distance between cities $i$ and $j$ , the smaller $d_{i j}$ , the larger $η_{i j}$ , the larger $p_{i j}$ becomes.

$α$ is the information-inspired factor, reflecting the amount of information accumulated on the path plays a guiding role in the movement of other ants, indicating the relative importance of the trajectory, the larger the value, the more the ant tends to choose the path that other ants pass through. In this paper we take $α = 1$ .

$β$ is the expected heuristic factor, which reflects the degree of importance of the heuristic information in the ant’s choice of path, and indicates the relative weight of the predicted value of the computational ability. In this paper we take $β = 5$ .

In the actual calculation process, if the amount of residual information on the path is not processed in a certain way, then as the ants search process, more and more pheromones on the path will overwhelm the inspired information, so each ant walks a path or completes the traversal of all the $n$ cities, it is necessary to use a certain strategy to adjust the pheromone, and gradually reduce it with the passage of time, we use the following rule to Adjust the amount of pheromone on path $(i, j)$ at the $t + n$ moment: 5 $τ_{i j} \begin{matrix} (t + n) = (1 - ρ) τ_{i j} (t) + Δ τ_{i j} (t) \end{matrix}$ 6 $Δ τ_{i j} (t) = \sum_{k = 1}^{m} Δ τ_{i j}^{k} (t)$ where:

$ρ$ denotes the pheromone volatilization factor, then $1 - ρ$ denotes the pheromone residual factor, in order to prevent the infinite accumulation of information, we limit the value range of $ρ$ to $ρ \subset [0, 1]$ , and in this paper we take $ρ = 0.5$ ;

$Δ τ_{i j} (t)$ denotes the residual pheromone concentration left on the path from $i$ to $j$ during the visit of ant $k$ in time $t$ to $t + n$ time, i.e., the pheromone increment on the path $(i, j)$ in this cycle, the initial moment $Δ τ_{i j} (0) = 0$ ;

In this paper, the pheromone updating strategy adopts the Ant-Cycle model: 7 $Δ τ_{i j} (t) = \{\begin{matrix} \frac{Q}{L_{k}} & I f t h e k t h a n t p a s s e s t h r o u g h (i, j) i n t h i s c y c l e \\ 0 & e l s e \end{matrix}$

Where: $Q$ denotes the pheromone strength, which affects the convergence speed of the algorithm to some extent; $L_{k}$ denotes the total length of the path taken by the $k$ rd ant in this loop.

3.2

Ideas of fusion algorithm implementation

Drosophila optimization algorithm and ant colony optimization algorithm are two different population intelligence optimization algorithms, Drosophila optimization algorithm is often used in continuous region and parameter optimization problems due to the simplicity of the algorithm and high accuracy of optimization, this paper synthesizes the advantages of the two algorithms and proposes the Drosophila ant colony fusion algorithm (FOA_ACA).

3.3

Fusion Algorithm Implementation Flow

3.3.1

Initializing the Ant Colony Algorithm Pheromone Concentration and Initial Position of Drosophila Population

The initial pheromone concentration matrix of the ant colony algorithm is set up according to the concept of weights, viz: 8 $\{\begin{matrix} τ_{i j} (0) = \frac{D_{i} + D_{i}}{2} & i \neq j \\ τ_{i j} (0) = 0 & e l s e \end{matrix}$

The initial position of the fruit fly population, i.e., a random path in the TSP problem, can only be used to initialize the path using heuristic information due to the scarcity of initial information in the algorithm.

3.3.2

Mechanism for updating state transfer probabilities

First, follow the state transfer probability function in the weighted ACO algorithm: 9 $p_{i j}^{k} (t) = \{\begin{matrix} \frac{{[τ_{i j} (t)]}^{α} \times {[η_{i j} (t)]}^{β}}{\sum_{s \subset a l l o w_{k}}^{k} {[τ_{i j} (t)]}^{α} \times {[η_{i j} (t)]}^{β}} & I f j b e l o n g s a l l o w_{k} \\ 0 & O t h e r w i s e \end{matrix}$

3.3.3

Pheromone updating mechanisms

The ant colony will leave pheromone behind during the transfer process, and at the same time, the residual information has to be updated and processed. Considering the volatilization of pheromone, the amount of information on the path $(i, j)$ at the moment $t + n$ can be adjusted according to the following rule: 10 $τ_{i j} (t + n) = (1 - ρ) \times τ_{i j} (t) + Δ τ_{i j} (t)$ 11 $Δ τ_{i j} (t) = \sum_{k = 1}^{m} Δ τ_{i j}^{k} (t)$

The case of locally optimal solutions is solved to some extent by pheromone quadratic reinforcement of longer paths on the optimal path to increase the probability of longer paths being selected: 12 $C D S P_{i j} = d_{i j} / L_{k}$ 13 $\{\begin{matrix} Δ τ_{i j}^{k} (n e w) = \frac{Q}{L_{k}} (1 + r a n d), C D S P_{i j} > q_{0} \\ Δ τ_{i j}^{k} (n e w) = Δ, e l s e \end{matrix}$

3.3.4

Algorithmic flow

1)

Parameter initialization. The main parameters in the Drosophila ant colony fusion algorithm are $\propto$ , $β$ and the number of ants (Drosophila) $m$ and so on. The magnitude of $α$ indicates the importance of the amount of information on each path; the magnitude of $β$ represents the importance of the heuristic factor.

2)

Initialize the location of the fruit fly population and use this path as the optimal solution for the time being. Place $m$ ant randomly on $n$ cities and select the next city according to the betting board algorithm that combines the fruit fly optimization algorithm.

3)

Perform the iterative process of the ant colony algorithm, where the ants calculate the state transfer probability to each city and select the next city according to the roulette wheel algorithm specific to the Drosophila ant colony fusion algorithm:

4)

Modify the taboo table.

5)

Repeat 3) and 4) until all ants have traveled to all cities.

6)

Calculate the fitness function for each fruit fly location (path traveled by each ant), find the optimal solution, and record the location of the optimal solution.

7)

Determine whether the iteration end condition $N_{c} > N_{\max}$ is satisfied or the difference between the two solution results is less than 0.1%, if it is satisfied, end the loop and output the results; if not, update the pheromone and empty the taboo table for the next loop.

4

Drosophila Ant Colony Fusion Algorithm for Resource Allocation in Scenic Cloud Computing Platforms

4.1

Resource allocation framework for scenic cloud computing platform based on FOA-ACA

Cloud computing scenic cloud computing platform resource allocation system usually needs two frameworks: first, cloud resource node management directory; second user task demand analysis center. To address these two issues, a cloud computing scenic spot cloud computing platform resource allocation model based on the FOA_ACA algorithm is proposed as shown in Figure 1.

User: interacts with the user smart tourism task demand analysis center, which is the initiation and end point of all tasks.

User task demand analysis center: collects and analyzes user input tasks, sorts the task queue according to the user bidding and the task completion time and other requirements proposed by the user, and looks for available cloud computing resources, calculates the resource consumption and execution time of the task queue on the corresponding resources.

Cloud Resource Catalog: It has two main functions: first, it regularly checks and updates the information of CPU, memory and broadband resources of each resource in the Cloud Resource Agent; second, it reviews the task sequences sent by the user’s Task Requirements Analysis Center, and then calls the corresponding resource nodes for the execution of the tasks.

Cloud Resource Agent: It is the resource pool for cloud computing to execute tasks, and after each execution of the task, it carries out the updating of pheromones and regularly sends the corresponding node computing resources to the cloud resource catalog for preservation.

4.2

Resource allocation process of scenic cloud computing platform based on FOA-ACA

In solving the TSP problem we learned that the core of the fusion algorithm is the ant colony algorithm, just using the fruit fly algorithm to improve the ant colony algorithm pheromone initialization and state transfer when the choice of mechanism, here the first use of the ant colony algorithm to describe the resource allocation process of the cloud computing platform of the cloud computing landscape, and then extended to the fruit fly ant colony algorithm.

4.2.1

Cloud Information Initialization

First calculate the comprehensive computing power of the cloud computing resource node and the total computing resources to be consumed by the task: $X a b i l i t y_{i}$ represents the comprehensive computing power of the node, and $X C_{i}$ , $X B_{i}$ , and $X M_{i}$ represent the mathematical abstraction of CPU, broadband, and storage capacity, respectively, where $a + b + c = 1$ represents the contribution of different resources in the node to the comprehensive computing power.

14 $X a b i l i t y_{i} = a * X C_{i} + b * X B_{i} + c * X M_{i}, 1 < i < m$ $$Xabilit{y_i} = a*X{C_i} + b*X{B_i} + c*X{M_i},1 ˂ i ˂ m$$

$Y s u m_{j}$ represents the total computational resources to be consumed by the task, $Y C_{j}$ , $Y B_{j}$ , and $Y M_{j}$ represent the mathematical abstraction of the CPU, broadband, and storage capacity to be consumed, respectively, where $a + b + c = 1$ represents the weight of the resources to be consumed by the task.

15 $Y s u m_{j} = a * Y C_{j} + b * Y B_{j} + c * Y M_{j}, 1 < j < m$ $$Ysu{m_j} = a*Y{C_j} + b*Y{B_j} + c*Y{M_j},1 < j < m$$

Initialize the Drosophila population position using the roulette method, refer to the state transfer probability of the ant colony algorithm when the task selects a node, the initial pheromone is zero, and the heuristic information is the comprehensive computational capacity of the resource node, when all the tasks have finished selecting the resource node, i.e., after the task allocation is completed, the execution time of all the tasks on each node is calculated, and then the pheromone concentration is initialized according to the execution time.

4.2.2

Probability that a task chooses the state of a resource

An iteration is started to initialize the task information and combine the pheromone information and heuristic information to compute the selection probability of the task to each resource node as follows.

16 $p_{i j}^{k} (t) = \{\begin{matrix} \frac{{[τ_{i j} (t)]}^{\propto} \times {[η_{i j} (t)]}^{β}}{\sum_{s \subset s a l l o w} {[τ_{i j} (t)]}^{\propto} \times {[η_{i j} (t)]}^{β}} & I f j belongs a l l o w_{k} \\ 0 & Otherwise \end{matrix}$

The specific roulette algorithm is then utilized to decide which resource node the task is to be assigned to.

The specific roulette algorithm has been described in detail again in the previous chapter, and this section will briefly explain the resource allocation process in the context of the Cloud Computing View cloud computing platform. When a task to exceed 3 times or even 5 times the probability of choosing a node, it so happens that the node in the last iteration of the best path, then we determine that this task must be selected this node, otherwise calculated in accordance with the normal roulette method, if there is no more special probability, but also in accordance with the normal roulette method. At the same time, considering the problem of resource load balancing, when a task selects a node in the cloud computing resource pool, we temporarily halve the pheromone concentration of that node, and wait until all the tasks have been assigned, and then return to the level when the task is first assigned.

4.2.3

Updated pheromone concentrations

Tasks leave pheromone on a resource node after selecting that node, in order to simplify the model, we wait for all tasks to be assigned to update the pheromone concentration [27]. Also considering the volatilization of pheromone concentration, the amount of information on path $(i, j)$ at moment $t + n$ can be adjusted according to the following rule: 17 $τ_{i j} (t + n) = (1 - ρ) \times τ_{i j} (t) + Δ τ_{i j} (t)$ 18 $Δ τ_{i j} (t) = \sum_{k = 1}^{m} Δ τ_{i j}^{k} (t)$

Where, denotes the $ρ$ pheromone volatilization coefficient, whose value range is $ρ \subset [0, 1]$ ; $Δ τ_{i j} (t)$ denotes the pheromone increment on the path $(i, j)$ in this cycle.

In order to solve the situation of falling into the local optimal solution in the resource allocation process of the scenic cloud computing platform, the concept of task resource contribution is introduced.

19 $C D S P_{i j} = d_{i j} / L_{k}$ 20 $\{\begin{matrix} Δ τ_{i j}^{k} (n e w) = \frac{Q}{L_{k}} (1 + r a n d), C D S P_{i j} > q_{0} \\ Δ τ_{i j}^{k} (n e w) = Δ, e l s e \end{matrix}$

$r a n d$ is the pheromone concentration randomly strengthened for paths with large path contribution, usually $r a n d$ can take a fixed value of 1-2, and $q_{0}$ is the path contribution threshold, which usually can take a value according to the number of cities.

Based on the above three algorithm process descriptions, the resource allocation process of scenic cloud computing platform based on FOA_ACA algorithm can be drawn as shown in Fig. 2:

5

Algorithm performance simulation experiment analysis

5.1

Pheromone-inspired factor setting

Self-organizing system evolution time consumes longer time. Practical applications require the time complexity of the algorithm, so the module of random search i.e. introduction of heuristic factor is introduced in the probability of state transfer in the forward direction of the ants. The process of giving initial guidance to the ant colony algorithm strengthens the effectiveness of the algorithm time.

The ants leave pheromones on the edges of the mapping while matching the VMs and tasks. The heuristic factor α indicates the relative importance of the accumulated pheromone in influencing the ants’ matching solutions. The larger the value of α, the higher the probability that the ants select the previous task-virtual machine mapping, and the randomness and diversity of the solutions are weakened; when the value of α is too small, the solutions searched for are similarly poor.

The effect of pheromone-inspired factor values in the ant colony algorithm on the performance of the scheduling algorithm is analyzed by simulation experiments with 100task-10vm and 200task-10vm as the study subjects. The relationship between the pheromone-inspired factor and the task completion measurement time CompleteTime is shown in Fig. 3 and Fig. 4. Where Figure 3 is at 100 tasks. Figure 4 is at 200 tasks. The relationship between the pheromone-inspired factor and the number of iterations is shown in Figure 5.

From the experimental results shown in Fig. 3 to Fig. 5, it can be seen that the value of the pheromone-inspired factor α has a large impact on the number of iterations of the algorithm and the completion time of the task, when α is less than 0.5, the algorithm converges slowly, the utilization of existing information is insufficient, and the solution is not optimal; when α is large, it is given to pheromone in the task and the virtual machine has been mapped paths of too much importance, which leads to a positive effect on the optimum completion time The feedback effect is strengthened, and the ability to utilize the existing information is enhanced, but some ants will stick to the existing allocation scheme, the innovation ability is weakened, and the algorithm will appear the local optimum phenomenon.

100task-10vm experiments show that when the value of α is in the range of 0.55 to 3.85, the comprehensive solution ability of this paper’s algorithm is better; when the value of α is greater than 4, the fluctuation of task completion time is larger, combined with Fig. 5 shows that the algorithm converges to the local optimal solution faster when the number of iterations is less than 100. 200task-10vm experiments show that when the value of α is in the range of 0.55 to 3.6, this paper’s algorithm is better; when the value of α is in the range of 0.55 to 3.6, the algorithm is better. 3.6, the algorithm of this paper has better comprehensive solution ability. When the value of α is greater than 3.6, the completion time of 200 tasks fluctuates widely, and the number of algorithm iterations fluctuates up and down around 90. In summary, it is concluded that it is better when the value of α is in the range of 0.5 to 3.5. After comprehensively considering the two indicators of iteration number and minimum task completion time, the value of α is selected to be equal to 3 in this paper.

5.2

Simulation Experiments

5.2.1

Environment setup

1)

Hardware environment: 1024G hard disk, 16G memory.

2)

Software environment: CloudSim 3.0, Windows 10 operating system, Eclipse 4.4, JDK1.7.

The main steps of the experiment are as follows:

Step 1: Configuration parameters, the emulator reads the local configuration file.

Step 2: Initialize the library function, mainly the list of configuration files, user parameters, tracking logs and so on.

Step 3: Create data center.

Step 4: Create the data center agent and get the ID of the agent.

Step 5: Create VMs based on configured parameters and put into VM list.

Step 6: Create cloud task and add to cloud task list.

Step 7: Start and end the simulation experiment.

Step 8: Print the results.

5.2.2

Experimental results and analysis

Experiment 1: Task Completion Time Comparison

This experiment will compare Genetic Algorithm (GA), Ant Colony Algorithm (ACO) and this paper’s algorithm FOA-ACA by task execution time. Set the number of tasks from 100 to 500 gradually, and the resource points are 200. The quality of service of resource points is differentiated by setting the resource point parameters CPU, network bandwidth, etc. At each task number stage, simulation trials are averaged over 10 times. Combined with the parameter selection in Section 5.1 of this paper, the parameter settings for this experiment are shown in Table 1.

Table 1.

Test parameter selection

Parameter	Numerical value	Parameter	Numerical value
Volatile factor	0.65	Population scale	60
Pheromones strengthening factor	0.45	Mutation rate	0.06
pheromones factor	1.2	Cross rate	0.45
Pheromone Inspiration factor	3	Minimum evolutionary rate	0.06
Expected value	1.2	Maximum iteration algebra	550
q0	0.8

Figure 6 shows the comparison between this paper’s algorithm and ACO algorithm in terms of task completion time.

Overall the completion time of this paper’s FOA-ACA algorithm is smaller than that of ACO. in the case of the number of tasks is not too large, the gap between FOA-ACA and ACO algorithm in the scheduling completion time is not very large, but with the gradual increase of the number of tasks, the gap in the completion time is more and more large, so FOA-ACA is more suitable for the application of large-scale resource allocation.

Figure 7 shows the comparison between the FOA-ACA algorithm and the genetic algorithm in terms of task completion time. It can be concluded that the elapsed time of this paper’s algorithm is more than 20s lower than the two baseline models when the number of tasks is 500.

The FOA-ACA algorithm by fusion is superior to the single algorithm in task assignment completion time, and this advantage is more obvious as the number of tasks increases. This further proves that the fusion strategy of FOA-ACA is correct, and there is a significant improvement in the search performance.

Experiment 2: System Load Balancing Comparison

This experiment will compare three algorithms by evaluating the load balancing rate. The number of tasks is set to rise gradually from 20 to 100, the number of resource points is 8, and other parameter settings refer to Experiment 1.

When the number of tasks is 60, the average load of each resource point in 10 experiments is shown in Figure 8. There are differences in the load profiles of each resource point, but FOA-ACA has a smaller gap in the load profiles on each resource point compared to GA and ACO. This also shows that the FOA-ACA algorithm is significantly better than GA and ACO in load balancing.

Then, by calculating the load balancing rate of each resource point under different number of tasks, we get the value of relative standard deviation under different number of tasks. The smaller the value of relative standard deviation, the better the load balancing of the system. The results are shown in Table 2.

Table 2.

Relative standard deviation

Resource nodes	FOA-ACA	GA	ACO
20	0.54	0.73	0.57
40	0.33	0.54	0.43
60	0.22	0.42	0.38
80	0.17	0.3	0.19
100	0.12	0.21	0.15

Figure 9 shows a comparison of the relative standard deviations of the results of the task assignment. The following points can be drawn:

First, the relative deviation values of all three algorithms are gradually decreasing with the increase of the number of tasks, which indicates that the larger the number of tasks, the better the load balancing of the system. Second, the relative deviation of the FOA-ACA algorithm is significantly lower than that of ACO and GA, which indicates that the load balancing of the system is significantly better than that of ACO and GA in the scheduling process of FOA-ACA. Thirdly, the relative deviation value of ACO is lower than that of GA, which is because ACO has better global search ability than GA, and this also provides the basis for this paper to use ACO as the main algorithm after fusion. From the above three points we can conclude that the algorithm designed in this paper is usable in resource allocation and can improve the load balancing of the system.

6

Conclusion

In order to realize the load balancing of scenic operation management system, this paper designs a cloud computing resource allocation model based on Drosophila ant colony fusion algorithm. Simulation experiments are conducted to analyze the performance of the algorithm. When the specific value ( ) is better in the range of 0.5 to 3.5, considering the number of iterations and the minimum task completion time, the value of is selected as 3. Compared with the baseline models ACO and GA, the algorithm in this paper takes less time to complete. In the case of low number of tasks, there is no significant difference between the time consumed by FOA-ACA and the baseline algorithm, but with the increase of the number of tasks, the gap in the completion time is getting bigger and bigger, which verifies that the designed FOA-ACA is more suitable to be applied in large-scale resource allocation. In terms of load balancing, the FOA-ACA algorithm is significantly better than GA and ACO.In summary, the designed model is able to realize cloud computing load balancing in the management of tourist attractions.

Sprache:: Englisch

Zeitrahmen der Veröffentlichung:: 1 Hefte pro Jahr
Fachgebiete der Zeitschrift:: Biologie, Biologie, andere, Mathematik, Angewandte Mathematik, Mathematik, Allgemeines, Physik, Physik, andere

Zeitschrift RSS Feed

Research on Optimization of Cloud Computing Resource Allocation Strategy in Scenic Area Operation and Management under Smart Tourism

Yanlong Wang

Online veröffentlicht: 24. Sept. 2025

Eingereicht: 28. Dez. 2024

Akzeptiert: 17. Apr. 2025

DOI: https://doi.org/10.2478/amns-2025-0976

SchlüsselwörterCloud computing, Resource allocation, Ant colony algorithm, Fruit fly algorithm

© 2025 Yanlong Wang, published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Schlüsselwörter
Cloud computing, Resource allocation, Ant colony algorithm, Fruit fly algorithm