A dynamic virtual machine integration method and system based on balancing energy consumption and service quality

Abstract

The present invention discloses a virtual resource integration method and system based on green energy saving, and a dynamic virtual machine integration method based on balancing energy consumption and service quality, including: 1) physical host load prediction: predicting the workload of the physical host at the next moment based on a prediction algorithm based on a cubic exponential smoothing model and an Elman neural network model; 2) physical host load state detection: identifying the current load state of the physical host through a hybrid load detection algorithm; 3) virtual machine selection: selecting virtual machines that need to be migrated on non-suitable hosts based on a virtual machine selection algorithm based on CPU and memory perception; 4) virtual machine placement: a virtual machine placement algorithm based on resource demand expansion, selecting a suitable physical host for placing the migrated virtual machine according to the resource demand of the virtual machine migration queue #imgabs0# and the resource information of the suitable hosts in the data center. The present invention reduces the number of virtual machine migrations, and maintains a high level of service quality while consuming lower energy.

Description

Technical Field

[0001] The invention relates to a dynamic virtual machine integration method and system based on balancing energy consumption and service quality, belonging to the technical field of dynamic virtual machine integration. Background Art

[0002] With the rapid development of cloud computing technology, Infrastructure as a Service (IaaS) has become an important service model. Users can rent resources including servers, networks, storage, etc. from IaaS providers on demand. Data centers with functions such as elastic resource supply, dynamic configuration of virtual services, virtualization and management of infrastructure resources have become an important carrier for building IaaS services.

[0003] However, in the past few years, extreme energy consumption in cloud data centers has become a considerable problem. Typically, data center energy is wasted due to a variety of reasons, including inefficient network equipment, server utilization, and data center cooling systems. According to a 2013 report by Gartner, cloud data center electricity consumption is often huge, equivalent to the electricity consumption of 25,000 homes. And between 2011-2035, global data center power demand is expected to grow by more than 66%. According to statistics, data center utilization is very low. For example, the average utilization of data centers is between 12% and 18%. Google's data center utilization has been between 10% and 50%, which is a waste of electricity because idle servers use an average of 70% of their maximum power. Low resource utilization leads to a lot of energy waste and complications, which further worsens resource waste by expanding the capacity of data centers.

[0004] Therefore, high power consumption and low utilization of cloud data centers are challenges facing cloud computing. An effective and commonly used method to solve this problem is virtual machine consolidation. Virtual machine consolidation refers to placing virtual machines on fewer servers through virtual machine migration according to their resource requirements, thereby changing some servers to a dormant state, thereby reducing the energy overhead of the data center. However, virtual machine migration increases the cost of computing resources, and large-scale virtual machine migration may cause additional workloads, SLA violations, and considerable energy consumption. At the same time, services will be suspended during virtual machine migration, and long-term migration may further affect service quality. Summary of the invention

[0005] The present invention provides a virtual resource integration system based on green energy saving. The model defines the specific process and related attributes of virtual resource integration. On the basis of the model, the present application proposes a dynamic virtual machine integration method based on balancing energy consumption and service quality to achieve efficient integration of virtual resources. When the method is applied to virtual machine integration, it not only reduces the number of virtual machine migrations, but also maintains a high level of service quality (Quality of Service, QoS) while consuming lower energy, thereby achieving a balance between energy consumption and QoS.

[0006] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0007] A dynamic virtual machine integration method based on balancing energy consumption and service quality includes the following steps:

[0008] 1) Physical host load prediction: The prediction algorithm HCESEA based on the triple exponential smoothing model and the Elman neural network model is used to predict the workload of the physical host at the next moment;

[0009] 2) Physical host load status detection: The hybrid load detection algorithm HLDA is used to identify the current load status of the physical host;

[0010] 3) VM selection: The VM selection algorithm CM-VMSA based on CPU and memory awareness is used to select VMs that need to be migrated on non-suitable hosts.

[0011] 4) Virtual machine placement: The virtual machine placement algorithm RDS-VMPA based on resource demand scaling is based on the virtual machine migration queue. The resource requirements of the virtual machine and the resource information of suitable hosts in the data center are combined to select a suitable physical host for accommodating the migrated virtual machine.

[0012] The above method realizes the efficient integration of virtual resources, reduces the number of virtual machine migrations and the number of servers used, improves the service resource utilization rate of the data center, and reduces energy consumption, thereby meeting the energy-saving and environmental protection needs of the data center.

[0013] The above method takes into account the possibility of potential overload and the resource demand expansion of the physical host into consideration during the virtual machine placement, which significantly improves the accuracy of selecting a suitable host.

[0014] In the above step 1), in view of the dynamic and uncertain characteristics of the physical host load data, this application proposes a prediction algorithm HCESEA (Predictional algorithm based on Hybrid cubic exponential smoothing model and Elman model, HCESEA) based on the cubic exponential smoothing model and the Elman neural network model to predict the workload of the physical host at the next moment. The prediction algorithm uses the cubic exponential smoothing model CES (Cubic Exponential Smoothing Model, CES) for prediction, and then uses the Elman neural network model ENN (Elman Neural Network Model, ENN) to predict the error of the cubic exponential smoothing model CES, and finally obtains a prediction value after correcting the error. This not only alleviates the impact of model parameters on the overall performance, but also the ENN model is a prediction of the CES model error, and its prediction accuracy is better than the prediction accuracy of the original data set, thereby further improving the prediction performance of the model. The following is an introduction to the model involved in the HCESE model.

[0015] Triple exponential smoothing model CES:

[0016] The calculation formula of the triple exponential smoothing value is as follows:

[0017]

[0018] Where: α represents the smoothing coefficient, 0<α<1; represents the exponential smoothing value of period t; represents the quadratic exponential smoothing value of period t; represents the triple exponential smoothing value for period t.

[0019] The CES prediction model is as follows:

[0020] Y t+T =A t +B t T+C t T 2

[0021]

[0022] Among them, T is the forecast period, A t , B t , C t is the prediction parameter.

[0023] Elman neural network model ENN:

[0024] The basic structure of the Elman network consists of four parts: input layer, hidden layer, output layer and association layer. Unlike general neural networks, the Elman neural network has an additional association layer, and the input of the association layer comes from the output of the hidden layer. This internal feedback mechanism enhances the network's ability to process dynamic time data.

[0025] Mathematical model of Elman neural network:

[0026] x(k)=f(w1x c (k)+w2u(k-1)) (4)

[0027] x c (k) = x(k-1) (5)

[0028] y(k)=g(w3x(k)) (6)

[0029] Among them, w1 is the connection weight matrix between the association layer and the hidden layer, w2 is the connection weight matrix between the input layer and the hidden layer, w3 is the connection weight matrix between the hidden layer and the output layer, x c (k) and x(k) represent the outputs of the association layer and the hidden layer respectively, y(k) represents the output of the output layer, f is the activation function, which is usually the Sigmoid function, u is the input vector, and g is the transfer function of the output neuron;

[0030] Prediction algorithm HCESEA:

[0031] Assume that the data set L′ h = {l t1 , l t2 ,…,l tn} for t1 to t n The actual load of physical host h during the time period, Y′ h Represents the triple exponential smoothing model CES based on the data set L′ h The load prediction sequence of physical host h with a length of m is obtained. Then the load prediction sequence error of physical host h can be expressed as E h =Y′ h -L′ h ={e1, e2, ..., e m}; Elman neural network model ENN according to E h Get the error correction sequence E′ h , then the modified load prediction of the physical host h is expressed as.

[0032] Assume m CES 、m ENN They are represented by the cubic exponential smoothing model CES and the Elman neural network model ENN, respectively. CES 、xENN are represented as the input vectors of the two models respectively, then at t n+1 At this moment, the output of the model can be expressed as:

[0033] Y * =m CES (x CES )-m ENN (x ENN ).

[0034] In the above step 2), since the load of each physical host in the data center changes dynamically, the load of different physical hosts at the same time will be different. To this end, the present application proposes to identify the load status of the current physical host through a hybrid load detection algorithm HLDA (Hybrid load detection algorithm, HLDA), first obtain the expected load and real-time load of each physical host in the data center, and then divide the load status of the physical host into the following four states according to the set threshold: underload state, appropriate load state, potential overload state and overload state.

[0035] The specific process of the hybrid load detection algorithm HLDA is as follows: When the load of a physical host is higher than the upper threshold, the physical host is added to the overloaded host queue When the load of a physical host falls below the lower threshold, the physical host is added to the underloaded host queue. When the load of a physical host is within the threshold, it is necessary to make a judgment based on the predicted load of the physical host; if the expected load of the physical host is higher than the upper threshold, the physical host is added to the potential overload host queue. Otherwise, the physical host is added to the appropriate host queue. middle.

[0036] In step 3), virtual machine selection: Real-time migration of virtual machines will have a negative impact on the performance of applications running on the virtual machines, thereby causing some SLA conflicts. In addition, the performance degradation during virtual machine migration is related to the migration time of the virtual machine. Therefore, the present application proposes a virtual machine selection algorithm based on CPU and memory perception, CM-VMSA (Virtual machine selection algorithm based on CPU and memory perception, CM-VMSA), to select virtual machines that need to be migrated on non-suitable hosts. This algorithm can reduce the migration time of virtual machines as much as possible while reducing the number of virtual machine migrations, thereby improving the quality of service.

[0037] There are three types of physical hosts that need to migrate virtual machines. They are located in the underloaded host queue. Overloaded host queue and potentially overloaded host queues Therefore, this application performs the following operations on these three types of queues:

[0038] Underloaded host queue Add all virtual machines on this type of physical host to the virtual machine migration queue After all virtual machines are migrated to other physical hosts, the node is switched to sleep mode to reduce energy consumption.

[0039] Overloaded host queue First, all virtual machines on this type of physical host are sorted in descending order according to CPU utilization; second, they are sorted in ascending order according to the amount of memory occupied by the virtual machines in the relative order of descending CPU utilization; finally, each virtual machine is tried to be migrated out in turn, and the load L of the physical host is calculated in real time. c , if L c <η u , then stop the above operations and add the virtual machine you are trying to migrate to the virtual machine migration queue

[0040] Potentially overloaded host queues First, all virtual machines on this type of physical host are sorted in descending order according to CPU utilization; second, they are sorted in ascending order according to the amount of memory used by the virtual machines in the relative order of descending CPU utilization; finally, each virtual machine is tried to be migrated out in turn, and the load L of the physical host is calculated in real time. E ,like Then stop the above operation and add the virtual machine you are trying to migrate to is the CPU load occupied by virtual machine i on physical host h.

[0041] In step 4), virtual machine placement: Virtual machine placement algorithm based on resource demand scaling (RDS-VMPA) is used to place virtual machines according to the virtual machine migration queue. The resource requirements of the host and the resource information of the suitable hosts in the data center are combined to select the appropriate physical host for accommodating the migrated virtual machine. RDS-VMPA calculates the workload of all physical hosts that meet the virtual machine resource allocation requirements at future times, divides the resource demand reduction queue and resource demand growth queue according to the forecast situation, sorts the queues in a specific way, and determines the destination host through further screening. The specific algorithm steps are as follows:

[0042] 1) Calculate the resource demand expansion on each physical host. The value is the predicted load of the physical host minus the current workload. This value reflects the changing trend of the virtual machine resource demand on the physical host.

[0043] 2) Calculate the amount of unallocated resources by subtracting the current workload from the total amount of physical host resources, and select the difference between the amount of unallocated resources and the amount of resource demand expansion, that is, the list of physical hosts with estimated remaining resources greater than zero, to form a list of candidate physical hosts;

[0044] 3) If the resource demand expansion amount is a negative value, the physical host is added to the resource demand reduction queue, and the difference between the remaining resources and the resource demand expansion amount is calculated and arranged in descending order; if the resource demand expansion amount is a positive value, the physical host is added to the resource demand expansion queue, and the ratio of the resource demand expansion amount to the remaining resources is calculated as the safety factor SF, and the queue is arranged in ascending order according to the value of the factor;

[0045] 4) After the two queues are generated, if one queue is empty, the head host of the other queue is directly selected as the placement host; if neither is empty, the priority factors δ of the two head hosts are compared.

[0046] A virtual resource integration system based on green energy saving includes a global manager and more than two local managers;

[0047] The global manager is equipped with a scheduling module, a data information module, a load prediction center, a load detection center, a virtual machine selection center, and a virtual machine placement center, and the local manager is equipped with a monitoring unit;

[0048] The local manager collects the real-time CPU and memory load on the physical host and virtual machine through the monitoring unit and sends the data set to the data information module in the global manager;

[0049] The scheduling module in the global manager uses the load prediction center to predict the load status of each physical host at the next moment based on the data set in the data information module;

[0050] The load detection center detects and divides the load status of the physical host by combining the real-time load and expected load of the physical host;

[0051] Select the virtual machines to be migrated from the non-suitable hosts through the virtual machine selection center;

[0052] The virtual machine placement center selects a suitable host for the virtual machine to be migrated, and sends the migration plan to each local manager for migration.

[0053] Virtual resource integration model:

[0054] Definition 1: The physical host in the data center is represented by Q H ={h1, h2, ..., h i ,…,h m}, h i represents the i-th physical host;

[0055] Definition 2: Physical host h i The virtual machine queue in is represented as v j represents the jth virtual machine;

[0056] Definition 3: During the virtual machine migration process, about 10% CPU overhead will be generated, which will cause the performance of the virtual machine to degrade. Therefore, each virtual machine migration may cause some SLA violations. This application needs to reduce the number of virtual machine real-time migrations as much as possible to ensure the quality of service provided. The formula for virtual machine migration time and performance degradation is as follows:

[0057]

[0058]

[0059] in, It is a virtual machine v j The time it took to complete the migration, It is a virtual machine v j The amount of memory used, B j is the available network bandwidth, is the performance degradation caused by virtual machine migration, t0 is the migration start time, It is a virtual machine v j CPU utilization;

[0060] Definition 4: The energy consumption of a data center is expressed as:

[0061]

[0062] Among them, EC is the energy consumption of the data center, For the physical host h i The load at time t, E is the power consumption corresponding to the physical host load.

[0063] The definitions of the symbols in this application are shown in Table 1.

[0064] Table 1 Symbol Definition

[0065]

[0066]

[0067] The technologies not mentioned in the present invention are all referred to the prior art.

[0068] Aiming at the existing problems of virtual machine integration, the present invention firstly constructs a virtual resource integration system based on green energy saving. The model defines the specific process and related attributes of virtual resource integration. Secondly, based on the model, the present application proposes a dynamic virtual machine integration method based on balancing energy consumption and service quality to achieve efficient integration of virtual resources. The strategy includes four parts: physical host load prediction, physical host load status detection, virtual machine selection and virtual machine placement.

[0069] Physical host load prediction: By sensing the load information in the data center, this application proposes a prediction model based on the cubic exponential smoothing model and Elman neural network model (HCESE) to predict the workload of the physical host at the next moment. This model performs error prediction and correction based on CES, so that the load status of the physical host can be predicted more accurately.

[0070] Physical host load status detection: This application proposes a hybrid load detection algorithm (HLDA) to identify the current load status of the physical host, and divides the load status of the physical host into the following four states: underload state, appropriate load state, potential overload state and overload state. This method divides the physical host state in detail, which can reduce potential SLA violations and thus improve service quality.

[0071] Virtual machine selection: This application proposes a virtual machine selection algorithm based on CPU and memory perception (CM-VMSA) to select virtual machines that need to be migrated on non-suitable hosts. This algorithm can reduce the number of virtual machine migrations and reduce the migration time of virtual machines as much as possible, thereby improving service quality.

[0072] Virtual machine placement: This application proposes a virtual machine placement algorithm based on resource demand scaling (RDS-VMPA), which selects a suitable physical host for migrating virtual machines based on the resource demand of the virtual machine migration queue and the resource information of the suitable hosts in the data center. The suitable hosts selected by the algorithm take into account the resource demand scaling of the physical host, which can effectively prevent overload caused by workload fluctuations, thereby reasonably allocating virtual machines to physical hosts and improving the resource utilization of the data center.

[0073] The present application proposes a dynamic virtual machine integration method based on balancing energy consumption and service quality, which not only reduces the number of virtual machine migrations and energy consumption, but also maintains a high level of service quality (Quality of service, QoS), achieving a balance between energy consumption and QoS. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a schematic diagram of the virtual resource integration system based on green energy saving of the present invention.

[0075] Figure 2 Schematic diagram of the ENN model of the present invention.

[0076] Figure 3 This is the structural diagram of HCESEA of the present invention. DETAILED DESCRIPTION

[0077] In order to better understand the present invention, the content of the present invention is further explained below in conjunction with the embodiments, but the content of the present invention is not limited to the following embodiments.

[0078] The present invention proposes a dynamic virtual machine integration method based on balancing energy consumption and service quality. The method first constructs a virtual resource integration system based on green energy saving. The model defines the specific process and related attributes of virtual resource integration. Secondly, based on the model, the present application proposes a dynamic virtual machine integration method based on balancing energy consumption and service quality to achieve efficient integration of virtual resources.

[0079] Figure 1 The figure is a schematic diagram of the green and energy-saving virtual resource integration system described in the present invention. The model consists of a global manager and several local managers. The local manager collects the real-time load of the CPU and memory on the physical host and virtual machine through the monitoring unit, and sends the data set to the data information module in the global manager. The scheduling process steps are as follows:

[0080] Step 101: The scheduling module in the global manager first uses the load prediction center to predict the load status of each physical host at the next moment based on the data set in the data information module;

[0081] Step 102: The load detection center detects and divides the load status of the physical host in combination with the real-time load and the expected load of the physical host;

[0082] Step 103: Select virtual machines to be migrated from the non-suitable hosts through the virtual machine selection center.

[0083] Step 104: The virtual machine placement center selects a suitable host for the virtual machine to be migrated, and sends the migration plan to each local manager for migration.

[0084] A dynamic virtual machine integration method based on balancing energy consumption and service quality, characterized in that it includes the following steps:

[0085] 1) Physical host load prediction:

[0086] In view of the dynamic and uncertain characteristics of physical host load data, this application proposes a prediction algorithm based on hybrid cubic exponential smoothing model and Elman neural network model HCESEA (Prediction algorithm based on Hybrid cubic exponential smoothing model and Elman model, HCESEA) to predict the workload of the physical host at the next moment. The prediction algorithm uses the cubic exponential smoothing model CES (Cubic Exponential Smoothing Model, CES) for prediction, and then uses Figure 2 The Elman Neural Network Model ENN (ENN) shown in the figure predicts the error of the triple exponential smoothing model CES, and finally obtains a predicted value after the error is corrected. This not only alleviates the impact of model parameters on the overall performance, but also the ENN model predicts the error of the CES model, and its prediction accuracy is better than the prediction accuracy of the original data set, thereby further improving the prediction performance of the model. The following is an introduction to the models involved in the HCESE model;

[0087] The calculation formula of the triple exponential smoothing value is as follows:

[0088]

[0089] Where: α represents the smoothing coefficient, 0<α<1; represents the exponential smoothing value of period t; represents the quadratic exponential smoothing value of period t; represents the triple exponential smoothing value for period t.

[0090] The CES prediction model is as follows:

[0091] Y t+T =A t +B t T+C t T 2

[0092]

[0093] Among them, T is the forecast period, A t , B t , C t is the prediction parameter.

[0094] Elman neural network model ENN:

[0095] The basic structure of the Elman network consists of four parts: input layer, hidden layer, output layer and association layer. Unlike general neural networks, the Elman neural network has an additional association layer, and the input of the association layer comes from the output of the hidden layer. This internal feedback mechanism enhances the network's ability to process dynamic time data.

[0096] Mathematical model of Elman neural network:

[0097] x(k)=f(w1x c (k)+w2u(k-1)) (4)

[0098] x c (k) = x(k-1) (5)

[0099] y(k)=g(w3x(k)) (6)

[0100] Among them, w1 is the connection weight matrix between the association layer and the hidden layer, w2 is the connection weight matrix between the input layer and the hidden layer, w3 is the connection weight matrix between the hidden layer and the output layer, x c (k) and x(k) represent the outputs of the association layer and the hidden layer respectively, y(k) represents the output of the output layer, f is the activation function, which is usually the Sigmoid function, u is the input vector, and g is the transfer function of the output neuron;

[0101] Prediction algorithm HCESEA:

[0102] Assume that the data set L′ h = {l t1 , l t2 ,…,l tn} for t1 to t nThe actual load of physical host h during the time period, Y′ h Represents the triple exponential smoothing model CES based on the data set L′ h The load prediction sequence of physical host h with a length of m is obtained. Then the load prediction sequence error of physical host h can be expressed as E h =Y′ h -L′ h ={e1, e2, ..., e m}; Elman neural network model ENN according to E h Get the error correction sequence E′ h , then the modified load prediction of physical host h is expressed as The structure of the algorithm is as follows Figure 3 shown.

[0103] Assume m CES 、m ENN They are represented by the cubic exponential smoothing model CES and the Elman neural network model ENN, respectively. CES 、x ENN are represented as the input vectors of the two models respectively, then at t n+1 At this moment, the output of the model can be expressed as:

[0104] Y * =m CES (x CES )-m ENN (x ENN ).

[0105] Figure 3 This is the HCESEA structure diagram. The specific steps of physical host prediction HCESEA are as follows:

[0106]

[0107]

[0108] 2) Physical host load status detection:

[0109] Since the load of each physical host in the data center changes dynamically, the load of different physical hosts at the same time may be different. Therefore, the present application proposes to identify the load status of the current physical host through a hybrid load detection algorithm HLDA (Hybrid load detection algorithm, HLDA), first obtain the expected load and real-time load of each physical host in the data center, and then divide the load status of the physical host into the following four states according to the set threshold: underload state, appropriate load state, potential overload state and overload state.

[0110] The specific process of the hybrid load detection algorithm HLDA is as follows: when the load of a physical host is higher than the upper threshold, the physical host is added to the overloaded host queue When the load of a physical host falls below the lower threshold, the physical host is added to the underloaded host queue. When the load of a physical host is within the threshold, it is necessary to make a judgment based on the predicted load of the physical host; if the expected load of the physical host is higher than the upper threshold, the physical host is added to the potential overload host queue. Otherwise, the physical host is added to the appropriate host queue. middle.

[0111] The specific steps of physical host load status detection HLDA are as follows:

[0112]

[0113]

[0114] 3) Virtual machine selection: Real-time migration of virtual machines will have a negative impact on the performance of applications running on the virtual machines, resulting in some SLA conflicts. In addition, the performance degradation during virtual machine migration is related to the migration time of the virtual machine. Therefore, this application proposes a virtual machine selection algorithm based on CPU and memory perception, CM-VMSA (Virtualmachine selection algorithm based on CPU and memory perception, CM-VMSA), to select virtual machines that need to be migrated on non-suitable hosts. This algorithm can reduce the migration time of virtual machines as much as possible while reducing the number of virtual machine migrations, thereby improving the quality of service;

[0115] There are three types of physical hosts that need to migrate virtual machines. They are located in the underloaded host queue. Overloaded host queue and potentially overloaded host queues Therefore, this application performs the following operations on these three types of queues:

[0116] Underloaded host queue Add all virtual machines on this type of physical host to the virtual machine migration queue After all virtual machines are migrated to other physical hosts, the node is switched to sleep mode to reduce energy consumption.

[0117] Overloaded host queue First, all virtual machines on this type of physical host are sorted in descending order according to CPU utilization; second, they are sorted in ascending order according to the amount of memory occupied by the virtual machines in the relative order of descending CPU utilization; finally, each virtual machine is tried to be migrated out in turn, and the load L of the physical host is calculated in real time. c , if L c <η u , then stop the above operations and add the virtual machine you are trying to migrate to the virtual machine migration queue

[0118] Potentially overloaded host queues First, all virtual machines on this type of physical host are sorted in descending order according to CPU utilization; second, they are sorted in ascending order according to the amount of memory used by the virtual machines in the relative order of descending CPU utilization; finally, each virtual machine is tried to be migrated out in turn, and the load L of the physical host is calculated in real time. E ,like Then stop the above operation and add the virtual machine you are trying to migrate to

[0119] The specific steps for selecting CM-VMSA for a virtual machine are as follows:

[0120]

[0121]

[0122] 4) Virtual Machine Placement:

[0123] Virtual machine placement algorithm based on resource demand scaling (RDS-VMPA) is used to place virtual machines according to the virtual machine migration queue. The resource requirements of the host and the resource information of the suitable hosts in the data center are combined to select the appropriate physical host for accommodating the migrated virtual machine. RDS-VMPA calculates the workload of all physical hosts that meet the virtual machine resource allocation requirements at future times, divides the resource demand reduction queue and resource demand growth queue according to the forecast situation, sorts the queues in a specific way, and determines the destination host through further screening. The specific algorithm steps are as follows:

[0124] 1) Calculate the resource demand expansion on each physical host. The value is the predicted load of the physical host minus the current workload. This value reflects the changing trend of the virtual machine resource demand on the physical host.

[0125] 2) Calculate the amount of unallocated resources by subtracting the current workload from the total amount of physical host resources, and select the difference between the amount of unallocated resources and the amount of resource demand expansion, that is, the list of physical hosts with estimated remaining resources greater than zero, to form a list of candidate physical hosts;

[0126] 3) If the resource demand expansion amount is a negative value, the physical host is added to the resource demand reduction queue, and the difference between the remaining resources and the resource demand expansion amount is calculated and arranged in descending order; if the resource demand expansion amount is a positive value, the physical host is added to the resource demand expansion queue, and the ratio of the resource demand expansion amount to the remaining resources is calculated as the safety factor SF, and the queue is arranged in ascending order according to the value of the factor;

[0127] 4) After the two queues are generated, if one queue is empty, the head host of the other queue is directly selected as the placement host; if neither is empty, the priority factors δ of the two head hosts are compared.

[0128] The specific steps for placing a virtual machine in RDS-VMPA are as follows:

[0129]

[0130]

[0131] In summary, the dynamic virtual machine integration method based on balancing energy consumption and service quality proposed in the present invention is suitable for solving the virtual machine integration problem. This strategy not only reduces the number of virtual machine migrations and energy consumption, but also maintains a high level of service quality, achieving a balance between energy consumption and service quality. This enables cloud service providers to reduce data center costs and improve user service experience, thereby further promoting the development of cloud computing.

[0132] The CloudSim simulation platform is used to evaluate the dynamic virtual machine consolidation method based on balancing energy consumption and quality of service (EQ-DVMCA) proposed in this application and compare it with some benchmark algorithms. We use CloudSim version 4.0, which is an event-driven simulator for simulating cloud computing infrastructure and application services. It supports virtualization resource management and modeling, energy consumption, virtual machine migration, and other functions.

[0133] The experiment simulates a data center consisting of 800 heterogeneous physical hosts, including three types of hosts: HP ProLiant ML110 G4 and HP ProLiant ML110 G5. The host configuration is shown in Table 2.

[0134]

[0135] In the experiment, we selected four types of virtual machines: high CPU medium instance, extra large instance, small instance and micro instance. The properties of the virtual machines are shown in Table 3.

[0136] Table 3 Virtual machine attributes

[0137]

[0138] In order to make the simulation results realistic, we used a part of the data provided by the CoMon project for simulation experiments. After creating PM and VM instances on the CloudSim platform, the PlanetLab dataset was used to generate VM workloads, which were then randomly deployed on PMs according to the VM resource requirements.

[0139] Table 4 Workload information of PlanetLab dataset

[0140]

[0141]

[0142] Benchmark Algorithm Settings

[0143] Including host load detection algorithms: Static Threshold (THR), Interquartile Range (IQR), Local Regression Robust (LRR), Median Absolute Deviation (MAD), virtual machine selection algorithms: Minimum Migration Time (MMT), Random Choice (RC), Maximum Correlation (MC), virtual machine placement algorithm: Power Aware Best Fit Decreasing (PABFD). By combining these algorithms, 12 different virtual machine integration algorithms can be obtained, and each integration method includes a host load detection algorithm, a virtual machine selection algorithm, and a virtual machine placement algorithm. The safety parameters of THR, IQR, LRR, and MAD are set to 0.8, 1.5, 1.2, and 2.5 respectively.

[0144] Performance indicators: number of virtual machine migrations (VMM), SLA violation time per active host (SLATAH), performance degradation due to migration (PDM), service level agreement violations (SLAV), energy consumption (EC), and comprehensive evaluation index of service quality and energy consumption (ESV).

[0145] The experimental results are as follows:

[0146] Table 5 Simulation results of EQ-DVMCA and benchmark algorithms under PlanetLab dataset

[0147]

[0148]

[0149] As can be seen from the above table, EQ-DVMCA performs best in all 6 performance tests. Compared with the baseline algorithm, the EQ-DVMCA of the present application not only significantly reduces the number of virtual machine migrations and energy consumption, but also provides reliable QoS. It can be calculated from Table 10 that compared with the baseline algorithm, the EQ-DVMCA of the present application has improved by 8.40%-21.60% in VMM indicators, 10.83%-36.93% in SLATAH indicators, 33.33%-62.96% in PDM indicators, 47.21%-75.80% in SLAV indicators, 9.78%-27.49% in EC indicators, and 51.97%-82.24% in ESV indicators, that is, the EQ-DVMCA of the present application has significant superiority.

Claims

1. A dynamic virtual machine integration method based on balancing energy consumption and service quality, characterized in that: The steps include: 1) Physical host load prediction: The prediction algorithm HCESEA based on the cubic exponential smoothing model CES and the Elman neural network model ENN is used to predict the workload of the physical host at the next moment; 2) Physical host load status detection: The hybrid load detection algorithm HLDA is used to identify the current load status of the physical host; 3) VM selection: The VM selection algorithm CM-VMSA based on CPU and memory awareness is used to select VMs that need to be migrated on non-suitable hosts. 4) Virtual machine placement: The virtual machine placement algorithm RDS-VMPA based on resource demand scaling is based on the virtual machine migration queue. The resource requirements of the virtual machine and the resource information of the suitable hosts in the data center are combined to select a suitable physical host for accommodating the migrated virtual machine; In step 1), the triple exponential smoothing model CES is used for prediction, and then the Elman neural network model ENN is used to predict the error of the triple exponential smoothing model CES, and finally a prediction value after correcting the error is obtained.

2. The method for dynamic virtual machine integration based on balancing energy consumption and service quality as claimed in claim 1, characterized in that: In step 1), the triple exponential smoothing model CES: The calculation formula of the triple exponential smoothing value is as follows: Where: ɑ represents the smoothing coefficient, 0<ɑ<1; represents the exponential smoothing value of period t; represents the quadratic exponential smoothing value of period t; represents the triple exponential smoothing value of period t; The CES prediction model is as follows: Y t+T =A t +B t T+C t T 2 Among them, T is the forecast period, A t , B t , C t is the prediction parameter; Elman neural network model ENN: The basic structure of the Elman network consists of four parts: input layer, hidden layer, output layer and association layer; Mathematical model of Elman neural network: x(k)=f(w1x c (k)+w2u(k-1)) (4) x c (k)=x(k-1) (5) y(k)=g(w3x(k)) (6) Among them, w1 is the connection weight matrix between the association layer and the hidden layer, w2 is the connection weight matrix between the input layer and the hidden layer, w3 is the connection weight matrix between the hidden layer and the output layer, x c (k) and x(k) represent the outputs of the association layer and the hidden layer respectively, y(k) represents the output of the output layer, f is the activation function, which is usually the Sigmoid function, u is the input vector, and g is the transfer function of the output neuron; Prediction algorithm HCESEA: Assume that the data set L′ h =(l t1 , l t2 ,…,l tn } for t1 to t n The actual load of physical host h during the time period, Y′ h Represents the triple exponential smoothing model CES based on the data set L′ h The load prediction sequence of physical host h with a length of m is obtained. Then the load prediction sequence error of physical host h can be expressed as E h =Y′ h -L′ h ={e1, e2, ..., e m }; Elman neural network model ENN according to E h Get the error correction sequence E′ h , then the modified load prediction of physical host h is expressed as Assume m CES 、m ENN They are represented by the cubic exponential smoothing model CES and the Elman neural network model ENN, respectively. CES 、x ENN Respectively represent the input vectors of these two models, then at t n+1 At this moment, the output of the model can be expressed as: Y * =m CES (x CES )-m ENN (x ENN )。 3. The method for dynamic virtual machine integration based on balancing energy consumption and service quality as claimed in claim 2, characterized in that: In step 2), the expected load and real-time load of each physical host in the data center are first obtained, and then the load state of the physical host is divided into the following four states according to the set threshold: underload state, appropriate load state, potential overload state and overload state.

4. The method for dynamic virtual machine integration based on balancing energy consumption and service quality as claimed in claim 3, characterized in that: In step 2), the specific process of the hybrid load detection algorithm HLDA is as follows: when the load of a physical host is higher than the upper threshold, the physical host is added to the overloaded host queue When the load of a physical host falls below the lower threshold, the physical host is added to the underloaded host queue. When the load of a physical host is within the threshold, it is necessary to make a judgment based on the predicted load of the physical host; if the expected load of the physical host is higher than the upper threshold, the physical host is added to the potential overload host queue. Otherwise, the physical host is added to the appropriate host queue. middle.

5. The method for dynamic virtual machine integration based on balancing energy consumption and service quality as claimed in claim 4, characterized in that: In step 3), there are three types of physical hosts that need to migrate virtual machines. They are located in the underloaded host queue. Overloaded host queue and potentially overloaded host queues Perform the following operations on these three types of queues: Underloaded host queue Add all virtual machines on this type of physical host to the virtual machine migration queue After all virtual machines are migrated to other physical hosts, they are switched to sleep mode to reduce energy consumption. Overloaded host queue First, all virtual machines on this type of physical host are sorted in descending order according to CPU utilization; second, they are sorted in ascending order according to the amount of memory occupied by the virtual machines in the relative order of descending CPU utilization; finally, each virtual machine is tried to be migrated out in turn, and the load L of the physical host is calculated in real time. c , if L c <η u , then stop the above operations and add the virtual machine you are trying to migrate to the virtual machine migration queue Potentially overloaded host queues First, all virtual machines on this type of physical host are sorted in descending order according to CPU utilization; second, they are sorted in ascending order according to the amount of memory used by the virtual machines in the relative order of descending CPU utilization; finally, each virtual machine is tried to be migrated out in turn, and the load L of the physical host is calculated in real time. E ,like Then stop the above operation and add the virtual machine you are trying to migrate to 6. The method for dynamic virtual machine integration based on balancing energy consumption and service quality as claimed in claim 5, characterized in that: In step 4), the specific steps of the virtual machine placement algorithm RDS-VMPA based on resource demand expansion are as follows: 1) Calculate the resource demand expansion on each physical host. The value is the predicted load of the physical host minus the current workload. This value reflects the changing trend of the virtual machine resource demand on the physical host. 2) Calculate the amount of unallocated resources by subtracting the current workload from the total amount of physical host resources, and select the difference between the amount of unallocated resources and the amount of resource demand expansion, that is, the list of physical hosts with estimated remaining resources greater than zero, to form a list of candidate physical hosts; 3) If the resource demand expansion amount is a negative value, the physical host is added to the resource demand reduction queue, and the difference between the remaining resources and the resource demand expansion amount is calculated and arranged in descending order; if the resource demand expansion amount is a positive value, the physical host is added to the resource demand expansion queue, and the ratio of the resource demand expansion amount to the remaining resources is calculated as the safety factor SF, and the queue is arranged in ascending order according to the value of the factor; 4) After the two queues are generated, if one queue is empty, the head host of the other queue is directly selected as the placement host; If neither is empty, compare the priority factors δ of the hosts at the head of the two queues.

7. A dynamic virtual machine integration system based on balancing energy consumption and service quality, characterized in that: Includes a global manager and two or more local managers; The global manager is equipped with a scheduling module, a data information module, a load prediction center, a load detection center, a virtual machine selection center, and a virtual machine placement center, and the local manager is equipped with a monitoring unit; The local manager collects the real-time load of the CPU and memory on the physical host and virtual machine through the monitoring unit, and sends the real-time load to the data information module in the global manager; The scheduling module in the global manager uses the load prediction center to predict the load status of each physical host at the next moment based on the data set in the data information module; The load detection center detects and divides the load status of the physical host by combining the real-time load and expected load of the physical host; Select the virtual machines to be migrated from the non-suitable hosts through the virtual machine selection center; The virtual machine placement center selects a suitable host for the virtual machine to be migrated, and sends the migration plan to each local manager for migration.

Images