A multi-objective optimization container migration method based on resource balance
By using the improved binary gray wolf optimization algorithm in mobile edge computing network for container migration method, the problems of load imbalance, resource fragmentation and high energy consumption in the existing technology are solved, and resource balance and energy consumption are achieved Reduced effect.
Patent Information
- Application Number
- CN202210532197.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-09
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art ignores the overall network conditions in mobile edge computing networks and only relies on static thresholds or linear regression to predict container migration timing, resulting in load imbalance, resource fragmentation and high energy consumption problems.
A container migration method that jointly considers the system's energy consumption, resource equalization and migration cost is proposed. Multi-objective optimization is performed through the improved binary gray wolf optimization algorithm (BGWO) to optimize the container migration strategy to achieve resource equalization and energy consumption reduction.
It has achieved the improvement of resource equalization in mobile edge computing networks, reduced the energy consumption of data centers, and optimized the cost of container migration to minimize system energy consumption.
Smart Images

Figure CN115022926B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of mobile communications, and in particular relates to a multi-objective optimization container migration method based on resource balance. Background Art
[0002] With the continuous development of the fifth generation mobile communication technology (5G), human demand for automation, artificial intelligence, augmented reality, smart homes and cities, real-time analysis, games and various other industrial and consumer applications continues to increase. The number of things connected to the Internet is continuing to grow, and the mobile data traffic generated is increasing day by day.
[0003] In a mobile edge computing network, the limited resources of edge nodes and the uneven distribution of end-user service requests will lead to significant differences in the service busyness among edge nodes. Containers can achieve high resource utilization and resource isolation. Therefore, containers play an important role in mobile edge computing.
[0004] In order to reduce the operating costs of service providers, one of the main challenges facing service providers is to reduce energy consumption. Servers, network components and cooling systems are the three major energy consumers in data centers, among which server energy consumption accounts for about 50% of the total energy consumption.
[0005] Container migration has become one of the key technologies for reducing data center energy consumption, balancing edge node loads, and optimizing data center resource utilization due to the cross-platform capabilities of containers. Running large-scale data centers consumes a lot of energy. Studies have shown that the server status of data centers varies significantly. Some servers are idle for a long time or have low resource utilization, but the energy consumption is still high. Some servers have high resource utilization and the probability of server downtime is also increased.
[0006] Traditional resource scheduling methods can lead to problems such as high load and high energy consumption. Therefore, it is necessary and meaningful to study how to reduce the number of running servers and improve the energy efficiency of data centers through container integration.
[0007] Reference 1 proposed three zero-downtime migration technologies based on state replication and state replication technology, and proposed an optimal migration technology selection algorithm by comprehensively considering the migration time and the traffic load injected into the network as a result of the migration.
[0008] Reference 2 proposes a flow-aware container migration method, which is a flow-aware container migration algorithm using LXD and CRIU (checkpoint recovery in user space).
[0009] Reference 3 proposes a decision-making mechanism based on container migration in the power Internet of Things; combining local pheromone evaporation and global pheromone update, a migration algorithm based on the improved ant colony algorithm is proposed.
[0010] Reference 4 proposes a container scale layout algorithm through reasonable container static layout and dynamic migration capabilities to reduce the fragmentation of container layout, thereby effectively improving the utilization of machine resources and achieving the goal of cost savings.
[0011] Reference 5 implements parallel multi-container migration in a WAN environment using multiple TCP connections (also called parallel TCP) on a single direct path. The application migration middleware framework uses a feedback controller to determine the appropriate number of parallel container migrations (i.e., parallel window) according to changes in network bandwidth.
[0012] The above articles have proposed relevant strategies for container migration. There are two reasons for container migration: user mobility and current system load imbalance. Regardless of the reason for container migration, the migration cost generated during the migration process is always the primary issue for scholars to study.
[0013] The migration costs mainly include migration delay, SLA during the migration process, and the number of migrations. Migration delay mainly refers to the downtime of the container. The main research is divided into cold migration and hot migration. The difference is that cold migration is to stop the container first, wait for the data migration to complete, and then restart the container; hot migration is to migrate data while the container does not need to be shut down.
[0014] The characteristics of containers make them more suitable for mobile edge computing networks. The existing container migration strategy ignores the overall network status of the current edge computing network and only relies on static thresholds or linear regression to predict the migration timing of containers.
[0015] In order to solve the load balancing problem, the migration decision of the container is only considered from the perspective of a single resource, ignoring the resource fragmentation problem caused by the correlation between resources, the energy consumption problem of the data center in the edge computing network, and the migration cost of the container.
[0016] [1]T.Kim, M.Al-Tarazi, J.-W.Lin and W.Choi. Optimal Container Migration for Mobile Edge Computing: Algorithm, System Design and Implementation[J]. IEEEAccess, 2021, pp.158074-158090.
[0017] [2] Maheshwari S, Choudhury S, Seskar I, et al. Traffic-Aware Dynamic Container Migration for Real-Time Support in Mobile Edge Clouds[C] / / IEEE International Conference on Advanced Networks and Telecommunications Systems. IEEE, 2018.
[0018] [3] Ma Z, Shao S, Guo S, et al. Container Migration Mechanism for Load Balancing in Edge Network Under Power Internet of Things[J]. IEEE Access, 2020, PP(99): 1-1.
[0019] [4] Li K, Chang C, Yun K, Zhang J. Research on Container Migration Mechanism of Power Edge Computing on Load Balancing[C]. 2021 IEEE 6th International Conference on Cloud Computing and Big Data Analytics(ICCCBDA), 2021, pp. 386-390.
[0020] [5] Thongthavorn W, Rattanatamrong P. Multi-Container Application Migration with Load Balanced and Adaptive Parallel TCP[C] / / 2019 International Conference on High Performance Computing&Simulation(HPCS). 2019. Summary of the Invention
[0021] In order to solve the problems of single optimization target and resource balance, the present invention proposes a container migration method that jointly considers system energy consumption, resource balance and latency; in order to obtain the global optimal container migration strategy result, an improved Binary Grey Wolf Optimizer (BGWO) algorithm based on joint resource balance and energy consumption of container migration strategy is designed for verification; that is, it fully utilizes node resources and achieves the purpose of reducing migration costs, and uses container migration to reduce the energy consumption of data centers in mobile edge computing networks.
[0022] The container migration method that jointly considers system energy consumption, resource balance and migration cost has the following specific steps:
[0023] Step 1: Build a container migration network scenario including n edge nodes and m containers; deploy the corresponding initial container on each edge node;
[0024] The set of edge nodes is S = {S1, S2, ... S j ,...,S n}; The set of containers is C = {C1, C2, ... C i ,...,C m};
[0025] For the jth edge node S j , the resource vector owned by the node is k represents the resource dimension on the edge node; the resource utilization of the node is Indicates the utilization rate of the kth resource of the jth edge node.
[0026] The i-th container is defined as in Represents the various resources required by the i-th container, where Represents the kth resource required by the i-th container; represents the computing delay when the i-th container meets the resource requirements, Tol i Represents the latency threshold of the task carried by the i-th container.
[0027] Step 2: Detect the load of each edge node, calculate the resource utilization balance of each node, and obtain the initial node where the container needs to be migrated;
[0028] The initial node meets the following requirements: Greater than the load constant c2 and resource utilization balance Λ j Greater than the resource balance constant c1;
[0029] Edge Node S j Load The calculation formula is:
[0030] in, is the edge node S j Have resources The utilization rate, x i,j Indicates whether the i-th container is deployed on the j-th edge node. If it is x i,j =1, otherwise 0; γ k is the resource utilization At the edge node S j The load weight must satisfy constraint;
[0031] Resource utilization balance j The calculation formula is:
[0032]
[0033] Λ j ∈(0,1), edge node S j The more balanced the resource utilization, the j The closer it is to 0, otherwise Λ j The closer it is to 1;
[0034] is the edge node S j The average utilization rate of various resources owned,
[0035] Step 3: Calculate the probability of migration of each deployed container on the initial node according to the container migration strategy, obtain the priority set of the containers to be migrated of each initial node respectively, and form a container initial deployment matrix with all the containers to be migrated;
[0036] Edge Node S j Upload container C to be migrated i The priority P i,j , the calculation formula is as follows:
[0037]
[0038] Where η is the weight based on migration time, ρ is the weight based on resource balance, and σ is the weight based on the amount of migrated data, satisfying η+ρ+σ=1;
[0039] is the edge node S j Upper container C i Migration probability based on migration time;
[0040]
[0041] Among them, λ is the value of the container Ci The weight factor of the CPU utilization of container C on the migration downtime. i The weight factor of the memory utilization on the migration downtime satisfies λ+μ=1;
[0042] Represents container C i The demand for CPU resources, Represents container C i The amount of memory resources required.
[0043] is the edge node S j Upper container C i The migration probability based on resource balance is expressed as:
[0044]
[0045] corr(i,j) represents the resource complementarity based on similarity: d i,j Indicates the resource similarity between the i-th container and the j-th edge node; S k Indicates the variance of a certain dimension of resources in the edge node. The calculation formula is:
[0046] is the edge node S j Upper container C i The migration probability based on the amount of migrated data is calculated as follows:
[0047]
[0048] t i is container C i The time it has been executed.
[0049] Step 4: Build a container migration model that comprehensively considers resource balance, data center energy consumption, and migration cost, and use the binary gray wolf optimization algorithm to solve it to obtain the migration target nodes corresponding to all containers in the container initial deployment matrix;
[0050] First, the container migration cost is defined as the container migration time, which includes the data transmission time and container downtime during the migration process. The calculation formula is as follows:
[0051]
[0052]
[0053] Describes the migration time of the i-th container from the j-th edge node to the j′-th edge node, where Data i The amount of data that needs to be migrated during container migration. is the memory of the i-th container, B jj′ is the communication bandwidth between the jth edge node and the j′th edge node.
[0054] Mig cost is the total migration time of all containers; M jj′ is the migration decision of whether the i-th container is migrated from the j-th edge node to the container on the j′-th edge node. If yes, then M jj′ =1, otherwise M jj′ =0.
[0055] Then, the resource balance includes resource load balance and resource fragmentation; the calculation formula is:
[0056] Balance=U b +V b
[0057] Among them, the load balancing degree U b The calculation formula is: is the load average of all edge nodes;
[0058] Resource fragmentation level V b The calculation formula is: is the edge node S j The degree of resource fragmentation on:
[0059] is the edge node S j The remaining resource rate of the k-th resource, Describes the edge node S j The remaining amount of the k-th resource; Describes the total amount of the kth resource on edge node j; is the edge node S j The average remaining resource rate of all resources above.
[0060] Next, the system energy consumption En in the system T time is calculated as:
[0061]
[0062] P dc (t) represents the instantaneous power of the system at time t; P j (t) is the edge node S corresponding to time t j The instantaneous power:
[0063] in, is the edge node S j Idle power consumption, is the edge node S j Maximum power consumption, U j (t) is the edge node S at time t j The CPU utilization of the corresponding server, N e is the number of edge nodes in the scene.
[0064] Finally, using the total migration time of the container, resource balance and system energy consumption, a multi-objective optimization problem P under QoS constraints is established:
[0065] P:min{αMig cost +βBalance+γEn}
[0066] st C1:
[0067] C2:
[0068] C3:
[0069] Among them, α, β, γ are the weight coefficients corresponding to each item, and they satisfy α+β+γ=1;
[0070] Constraint C1 stipulates that the total demand for a certain type of resource by the containers deployed on the edge nodes must be lower than the maximum capacity of the resource of that type;
[0071] Constraint C2 indicates that the total delay of calculating the tasks carried by the container to be migrated should not exceed the delay threshold that the task can tolerate; Indicates whether the i-th container is migrated from the j-th edge node to the j′-th edge node. If so, otherwise,
[0072] Constraint C3 stipulates that a container can only be mapped to one edge node.
[0073] The improved binary gray wolf algorithm solves the multi-objective optimization problem P and obtains the target edge node S l ;
[0074] S l =argmin{αMig cost +βBalance+γEn}
[0075] st C1,C2,C3
[0076] The improved binary gray wolf algorithm uses the Piecewise mapping function to initialize the population. The function is as follows:
[0077]
[0078] Among them, P∈(0,1) is the control parameter, x∈(0,1);
[0079] Container migration decision jj′ For 0-1 decision, the wolf pack position is transformed between 0 and 1 by means of a transformation function to solve the discrete problem; the transformation function used in the present invention is:
[0080]
[0081] Where D represents the distance between the gray wolf and its prey, and A is the coefficient vector.
[0082] Step 5: Each initial node migrates each container to the corresponding migration target node according to the migration order of priority until each initial node achieves resource balance.
[0083] The advantages of the present invention are:
[0084] 1. A multi-objective optimization container migration method based on resource balance reduces the energy consumption of the mobile edge computing network system. The simulation results show that by comparing the system energy consumption before and after container migration, the energy consumption of the entire system can be minimized.
[0085] 2. A multi-objective optimization container migration method based on resource balance can balance the resource balance of edge nodes in the mobile edge computing network.
[0086] 3. A multi-objective optimization container migration method based on resource balance achieves a good balance between optimizing system energy consumption, edge node resource balance and migration time. BRIEF DESCRIPTION OF THE DRAWINGS
[0087] Figure 1 A schematic diagram of a multi-objective optimization container migration method based on resource balance according to the present invention;
[0088] Figure 2 A diagram of the container migration network scenario model built for the present invention.
[0089] Figure 3 This is a relationship diagram between the optimization target value and the number of iterations under different algorithms of the present invention.
[0090] Figure 4 This is the energy consumption diagram of the system before and after the container migration of the present invention.
[0091] Figure 5This is the container migration cost graph under different migration strategies of the present invention.
[0092] Figure 6 This is a diagram of resource balance under different migration strategies of the present invention.
[0093] Figure 7 This is a diagram of the optimization target values under different migration strategies of the present invention. DETAILED DESCRIPTION
[0094] The present invention is further explained in detail below with reference to the accompanying drawings and embodiments.
[0095] In order to solve the problem of single continuous optimization target and resource balance, the present invention proposes a multi-objective joint optimization container migration strategy. First, considering the resource utilization of the server, a resource balance model and an energy consumption model are established to evaluate the overall state of the mobile edge computing network, and based on this, edge nodes with unbalanced resource utilization and containers deployed on the node are found; secondly, in order to avoid the resource fragmentation problem caused by container migration, a node complementarity matrix model is established to determine the selection of container migratable nodes; finally, the probability of container migration on the edge node is calculated according to the migration strategy, so as to obtain the containers that need to be migrated first, and comprehensively consider multiple objectives such as resource balance, data center energy consumption and migration cost, and migrate the container to the migratable node according to the container migration order until resource balance is achieved on the node. Finally, the improved Binary Grey Wolf Optimizer (BGWO) algorithm based on the joint resource balance and energy consumption of the container migration strategy is used for simulation experiments.
[0096] The multi-objective optimization container migration method based on resource balance is as follows: Figure 1 As shown, the specific steps are as follows:
[0097] Step 1: Build a container migration network scenario including n edge nodes and m containers, and deploy the corresponding initial container on each edge node;
[0098] The network scenario for building container migration is as follows;
[0099] The set of edge nodes is S = {S1, S2, ... S j ,...,S n}; The set of containers is C = {C1, C2, ... C i ,...,C m}, each container carries one task;
[0100] For the jth edge node S j , the resource vector owned by the node is k represents the resource dimension on the edge node, Indicates that the maximum number of resources that the jth edge node can provide is k;
[0101] The resource utilization of the edge node is Indicates the utilization rate of the kth resource of the jth edge node. The resource types required by the container correspond to the resource types of the edge node.
[0102] The wireless communication bandwidth between the jth edge node and the j′th edge node is represented as B jj′ .
[0103] The i-th container is defined as in Represents the various resources required by the i-th container, where Represents the kth resource required by the i-th container; represents the computing delay when the i-th container meets the resource requirements, Tol i Represents the latency threshold of the task carried by the i-th container.
[0104] Step 2: By detecting the load of each edge node, the resource utilization balance of each node is calculated, and the initial node that needs to be migrated is obtained to form a set of resource fragmentation nodes Φ;
[0105] The initial node meets the following requirements: Greater than the load constant c2 and resource utilization balance Λ j Greater than the resource balance constant c1;
[0106] Define resource balance constant c1 and load constant c2. If And j When >c1, obtain the set Φ of resource fragmentation nodes;
[0107] First, for the edge node S j , calculate the kth resource on the edge node Utilization rate
[0108]
[0109] x i,j Represents the i-th container C i Whether to deploy on edge node S j If yes, then x i,j =1, otherwise, x i,j =0;
[0110] Then, calculate the edge node s j Average utilization of various resources on
[0111]
[0112] Finally, using the edge node s j The average utilization rate of various resources on the node is used to calculate the resource utilization balance of the node Λ j ;
[0113]
[0114] Λ j ∈(0,1), edge node S j The more balanced the resources, the j The closer it is to 0, otherwise Λ j The closer it is to 1;
[0115] At the same time, the utilization rate of various resources is used to calculate the edge node S j Load
[0116]
[0117] Among them, γ k It is a resource At the edge node S j The load weight must satisfy constraint;
[0118] Step 3: Calculate the probability of migration of each deployed container on the initial node according to the container migration strategy, obtain the priority set of the containers to be migrated of each initial node respectively, and form a container initial deployment matrix with all the containers to be migrated;
[0119] Edge Node S j Upload container C to be migrated i The priority P i,j , the calculation formula is as follows:
[0120]
[0121] Where η is the weight based on migration time, ρ is the weight based on resource balance, and σ is the weight based on the amount of migrated data, satisfying η+ρ+σ=1;
[0122] The shorter the downtime of container migration, the smaller the impact on business. is the edge node S j Upper container C i Migration probability based on migration time;
[0123]
[0124] Among them, λ is the value of the container C i The weight factor of the CPU utilization of container C on the migration downtime.i The weight factor of the memory utilization on the migration downtime satisfies λ+μ=1;
[0125] Represents container C i The demand for CPU resources, Represents container C i The amount of memory resources required.
[0126] Since minimizing the container migration time does not improve the resource balance of edge nodes, on the contrary, frequent container migration may increase the total migration downtime. Therefore, the resource complementarity between containers and nodes is used to measure the impact of containers on the resource balance of edge nodes.
[0127] is the edge node S j Upper container C i The migration probability based on resource balance is expressed as:
[0128]
[0129] For n edge nodes and m containers in the scenario, an N×M dimensional complementarity matrix model is constructed to determine the selection of nodes to which containers can be migrated, thus avoiding resource fragmentation caused by container migration;
[0130] The node complementarity matrix model is:
[0131] corr(i,j) represents the resource complementarity between the i-th container and the j-th edge node based on similarity; the value range of corr(i,j) is usually between 0 and 1. The closer the value of corr(i,j) is to 1, the higher the complementarity of resource demand characteristics between the i-th container and the j-th edge node; conversely, the smaller the complementarity between the i-th container and the j-th edge node, the closer corr(i,j) is to 0. The calculation formula is:
[0132] d i,j Indicates the resource similarity between the i-th container and the j-th edge node;
[0133] Among them, S k Indicates the variance of a certain dimension of resources in the edge node.
[0134] Container migration includes the migration of business data. The execution progress of the business affects the amount of data that needs to be transmitted. The faster the execution progress of the business, the smaller the amount of data that needs to be transmitted.
[0135] is the edge node Sj Upper container C i The migration probability based on the amount of migrated data is calculated as follows:
[0136]
[0137] t i is container C i The time it has been executed.
[0138] According to the set of resource fragmentation nodes Φ, the initial deployment matrix X of the container on each node is obtained as [x ij ] I×J , where when container C i Deployed on edge node S j When on, x i,j =1, otherwise 0, the expression is:
[0139]
[0140] Define edge node S j The initial container set deployed on is V(S j ).
[0141] Step 4: Build a container migration model that comprehensively considers resource balance, data center energy consumption, and migration cost, and use the binary gray wolf optimization algorithm to solve it to obtain the migration target nodes corresponding to all containers in the container initial deployment matrix;
[0142] First, the container migration cost is defined as the container migration time, which includes the data transmission time and container downtime during the migration process; the total migration time of all containers is Mig cost , that is, the container migration cost calculation formula is as follows:
[0143]
[0144]
[0145] Describes the migration time of the i-th container from the j-th edge node to the j′-th edge node, where Data i The amount of data that needs to be migrated during container migration. is the memory of the i-th container, B jj′ is the communication bandwidth between the jth edge node and the j′th edge node.
[0146] M jj′ is the migration decision of whether the i-th container is migrated from the j-th edge node to the container on the j′-th edge node. If yes, then M jj′ =1, otherwise Mjj′ =0; that is:
[0147]
[0148] Then, the resource balance Balance includes the resource load balance U b and resource fragmentation level V b ; The calculation formula is:
[0149] Balance=U b +V b
[0150] Among them, the load balancing degree U b The calculation formula is: is the load average of all edge nodes;
[0151] Resource fragmentation level V b It is measured by the ratio of the remaining resources of edge nodes, and the calculation formula is: is the edge node S j The degree of resource fragmentation on:
[0152] is the edge node S j The remaining resource rate of the k-th resource, Describes the edge node S j The remaining amount of the k-th resource; Describes the total amount of the kth resource on edge node j; is the edge node S j The average remaining resource rate of all resources above.
[0153] Next, according to the system energy consumption model, the main factor affecting the energy consumption of the mobile edge computing network is the CPU utilization, so the system energy consumption En within the time T is calculated as:
[0154]
[0155] P dc (t) represents the instantaneous power of the system at time t; P j (t) is the edge node S corresponding to time t j The instantaneous power:
[0156] in, is the edge node S j Idle power consumption, is the edge node S j Maximum power consumption, U j (t) is the edge node S at time t jThe CPU utilization of the corresponding server, N e is the number of edge nodes in the scene.
[0157] Finally, using the total migration time of the container, resource balance and system energy consumption, in order to reduce the cost of container migration and reduce system energy consumption, the container migration problem is described as a multi-objective optimization problem P under QoS constraints:
[0158] P:min{αMig cost +βBalance+γEn}
[0159] st C1:
[0160] C2:
[0161] C3:
[0162] Among them, α, β, γ are the weight coefficients corresponding to each item, and they satisfy α+β+γ=1;
[0163] Constraint C1 stipulates that the total demand for a certain type of resource by the containers deployed on the edge nodes must be lower than the maximum capacity of the resource of that type;
[0164] Constraint C2 indicates that the total delay of calculating the tasks carried by the container to be migrated should not exceed the delay threshold that the task can tolerate; Indicates whether the i-th container is migrated from the j-th edge node to the j′-th edge node. If so, otherwise,
[0165] Constraint C3 stipulates that a container can only be mapped to one edge node.
[0166] The improved binary grey wolf algorithm solves the multi-objective optimization problem P, reduces the cost of container migration, reduces system energy consumption and reduces the degree of resource fragmentation, and obtains the target edge node S l ;
[0167] S l =argmin{αMig cost +βBalance+γEn}
[0168] st C1,C2,C3
[0169] Since the gray wolf optimization algorithm solves the problem of continuous random variables, and the model defined in the container migration process is in binary form, the gray wolf optimization algorithm needs to be improved to adapt to the optimization of discrete problems. Generally, in the process of initializing the population, a random initialization method is adopted to generate the initial population. Although this method can quickly obtain the initial solution, the problem is that it is easy to fall into the local optimal solution when solving complex optimization. After research, it is found that chaotic variables have the characteristics of uncertainty, regularity and ergodicity. Compared with random sequences, chaotic sequences can realize ergodic search of the search space, and to a certain extent, they can avoid falling into the local optimal solution by increasing a certain search range.
[0170] This embodiment uses the Piecewise mapping function as the chaotic function to initialize the population. The function is as follows:
[0171]
[0172] Among them, P∈(0,1) is the control parameter, x∈(0,1);
[0173] Container migration decision jj′ It is a 0-1 decision, 1 represents the migration of the i-th container from the j-th edge node to the j′-th edge node; 0 represents no migration; in the gray wolf optimization algorithm, the wolf pack uses continuous positions in the global search space to hunt prey, which is used to deal with continuous problems. However, in binary space, this position update method is not applicable. In order to better solve the adaptation problem of the wolf pack algorithm in binary, it is necessary to use the position information of α, β, δ to realize the conversion of the wolf pack position between 0 and 1 to solve the discrete problem; the conversion function used in the present invention is:
[0174]
[0175] Where D represents the distance between the gray wolf and its prey, and A is the coefficient vector.
[0176] Step 5: Each initial node migrates each container to the corresponding migration target node according to the migration order of priority until each initial node achieves resource balance.
[0177] In the mobile edge computing network, the container is small in size, deployed quickly, and one of the characteristics of the container environment is frequent creation and destruction. Therefore, reasonable container migration, that is, making full use of node resources, can achieve the purpose of energy saving. The present invention uses container migration calculation to reduce the energy consumption of the data center in the mobile edge computing network in view of resource imbalance; first, the edge server load is detected, the resource balance of the current edge node is calculated, and the edge node that needs to be migrated in the current system is found, and then the priority of the container to be migrated on each edge node, such as the container migration strategy, is calculated, and a container migration model based on multi-objective optimization of migration cost, resource balance and energy consumption is built, and the node complementarity matrix is used to find the target edge node to be migrated by the container. After satisfying the constraints of resource balance, migration time and system energy consumption, the container is migrated in order according to the priority to achieve resource balance of each initial edge node. Finally, the improved binary gray wolf optimization algorithm that reduces resource fragmentation and energy consumption is used for verification, so that the container migration adapts to 0-1 decision-making.
[0178] Example:
[0179] In the edge computing network, the tasks carried by the container are offloaded to the edge node for computing. In order to balance the resources in the network and reduce the energy consumption of the data center, it is necessary to migrate the containers on the node. Considering the migration cost generated during the migration process, the following is constructed: Figure 2 The container migration model shown in the figure; there are n edge nodes in the edge computing network, each edge node has multiple types of resources such as CPU, memory, and hard disk; each container in m containers is defined to carry a task.
[0180] Considering the imbalance of edge node resources and high energy consumption of the system in the current scenario, container migration can solve this problem. At the same time, the migration process will also generate information interaction between edge nodes and containers and the migration cost of containers. During the migration process, the main research goal is to reduce the migration time as much as possible while reducing system energy consumption and balancing resource balance.
[0181] It is necessary to establish a multi-objective decision model based on migration time, resource balance, node complementarity and energy consumption to achieve multi-objective optimization results that reduce migration time and system energy consumption. The migration problem is divided into the following sub-problems: ① Node overload detection, select the host that needs to migrate the container. ② Detection of underloaded nodes, wait for the container migration to be completed, and put the node into sleep state to reduce energy consumption. ③ Select the container to be migrated from the overloaded and underloaded nodes. ④ Use different migration strategies for multi-objective optimization. ⑤ Determine whether to migrate the container and which edge node to migrate the container to.
[0182] In mobile edge computing, the resource dimensions of edge nodes are multidimensional. In order to improve the resource utilization of the entire system, in addition to considering the load of edge nodes, the degree of resource fragmentation of edge nodes should also be considered. When the load of edge nodes is too high, the lack of computing resources will cause the application performance of containers deployed on the edge nodes to decline. When the load of edge nodes is too low, edge nodes are idle, which will increase the energy consumption of data centers. The higher the degree of resource fragmentation, the easier it is to generate resource fragments.
[0183] Edge node resource fragmentation refers to the huge difference in resource utilization of different types of nodes. When a certain resource is consumed in large quantities, resulting in very little remaining resources, while another resource is consumed in small quantities and has more remaining resources, the edge node cannot deploy new containers due to the lack of a certain resource, resulting in unavailability of remaining resources, causing resource waste. Therefore, resource fragmentation occurs when resource edge nodes are prone to overload.
[0184] Similarity measurement is a measure that comprehensively evaluates the similarity between two objects. Commonly used methods for calculating similarity are distance and angle. This embodiment uses Euclidean distance as the similarity between the resources required by the calculation container and the remaining resources of the edge node.
[0185] Cluster energy optimization can be divided into two levels: hardware and software. Hardware-level optimization, such as merging traffic, sleeping idle switches or links, and hot data concentration, is suitable for cloud service tasks, but not for power simulation of high-performance computing tasks. Cluster energy optimization based on software methods is mainly based on the scheduling and migration of virtual machines and containers.
[0186] There are many factors that affect system energy consumption. In edge computing networks, the energy consumption of data centers mainly depends on factors such as CPU, memory, and disk.
[0187] In this embodiment, an edge computing scenario with 10 edge servers and 50 to 200 containers is simulated; the specific parameters are shown in Table 1 below:
[0188] Table 1
[0189]
[0190] Based on the container migration strategy (CMS) proposed by the present invention based on the balance of mobile edge network resources to reduce the cost of container migration and reduce the energy consumption of data centers in edge networks, the improved Binary Grey Wolf Optimizer (BGWO) is compared with other classic heuristic algorithms, namely, Genetic Algorithm (GA) and Ant Colony Algorithm (ACA); Figure 3 As shown in the figure, the relationship between the optimization target value and the number of iterations under different algorithms is compared. The smaller the optimization target value, the better the algorithm performance. The figure reflects the performance characteristics of the three algorithms as the number of iterations increases. It can be seen from the figure that the convergence speed of ACA and GA is slow, and the BGWO algorithm converges faster. This also indirectly proves that the gray wolf optimization algorithm has the characteristic of fast convergence. Among the three algorithms, the BGWO algorithm can effectively avoid the local optimal problem by changing the initialization of the chaotic sequence. When optimizing the target value, the performance of the BGWO algorithm is better than that of AG and ACA.
[0191] like Figure 4 As shown, from the comparative analysis of the energy consumption of the mobile edge computing network, the energy consumption of the mobile edge computing network is proportional to the CPU utilization of the edge node. It can be seen from the figure that before the container migration, the system energy shows a slight upward trend. As the number of containers increases, the CPU utilization of the edge node also increases. This embodiment integrates the containers in the system through container migration, and migrates the containers to the edge nodes with the smallest energy consumption growth, so that the energy consumption of the entire system is reduced.
[0192] like Figure 5 As shown in the figure, the migration cost of container migration under different migration strategies is described. As the number of containers increases, the resource utilization of edge nodes increases and the resource balance of the mobile edge computing network deteriorates, resulting in more containers needing to be migrated to relatively idle edge nodes to ensure the latency constraints of the services carried by the containers. As can be seen from the figure, the migration strategy proposed in this embodiment has a smaller migration cost than the migration cost based on the minimum migration time (MMT), resource balance (RB) and random migration (RS) strategies.
[0193] like Figure 6 As shown in the figure, the resource balance under different migration strategies is described. The resource balance is measured by the balance of the remaining resources of the edge nodes in the mobile edge computing network and the load balance between the edge nodes. It can be seen from the figure that the strategy proposed in the present invention reduces the container migration cost and reduces the system energy consumption while balancing the resource balance of the nodes.
[0194] like Figure 7 As shown in FIG. 1 , the optimization target values under different migration strategies are described, and the optimization target is the weighted impact of migration cost, system energy consumption, and resource balance. As can be seen from the figure, the migration strategy proposed in the present invention has better performance than the strategies based on minimum migration time (MMT), resource balance (RB), and random migration (RS).
Claims
1. A multi-objective optimization container migration method based on resource balance, characterized in that: The specific steps are as follows: First, a container migration network scenario is built, including n edge nodes and m containers. Each edge node deploys its own corresponding initial container. By detecting the load of each edge node and calculating the resource utilization balance of each node, the initial node that needs to be migrated is obtained. The set of edge nodes is S = {S1, S2, ... S j ,...,S n }; The set of containers is C = {C1, C2, ... C i ,...,C m }; For the jth edge node S j , the resource vector owned by the node is k represents the resource dimension on the edge node; the resource utilization of the node is represents the utilization rate of the kth resource of the jth edge node; The i-th container is defined as in Represents the various resources required by the i-th container, where Represents the kth resource required by the i-th container; represents the computing delay when the i-th container meets the resource requirements, Tol i Represents the latency threshold of the task carried by the i-th container; The initial node meets the following requirements: Greater than the load constant c2 and resource utilization balance Λ j Greater than the resource balance constant c1; Then, the probability of migration of each deployed container on the initial node is calculated according to the container migration strategy, and the priority set of the containers to be migrated on each initial node is obtained respectively, and all the containers to be migrated are combined into a container initial deployment matrix; Next, we built a container migration model that comprehensively considered resource balance, data center energy consumption, and migration costs, and used the binary gray wolf optimization algorithm to solve it, obtaining the migration target nodes corresponding to all containers in the container initial deployment matrix. The binary gray wolf optimization algorithm uses the Piecewise mapping function to initialize the population, and the function is as follows: Among them, P∈(0,1) is the control parameter, x∈(0,1); Container migration decision jj′ For 0-1 decision, the transformation function is used to transform the wolf pack position between 0 and 1 to solve the discrete problem; the transformation function is: Where D represents the distance between the individual gray wolf and the prey, and A is the coefficient vector; Finally, each initial node migrates each container to the corresponding migration target node in the order of priority until each initial node achieves resource balance.
2. A multi-objective optimization container migration method based on resource balance as claimed in claim 1, characterized in that: The edge node S j Load The calculation formula is: in, is the edge node S j Have resources The utilization rate, x i,j Indicates whether the i-th container is deployed on the j-th edge node. If it is x i,j =1, otherwise 0; γ k is the resource utilization At the edge node S j The load weight must satisfy constraint; The resource utilization balance degree Λ j The calculation formula is: Λ j ∈(0,1), edge node S j The more balanced the resource utilization, the j The closer it is to 0, otherwise Λ j The closer it is to 1; is the edge node S j The average utilization rate of various resources owned, 3. A multi-objective optimization container migration method based on resource balance as claimed in claim 2, characterized in that: The edge node S j Upload container C to be migrated i The priority P i,j , the calculation formula is as follows: Where η is the weight based on migration time, ρ is the weight based on resource balance, and σ is the weight based on the amount of migrated data, satisfying η+ρ+σ=1; is the edge node S j Upper container C i Migration probability based on migration time; Among them, λ is the value of the container C i The weight factor of the CPU utilization of container C on the migration downtime is μ. i The weight factor of the memory utilization on the migration downtime satisfies λ+μ=1; Represents container C i The demand for CPU resources, Represents container C i The amount of memory resources required; is the edge node S j Upper container C i The migration probability based on resource balance is expressed as: corr(i,j) represents the resource complementarity based on similarity: d i,j Indicates the resource similarity between the i-th container and the j-th edge node; S k Indicates the variance of a certain dimension of resources in the edge node; is the edge node S j Upper container C i The migration probability based on the amount of migrated data is calculated as follows: t i It is container C i The time it has been executed.
4. The multi-objective optimization container migration method based on resource balance as claimed in claim 1, characterized in that: The construction process of the container migration model is as follows: First, the container migration cost is defined as the container migration time, which includes the data transmission time and container downtime during the migration process. The calculation formula is as follows: Describes the migration time of the i-th container from the j-th edge node to the j′-th edge node, where Data i The amount of data that needs to be migrated during container migration. is the memory of the i-th container, B jj′ is the communication bandwidth between the jth edge node and the j′th edge node; Mig cost is the total migration time of all containers; M jj′ The migration decision of the container on the jth edge node to determine whether the i-th container should be migrated from the j-th edge node to the j′th edge node; Then, the resource balance includes resource load balance and resource fragmentation; The calculation formula is: Balance=U b +V b Among them, the load balancing degree U b The calculation formula is: Δ Sj is the edge node S j Load; is the load average of all edge nodes; Resource fragmentation level V b The calculation formula is: V sj is the edge node S j The degree of resource fragmentation on: is the edge node S j The remaining resource rate of the k-th resource, Describes the edge node S j The remaining amount of the k-th resource; Describes the total amount of the kth resource on edge node j; x i,j Indicates whether the i-th container is deployed on the j-th edge node; is the edge node S j The average remaining resource rate of all resources above; Next, the system energy consumption En in the system T time is calculated as follows: P dc (t) represents the instantaneous power of the system at time t; P j (t) is the edge node S corresponding to time t j The instantaneous power of Finally, using the total migration time of the container, resource balance and system energy consumption, a multi-objective optimization problem P under QoS constraints is established: Among them, α, β, γ are the weight coefficients corresponding to each item, and they satisfy α+β+γ=1; Constraint C1 stipulates that the total demand for a certain type of resource by the containers deployed on the edge nodes must be lower than the maximum capacity of the resource of that type; Constraint C2 indicates that the total delay of calculating the tasks carried by the container to be migrated should not exceed the delay threshold that the task can tolerate; Indicates whether the i-th container is migrated from the j-th edge node to the j′-th edge node; Constraint C3 stipulates that a container can only be mapped to one edge node.
Citation Information
Patent Citations
Electric power Internet of Things container migration method for edge network load balancing
CN111694636A
