Virtual machine placement strategy based on efficient energy awareness and service quality improvement in cloud computing platform
By introducing energy efficiency and resource utilization indicators into the cloud computing platform, optimizing the virtual machine placement strategy, the energy consumption and service quality problems of cloud data centers are solved, stable, balanced and efficient utilization of resources are achieved, and energy consumption and migration are reduced. Improved service quality.
Patent Information
- Application Number
- CN202411115650.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-12
Smart Images

Figure CN118939381A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of cloud computing technology, and in particular to a virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform. Background Art
[0002] With the rapid development of the Internet, cloud computing has gradually been widely used to meet users' needs for computing, networking, storage, platform and application services. Cloud computing is a model for providing computing services through the Internet. It allows users to access computing resources through the network without the need to own the entities of these resources locally. Users can dynamically adjust their resource usage as needed, thereby achieving higher efficiency and cost savings. With the popularity of cloud computing, large data centers containing thousands of computing nodes have been established all over the world. However, with the surge in the number of cloud users and the expansion of the scale of data centers, the energy consumption and operating costs of data centers are also rising. As of the end of June 2023, the total scale of data center racks in use in my country exceeded 7.6 million standard racks, with an average annual growth rate of nearly 30% in the past five years. The total power consumption of data centers in China in 2022 accounted for about 3% of the total social electricity consumption in the country during the same period. The large power demand of these data centers not only exacerbates energy consumption, but also brings a large amount of carbon emissions and other greenhouse gases, further exacerbating the problem of global warming. According to Amazon's estimates, 42% of the operating costs of data centers are due to their energy consumption. The high energy consumption of data centers is mainly due to their large-scale use of computing resources, while the energy utilization efficiency of these resources is very low. This is a potential threat to cloud providers because it increases total costs. However, efficient resource utilization increases the possibility of physical machine overload, leading to serious violations of the service level agreement (SLA). SLA is defined as the service level expected by the facility provider. Therefore, in order to respond to the country's dual carbon plan and achieve the goal of carbon neutrality as soon as possible, it is urgent to build data centers with high service quality and green energy-saving characteristics.
[0003] In cloud data centers, one of the main methods for reducing energy consumption is to optimize the resource utilization of physical machines (PMs) and switch idle physical machines to sleep mode. The dynamic virtual machine consolidation (DVMC) algorithm is an effective way to implement this method. Virtual machine placement (VMP) is a key component of dynamic virtual machine consolidation. It is responsible for real-time monitoring of the system load and dynamically adjusting the location of virtual machines and resource allocation as needed, which has an important impact on improving the productivity of cloud data centers. Most studies regard VMP as a multi-dimensional bundling problem. The common solution is to place as many virtual machines as possible on physical machines to improve the resource utilization of the entire data center. However, the high resource utilization of physical machines will cause virtual machines to compete for the same physical resources (CPU, RAM, BW), affecting the working performance of virtual machines. The efficient utilization of a single physical resource often tends to ignore the overall coordination and balance of resources, resulting in resource waste. At the same time, high resource utilization will also increase the risk of physical machine overload, affecting the quality of service (QoS) of cloud data centers. The resource requests of virtual machines show obvious fluctuation characteristics over time, and the resource allocation of physical machines also changes accordingly. When the resource requests of virtual machines fluctuate greatly, it may be difficult for the physical machine to quickly and accurately adapt to such changes, and it is easy to be overloaded or underloaded, affecting the performance and stability of the entire system. Therefore, it is crucial to ensure the working performance of virtual machines and the stability and balance of physical machine resource utilization while improving the utilization rate of physical machine resources. Summary of the invention
[0004] The present invention mainly considers the adverse effects on the working performance of virtual machines and the service quality of cloud data centers when the resource utilization of physical machines is high or fluctuates greatly, as well as the resource waste caused by the efficient utilization of a single physical resource, and proposes internal resource utilization balance deviation, safety factor, and fluctuation coefficient indicators. At the same time, considering the energy consumption problem of the cloud data center, the energy efficiency index of the physical machine is introduced. When looking for a target physical machine for migrating VMs, a virtual machine placement strategy (abbreviated as ESVMP strategy) based on efficient energy awareness and service quality improvement in a cloud computing platform is proposed. The strategy comprehensively considers the balance, security, volatility and energy efficiency of the utilization of physical machine resources (CPU, RAM, BW), aiming to deploy virtual machines on high-efficiency physical machines with safe, stable and balanced resource utilization.
[0005] In order to achieve the above object, the present invention adopts the following technical solutions:
[0006] The present invention proposes a virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform, comprising:
[0007] Step 1: Enter the list of physical machines and the list of virtual machines to be migrated;
[0008] Step 2: Loop through the list of virtual machines to be migrated, and repeat steps 3 to 7 for each virtual machine to be migrated;
[0009] Step 3: Loop through the physical machine list and repeat steps 4 to 7 for each physical machine;
[0010] Step 4: Determine whether the physical machine can meet the resource requirements of the virtual machine, and whether the physical machine will not be overloaded at present and after the virtual machine is deployed. If the conditions are met, proceed to step 5.
[0011] Step 5: Calculate the energy efficiency index of the physical machine;
[0012] Step 6: Calculate the internal resource utilization balance deviation index, safety factor index, and CPU resource utilization fluctuation coefficient index of the physical machine after the virtual machine is deployed, and calculate the comprehensive score of the physical machine after the virtual machine is deployed after standardization;
[0013] Step 7: Based on the obtained energy efficiency index of the physical machine and the comprehensive score, a physical machine is selected as a target physical machine for the virtual machine to be migrated.
[0014] Furthermore, in step 5, the energy efficiency index of the physical machine is calculated according to the following formula:
[0015]
[0016] Among them, e i Represents the physical machine p i Energy efficiency, P i max Indicates p i The maximum energy consumption, Indicates p i CPU resource capacity.
[0017] Furthermore, in step 6, the internal resource utilization balance deviation index of the physical machine after the virtual machine is deployed is calculated in the following manner:
[0018]
[0019] Among them, d i Represents the physical machine p i The resource utilization equilibrium deviation, For p i The resource utilization vector, Indicates p i Resource utilization, Respectively represent p i CPU, memory, and bandwidth resource utilization, θ i express and The angle between is the resource balance vector, U max =(1, 1, 1) indicates the maximum resource utilization of the physical machine.
[0020] Furthermore, u is obtained as follows i CPU 、u i RAM 、u i BW :
[0021]
[0022] In the formula
[0023]
[0024] Among them, x ij Represents the physical machine p i With virtual machine v j The mapping relationship between them, M is the number of virtual machines in the cloud data center, Respectively represent v j The requested CPU, memory, and bandwidth resource capacity, Respectively represent p i CPU, memory, and bandwidth resource capacity.
[0025] Furthermore, in step 6, the safety factor index of the physical machine after the virtual machine is deployed is calculated in the following manner:
[0026]
[0027] Among them, s i Represents the physical machine p i The safety factor, Indicates p i Resource utilization, Respectively represent p i CPU, memory, and bandwidth resource utilization, U max =(1, 1, 1) indicates the maximum resource utilization of the physical machine.
[0028] Furthermore, in step 6, the CPU resource utilization fluctuation coefficient index of the physical machine after the virtual machine deployment is calculated in the following manner:
[0029]
[0030] Among them, v ij Represents virtual machine v j After deployment, the physical machine p iThe CPU resource utilization fluctuation coefficient, N0 represents the set of all non-negative integers, f ij '(l) indicates f ij The slope at point l, m represents H ij The length, f ij To H ij The CPU resource utilization curve is fitted by the open source library Apache Commons Math in Java. ij For virtual machine v j Historical CPU utilization record vh j and candidate physical machine p i Historical CPU utilization records i Combined historical CPU utilization records
[0031]
[0032] Where a is vh j The length of , b is ph i length, n = min(a,b), Represents virtual machine v j Total CPU capacity, Represents the physical machine p i The total CPU capacity.
[0033] Furthermore, in step 6, each index obtained in step 6 is standardized according to maximum and minimum standardization.
[0034] Furthermore, in step 6, the comprehensive score of the physical machine after the virtual machine is deployed is calculated according to the following formula:
[0035]
[0036] Among them, score is the comprehensive score of the physical machine after the virtual machine is deployed, d', s', and v' respectively represent the internal resource utilization balance deviation index, safety factor index, and CPU resource utilization fluctuation coefficient index after standardization, and a is the virtual machine v j The length of the historical CPU utilization record, b is the candidate physical machine p i The length of the historical CPU utilization record.
[0037] Furthermore, in step 7, a physical machine is selected as a target physical machine for the virtual machine to be migrated in the following manner:
[0038] Maximize:score
[0039] Maximize:(e)
[0040] The constraints are:
[0041]
[0042] Among them, score is the comprehensive score of the physical machine after the virtual machine is deployed, e is the energy efficiency of the physical machine, N is the number of physical machines in the cloud data center, and r represents CPU, memory or bandwidth resources.
[0043] Compared with the prior art, the present invention has the following beneficial effects:
[0044] In order to deploy virtual machines on energy-efficient physical machines with more balanced, stable and secure resource utilization, the present invention introduces the internal resource utilization balance deviation index, safety factor index, CPU resource utilization fluctuation coefficient index and energy efficiency index of the physical machine after the virtual machine is deployed. While reducing energy consumption, the present invention also reduces the number of virtual machine migrations and the proportion of service level agreement violations, thereby ensuring the service quality of the cloud data center. A large number of experimental results using 10 PlanetLab workloads on the CloudSim simulator show that compared with the LBVMP method, the ESVMP method proposed in the present invention can reduce the energy consumption, number of migrations, SLAV and ESV of the cloud data center by an average of 1.7%, 64.3%, 98.4% and 98.4%, respectively. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 A system architecture diagram of a virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform according to an embodiment of the present invention;
[0046] Figure 2 A schematic diagram of the utilization of three-dimensional resources (CPU, RAM, BW) of a physical machine provided by an embodiment of the present invention;
[0047] Figure 3 A schematic diagram of a fluctuation curve calculation process provided by an embodiment of the present invention;
[0048] Figure 4 The total energy consumption comparison of PABFD, LBVMP, and ESVMP strategies provided in the embodiment of the present invention in combination with MMT and MC respectively;
[0049] Figure 5 The PABFD, LBVMP, and ESVMP strategies provided in the embodiment of the present invention are respectively combined with the number of virtual machine migrations of MMT and MC;
[0050] Figure 6 The PABFD, LBVMP, and ESVMP strategies provided in the embodiments of the present invention are combined with the SLAV comparison of MMT and MC respectively;
[0051] Figure 7The PABFD, LBVMP and ESVMP strategies provided in the embodiments of the present invention are respectively combined with the ESV comparison of MMT and MC. DETAILED DESCRIPTION
[0052] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments:
[0053] A virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform of the present invention comprises:
[0054] S101: Input a list of physical machines and a list of virtual machines to be migrated;
[0055] S102: loop through the list of virtual machines to be migrated, and repeat S103-S107 for each virtual machine to be migrated;
[0056] S103: loop through the physical machine list, and repeat S104-S107 for each physical machine;
[0057] S104: Determine whether the physical machine can meet the resource requirements of the virtual machine, and whether the physical machine will not be overloaded at present and after the virtual machine is deployed. If the conditions are met, execute S105;
[0058] S105: Calculate the energy efficiency index of the physical machine;
[0059] S106: Calculate the internal resource utilization balance deviation index, safety factor index, and CPU resource utilization fluctuation coefficient index of the physical machine after the virtual machine is deployed, and calculate the comprehensive score of the physical machine after the virtual machine is deployed after standardization;
[0060] S107: Based on the obtained energy efficiency index of the physical machine and the comprehensive score, a physical machine is selected as the target physical machine for the virtual machine to be migrated. Since there are restrictions on the calculation of the CPU resource utilization fluctuation coefficient index, if there is a physical machine that meets the conditions, a high-energy-efficiency physical machine that meets the conditions with the highest comprehensive score is selected as the target physical machine for the virtual machine to be migrated; if there is no physical machine that meets the conditions, a high-energy-efficiency physical machine that meets the conditions with the highest comprehensive score is selected as the target physical machine for the virtual machine to be migrated.
[0061] Specifically, Figure 1The system architecture of the ESVMP strategy is shown. We assume that a cloud data center (CDC) consists of N physical machines and M virtual machines. Each physical machine is embedded with a VM management system, which is responsible for real-time monitoring of the status of the physical machine. The management system consists of two core modules: the virtual machine monitoring (VMM) module and the virtual machine placement (VMP) module. These physical machines are divided into three working states: overload, normal, and underload according to resource utilization. When the VMM module detects that a physical machine is in an overload or underload state, the VMP module based on the ESVMP strategy will migrate the virtual machine to be migrated selected by the physical machine to the energy-efficient physical machine with the highest comprehensive score. The ESVMP strategy establishes a three-dimensional resource (CPU, RAM, BW) space, introduces resource utilization vectors and resource balance vectors, and defines the balance deviation and safety factor indicators of the resource utilization of the physical machine. The CPU resource utilization curve of the virtual machine and the physical machine is fitted by the open source library Apache Commons Math method in Java, and the fluctuation coefficient indicator of the CPU resource utilization is defined. Finally, the virtual machine placement problem is formulated as a mixed integer linear programming (MILP) problem, which comprehensively considers the energy efficiency of the physical machine and the stability, balance, and security of the physical machine resource utilization after the virtual machine is deployed, ensuring the performance of the virtual machine and the service quality of the cloud data center. At the same time, considering the dimensional impact between different indicators, we normalize these indicators through the Min-Max Normalization method in the process of scoring candidate physical machines to ensure that they are comparable in the analysis.
[0062] Table 1 lists some symbols used in the present invention to facilitate understanding of the present invention.
[0063] Table 1 List of some symbols used in the present invention
[0064]
[0065]
[0066] Physical machine: Assume PM = {p1, p2, ..., p N} is a collection of N physical machines in the data center, and physical machine p i The resource capacity is expressed as in Represents the physical machine p i CPU, RAM (memory), and BW (bandwidth) resource capacity.
[0067] Virtual machine: Assume VM = {v1, v2, ..., v M} is a collection of M virtual machines in a data center, virtual machine v j The request resource is represented as in Represents virtual machine v j Requested CPU, RAM, and BW resource capacity.
[0068] VMP problem: We define an N-dimensional M-column matrix X to represent the mapping relationship between virtual machines and physical machines, where the physical machine p i With virtual machine v j The mapping relationship between them is as follows:
[0069]
[0070] Unbalanced allocation of physical machine resources means that in a single or multiple resource dimensions, the resource utilization of some physical machines is high, but the available resources are insufficient to allocate to the migrated virtual machines to ensure their normal operation, which ultimately leads to a waste of physical machine resources. Therefore, by achieving balanced allocation of resources, it is possible to avoid performance degradation caused by overloading of some physical machines and prevent idleness of other physical machine resources, thereby ensuring the service quality of the cloud data center.
[0071] Physical machine i The resource utilization is expressed as in Represents the physical machine p i The CPU, RAM, and BW resource utilization are defined as follows:
[0072]
[0073] Among them, x ij Represents the physical machine p i With virtual machine v j The mapping relationship between them. Figure 2 As shown, in the three-dimensional resource space, U max =(1, 1, 1) represents the maximum resource utilization of the physical machine, and defines the vector is the resource balance vector. The resource utilization of the physical machine is represented by U, and the vector is defined as is the resource utilization vector of the physical machine. By measuring the point U and the vector The distance between the physical machines can be used to determine the imbalance in the utilization of physical machine resources. The larger the distance, the more imbalanced the utilization of physical machine resources. i The internal resource utilization balance deviation is expressed as d i It is defined as follows:
[0074]
[0075] Among them, θ i Representation vector With vector The angle between.
[0076] The utilization of physical resources (including CPU, RAM, and BW) largely reflects the impact on the performance of virtual machines. Deploying virtual machines on physical machines with high resource utilization will reduce the performance of virtual machines and cause adverse effects such as resource contention and invalid migration. Therefore, monitoring and managing physical machines to prevent overload is crucial to maintaining system performance and stability. We use the safety factor of physical machines to measure the overload risk of physical machines. The higher the safety factor, the lower the overload risk. i The safety factor is denoted by s i It means that the definition is as follows:
[0077]
[0078] The resource requests of virtual machines show obvious fluctuation characteristics over time, and the resource allocation of physical machines also changes accordingly. When the resource requests of virtual machines fluctuate greatly, it may be difficult for the physical machine to quickly and accurately adapt to such changes, and it is easy to be overloaded or underloaded, affecting the performance and stability of the entire system. Therefore, it is particularly important to consider the stability of physical machine resource utilization when deploying virtual machines. By analyzing and predicting historical CPU utilization data, the stability of CPU resource utilization and the status of physical machines can be obtained, inappropriate virtual machine placement can be avoided, and service quality can be guaranteed to a certain extent.
[0079] To migrate virtual machines v j In the process of finding the target physical machine for placement, we will place the virtual machine v j Historical CPU utilization records are recorded with vh j Indicates that the candidate physical machine p i Historical CPU utilization records are recorded using ph i By combining vh j and ph i The development trend of can predict the stability of physical machine resource utilization in the future after the virtual machine is deployed, which is crucial to ensure the performance of the virtual machine. We represent the historical CPU utilization record of the combination of the two as H ij , which is used as the basis for judging the stability of physical machine CPU resource utilization after virtual machine deployment, is defined as follows:
[0080]
[0081] Where a is vh j The length of , b is ph i length, n = min(a,b), Represents virtual machine v j Total CPU capacity, Represents the physical machine pi The total CPU capacity.
[0082] Record H ij Fitting to curve f using the open source library Apache Commons Math in Java ij , obtain the slope of each record corresponding point and calculate the average absolute value as the fluctuation coefficient of physical machine CPU resource utilization. ij The specific process is shown in Figure 3 .
[0083] Virtual Machine j After deployment, the physical machine p i The fluctuation coefficient of CPU resource utilization is expressed as v ij It is defined as follows:
[0084]
[0085] Where N0 represents the set of all non-negative integers, that is, including zero and all positive integers; f ij '(l) represents the curve f ij The slope at point l, m represents H ij Length.
[0086] After the virtual machine is deployed, the high CPU resource utilization of the physical machine is likely to cause physical machine overload and invalid migration. If the CPU resource utilization of the physical machine in the next stage is predicted to be too high after the virtual machine is deployed, we believe that the deployment of the virtual machine is likely to cause physical machine overload, and the physical machine will not be selected as the target physical machine.
[0087] In cloud data centers, energy efficiency is a common feature of all types of physical machines, representing the effective utilization of hardware resources and the optimization level of energy consumption. The higher the energy efficiency, the less energy a physical machine consumes while providing the same computing power. i The energy efficiency is defined as follows:
[0088]
[0089] Among them, P i max Represents the physical machine p i The maximum energy consumption of is the power consumption when the CPU resource utilization is 1.
[0090] Indicator standardization: Different indicators often have different dimensions and dimensional units. We need to eliminate the dimensional effects between them to ensure that they are comparable in analysis. To solve this problem, we use the Min-Max Normalization method, which can scale the data to a specific range, usually 0, 1 or -1, 1. The Min-Max Normalization is converted using the following formula:
[0091]
[0092] Among them, X represents the original data, min(X) represents the minimum value of the data, max(X) is the maximum value of the data, and X' is the result after standardization.
[0093] To apply this method, it is necessary to determine the maximum and minimum values of the indicator data. We standardize the above indicators except the energy efficiency indicator. The processed resource balance, safety factor, and volatility indicators are represented as d', s', and v', respectively, which are used to guide the virtual machine allocation plan.
[0094] Solve for the composite score as follows:
[0095]
[0096] Our goal is to deploy virtual machines on high-efficiency physical machines with stable, balanced and secure resource utilization to ensure the service quality of cloud data centers. Based on this goal, we designed three indicators: resource utilization balance deviation, safety factor, and volatility, which work together with energy efficiency indicators on virtual machine allocation plans. The problem definition is as follows:
[0097] Maximize:score (12)
[0098] Maximize:(e) (13)
[0099] The constraints are:
[0100]
[0101] Constraint (14) restricts a virtual machine to be placed on only one physical machine, and constraint (15) restricts the requested resources of all virtual machines deployed on a physical machine to not be greater than the resource capacity of the physical machine itself.
[0102] Finally, the migration target physical machine is obtained through the above method.
[0103] In order to better understand the present invention, the present invention is further described:
[0104] The virtual machine placement algorithm is an important part of the virtual machine dynamic integration algorithm, also known as the multi-dimensional packing process, which is an NP-hard problem for the mapping relationship between VMs and PMs. The efficiency of cloud computing depends largely on the efficiency of the virtual machine placement strategy used. When the virtual machine monitoring module (VMM) detects that a physical machine is in an overloaded or underloaded state, it triggers the migration process of the virtual machine. In this process, selecting the migration destination physical machine is a key decision, and multiple factors need to be considered to ensure the effect of the migration and the overall performance of the system.
[0105] Algorithm 1 gives the pseudo code of the ESVMP strategy. The input of the algorithm is the list of physical machines in the cloud data center and the selected virtual machine to be migrated, and the output is the target physical machine. The selection of the target physical machine is based on the indicators established by the present invention. In lines 1-6, the maximum score, maximum energy efficiency and target physical machine of the candidate physical machine are initialized. The two identical indicators are defined separately because the calculation of the fluctuation coefficient indicator needs to meet the condition of the minimum number of historical CPU utilization records of the physical machine or virtual machine. If there is a physical machine that meets the conditions, the virtual machine to be migrated is deployed on the physical machine with the maximum score and the highest energy efficiency that meets the conditions, otherwise it is deployed on the physical machine with the maximum score and the highest energy efficiency that does not meet the conditions. Then, in lines 7-36, the list of physical machines in the cloud data center is traversed to find the target physical machine. First, it is determined whether the candidate physical machine is in an active state and can meet the resource requirements of the virtual machine without being overloaded, and the physical machine that meets the conditions is subjected to the next step of judgment and processing. Secondly, the energy efficiency index of the candidate physical machine is calculated, and the index array is calculated by Algorithm 2. Since there are restrictions on the calculation of the fluctuation coefficient index, we discuss it in two cases.
[0106] The first case: If the length of the indicator array is less than 3, it means that the calculation of the volatility index does not meet the conditions. The ESVMP policy prioritizes the deployment of virtual machines on physical machines with stable resource utilization, and gives higher priority to physical machines that can calculate the volatility index. Therefore, in lines 16-18, if there is a target physical machine that considers the volatility index, then the physical machine that cannot calculate the volatility index is no longer used as a reference. If there is no target physical machine that considers the volatility index, in lines 20-23, we consider the comprehensive score and energy efficiency of the physical machine, and select the physical machine with the highest score and the highest energy efficiency as the target candidate physical machine.
[0107] The second case: the fluctuation coefficient indicator exists in the indicator array. From formula (8), we can see that the fluctuation coefficient indicator is not less than 0. If the fluctuation coefficient indicator is negative, it means that the deployment of the virtual machine is likely to cause the physical machine to be overloaded. See Algorithm 2 for details. At this time, the physical machine is no longer used as a reference. If the fluctuation coefficient indicator is normal, in lines 30-33, we consider the comprehensive score and energy efficiency of the physical machine and select the physical machine with the highest score and the highest energy efficiency as the target physical machine.
[0108] Finally, in lines 37-42, if the target physical machine considering the volatility index exists, the physical machine is returned. If not, the target candidate physical machine is returned.
[0109]
[0110] Algorithm 2 gives the process of calculating the physical machine index array. The input of the algorithm is the candidate physical machine and the selected virtual machine to be migrated, and the output is the index array. First, in lines 1-4, the index array of the candidate physical machine is initialized and defined, and the resource (CPU, RAM, BW) utilization of the physical machine after the virtual machine is deployed is calculated respectively. Secondly, in lines 5-11, the resource balance vector and resource utilization vector are defined, and the resource utilization balance deviation index and safety factor index are calculated according to formula (5) (Eq. (5)) and formula (6) (Eq. (6)) respectively, and these two indicators are standardized by formula (10) and stored in the index array. Thirdly, in lines 12-16, the historical CPU utilization records of the VM to be migrated and the candidate physical machine are obtained respectively. If the number of historical CPU utilization records of the migrated VM or the candidate physical machine is too small, it is easy to cause the curve fitting accuracy to be too low and affect the decision. At this time, we do not consider the fluctuation coefficient index of CPU resource utilization and directly return the index array. Here we set the minimum number of records to 5. Otherwise, in lines 18-28, the historical CPU utilization records of the two are combined and the curve is fitted. The fluctuation coefficient index is calculated according to formula (8) and normalized by formula (10) and added to the index array. In lines 21-23, if the predicted value of the curve in the next stage is too high, it means that the deployment of the virtual machine is likely to cause physical machine overload and invalid migration. At this time, the physical machine is not used as a candidate physical machine. The method of setting the fluctuation coefficient index to a negative value guides the ESVMP strategy to abandon the physical machine. Here we use 0.7 as the safety threshold for virtual machine deployment. Finally, return the index array.
[0111]
[0112]
[0113] Time complexity analysis: Assume that there are M virtual machines selected for migration and N physical machines in the cloud data center. For each virtual machine, we need to traverse all physical machines to make decisions. Therefore, the time complexity of our method is proportional to the number of physical machines N, that is, O(MN).
[0114] In order to verify the effectiveness of the strategy proposed in this invention, a simulation verification is carried out on the CloudSim simulation platform. 800 heterogeneous servers of two types, HP ProLiant ML110 G4 (Intel Xeon 3040) and HP ProLiant ML110 G5 (Intel Xeon 3075), are simulated respectively. The characteristics of the two types of servers are shown in Table 2.
[0115] Table 2 Server characteristics
[0116]
[0117] When a physical machine is active, there is a linear relationship between its energy consumption and its CPU resource utilization. The measured energy consumption is based on real data from the SPECpower benchmark test results. The power consumption of the two types of servers under different loads is shown in Table 3.
[0118] Table 3 Power consumption of the server under different loads (W)
[0119]
[0120] We use four types of Amazon EC2 VMs, the details of which are shown in Table 4.
[0121] Table 4 VM characteristics
[0122]
[0123] In order to verify the working performance of the present invention in a real cloud data center, 10 workloads provided by PlanetLab are used, which is a computer cluster project distributed all over the world. It collects CPU resource utilization data of virtual machines from servers in more than 500 locations around the world. The statistical cycle is 1 day, and the CPU resource utilization of the server is counted every 5 minutes. The specific information of the workload is shown in Table 5.
[0124] Table 5 Workload characteristics
[0125]
[0126]
[0127] (1) Performance evaluation:
[0128] In order to comprehensively evaluate the performance of the strategy proposed in this invention, we use a series of mainstream and comprehensive evaluation indicators. These indicators not only reflect the energy efficiency and operation efficiency of the system, but also comprehensively consider the service quality and system reliability, providing us with a multi-dimensional performance evaluation perspective. These indicators are described as follows:
[0129] 1. Energy consumption: Energy consumption directly reflects the energy efficiency and environmental friendliness of the data center, and is an important indicator for evaluating the performance of the virtual machine placement algorithm. Idle hosts will increase the total power consumption of the data center. The energy consumption of the physical machine is represented by P, which is defined as follows:
[0130]
[0131] Among them, P i idle Represents the physical machine p i Idle energy consumption, P max Represents the physical machine p i Energy consumption at full load, Indicates the CPU resource utilization of the physical machine.
[0132] From formula (16), we can see that the energy consumption of the physical machine fluctuates with the change of CPU resource utilization, and the CPU resource utilization of the physical machine fluctuates with the change of user demand. The energy consumption of the physical machine over a period of time is defined as follows:
[0133]
[0134] The total energy consumption of a data center is the sum of the energy consumption of all physical machines, represented by E, and defined as follows:
[0135]
[0136] Where N represents the total number of physical machines in the data center.
[0137] 2. Number of VM migrations: Frequent VM migrations may not only cause service response delays and reduced system performance, but also directly affect the service experience of end users.
[0138] 3. SLAV: SLAV is a comprehensive indicator that takes into account both PDM and SLATAH to reflect the degree of service unavailability of a service provider within a certain time frame. It is defined as follows:
[0139] SLAV=PDM*SLATAH(19)
[0140] 4. ESV: ESV is a comprehensive indicator of E and SLAV, used to evaluate the comprehensive value of cloud services, and is defined as follows:
[0141] ESV=E*SLAV(20)
[0142] (2) Comparison method
[0143] In order to verify the effectiveness of the strategy of the present invention, we used five different physical machine overload detection methods (IQR, LR, LRR, MAD, THR) and two migration virtual machine selection algorithms (MMT, MC) in CloudSim to compare the performance of the PABFD algorithm [Beloglazov A, Buyya R. Optimal online deterministic algorithms and adaptive heuristics for energy and performance efficient dynamic consolidation of virtual machines in Cloud data centers [J]. Concurrency and Computation: Practice and Experience, 2012, 24. DOI: 10.1002 / cpe.1867.], the LBVMP algorithm [Wang J, Yu J, Song Y, et al. An efficient energy-aware and service quality improvements strategy applied in cloud computing [J]. Cluster Computing, 2023, 26 (6): 4031-4049.] and the ESVMP strategy proposed in this invention. The safety parameters of the IQR, LR, LRR, and MAD algorithms are set to 1.2, and the safety parameter of the THR algorithm is set to 0.8 to be consistent with the LBVMP algorithm.
[0144] (3) Experimental results analysis
[0145] Table 6 shows the performance comparison of five overloaded physical machine detection methods (IQR, LR, LRR, MAD, THR) combined with the minimum migration time migration virtual machine selection algorithm (MMT) based on the ESVMP strategy proposed in the present invention and the advanced LBVMP strategy under the same experimental conditions. ESVMP is superior to the LBVMP strategy in terms of the number of VM migrations, SLAV, and ESV, which are improved by 62.5%, 98.4%, and 98.5%, respectively. In terms of energy consumption, the iqr_mmt, mad_mmt, and thr_mmt algorithms based on the ESVMP strategy are not as good as the LBVMP strategy. However, ignoring the difference in algorithms, on average, the energy consumption of the ESVMP strategy is 1.7% lower than that of the LBVMP strategy. This is because the ESVMP strategy aims to deploy virtual machines on energy-efficient physical machines with stable, balanced and secure resource utilization, while optimizing the resource utilization of physical machines and reducing the risk of overload of physical machines, while saving energy and reducing the number of migrations and SLA violations.
[0146] Table 6 Comparison of ESVMP and LBVMP strategies
[0147]
[0148]
[0149] Figure 4 The total energy consumption comparison of different virtual machine placement strategies under workloads W1 to W10 is shown, combining MMT and MC virtual machine selection algorithms. When the migration virtual machine selection algorithm is MMT, the total energy consumption generated by the ESVMP strategy is 34.1% lower than the PABFD strategy on average, and 1.7% lower than the LBVMP strategy. When the migration virtual machine selection algorithm is MC, the total energy consumption generated by the ESVMP strategy is 34.6% lower than the PABFD strategy on average, and 1.6% lower than the LBVMP strategy. The goal of the ESVMP strategy is to prioritize the deployment of virtual machines on physical machines with high energy efficiency, thereby improving the energy efficiency of the entire data center. The experimental results fully verify the effectiveness of the ESVMP strategy in reducing energy consumption.
[0150] Figure 5The comparison of the number of VM migrations generated by the MMT and MC virtual machine selection algorithms under workloads W1 to W10 is shown. When the migration virtual machine selection algorithm is MMT, the number of VM migrations generated by the ESVMP strategy is 73.7% higher than the PABFD strategy on average and 64.3% higher than the LBVMP strategy. When the migration virtual machine selection algorithm is MC, the number of VM migrations generated by the ESVMP strategy is 77.8% higher than the PABFD strategy on average and 60.8% higher than the LBVMP strategy. The ESVMP strategy is committed to deploying virtual machines on physical machines with stable resource usage and low overload risks, with the aim of minimizing the possibility of overload and invalid migration of physical machines in the data center. The experimental results fully verify the significant effectiveness of the ESVMP strategy in reducing the number of virtual machine migrations.
[0151] Figure 6 The comparison of SLAV indicators generated by combining MMT and MC virtual machine selection algorithms under workloads W1 to W10 is shown. When the migration virtual machine selection algorithm is MMT, the SLAV generated by the ESVMP strategy is 99.6% higher than the PABFD strategy and 98.4% higher than the LBVMP strategy on average. When the migration virtual machine selection algorithm is MC, the SLAV generated by the ESVMP strategy is 99.4% higher than the LBVMP strategy and 98.5% higher than the PABFD strategy on average. The optimization of SLAV is directly related to the reduction of PDM and SLATAH. The experimental results fully verify that the ESVMP strategy has a significant effect in reducing service level agreement (SLA) violations.
[0152] Figure 7 The comparison of ESV indicators generated by combining MMT and MC virtual machine selection algorithms under workloads W1 to W10 is shown. When the migration virtual machine selection algorithm is MMT, the ESV generated by the ESVMP strategy is 99.7% higher than the PABFD strategy and 98.3% higher than the LBVMP strategy on average. When the migration virtual machine selection algorithm is MC, the ESV generated by the ESVMP strategy is 99.6% higher than the LBVMP strategy and 98.7% higher than the PABFD strategy on average. The optimization of ESV is directly related to the reduction of SLAV and energy consumption. The experimental results fully demonstrate the remarkable effectiveness of the ESVMP strategy in optimizing comprehensive energy indicators.
[0153] In summary, in order to deploy virtual machines on energy-efficient physical machines with more balanced, stable and secure resource utilization, the present invention introduces the internal resource utilization balance deviation index, safety factor index, CPU resource utilization fluctuation coefficient index and energy efficiency index of the physical machine after the virtual machine is deployed. While reducing energy consumption, the present invention also reduces the number of virtual machine migrations and the proportion of service level agreement violations, thereby ensuring the service quality of the cloud data center. A large number of experimental results using 10 PlanetLab workloads on the CloudSim simulator show that compared with the LBVMP method, the ESVMP method proposed in the present invention can reduce the energy consumption, number of migrations, SLAV and ESV of the cloud data center by an average of 1.7%, 64.3%, 98.4% and 98.4%, respectively.
[0154] The above is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.
Claims
1. A virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform, characterized in that: include: Step 1: Enter the list of physical machines and the list of virtual machines to be migrated; Step 2: Loop through the list of virtual machines to be migrated, and repeat steps 3 to 7 for each virtual machine to be migrated; Step 3: Loop through the physical machine list and repeat steps 4 to 7 for each physical machine; Step 4: Determine whether the physical machine can meet the resource requirements of the virtual machine, and whether the physical machine will not be overloaded at present and after the virtual machine is deployed. If the conditions are met, proceed to step 5. Step 5: Calculate the energy efficiency index of the physical machine; Step 6: Calculate the internal resource utilization balance deviation index, safety factor index, and CPU resource utilization fluctuation coefficient index of the physical machine after the virtual machine is deployed, and calculate the comprehensive score of the physical machine after the virtual machine is deployed after standardization; Step 7: Based on the obtained energy efficiency index of the physical machine and the comprehensive score, a physical machine is selected as a target physical machine for the virtual machine to be migrated.
2. The virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform according to claim 1, characterized in that: In step 5, the energy efficiency index of the physical machine is calculated according to the following formula: Among them, e i Represents the physical machine p i Energy efficiency, P i max Indicates p i The maximum energy consumption, Indicates p i CPU resource capacity.
3. The virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform according to claim 1, characterized in that: In step 6, the internal resource utilization balance deviation index of the physical machine after the virtual machine is deployed is calculated in the following manner: Among them, d i Represents the physical machine p i The resource utilization equilibrium deviation, For p i The resource utilization vector, Indicates p i Resource utilization, Respectively represent p i CPU, memory, and bandwidth resource utilization, θ i express and The angle between is the resource balance vector, U max =(1, 1, 1) indicates the maximum resource utilization of the physical machine.
4. The virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform according to claim 3, characterized in that: Derived in the following way In the formula Among them, x ij Represents the physical machine p i With virtual machine v j The mapping relationship between them, M is the number of virtual machines in the cloud data center, Respectively represent v j Requested CPU, memory, and bandwidth resource capacity, Respectively represent p i CPU, memory, and bandwidth resource capacity.
5. The virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform according to claim 1, characterized in that: In step 6, the safety factor index of the physical machine after the virtual machine is deployed is calculated in the following manner: Among them, s i Represents the physical machine p i The safety factor, Indicates p i Resource utilization, Respectively represent p i CPU, memory, and bandwidth resource utilization, U max =(1, 1, 1) indicates the maximum resource utilization of the physical machine.
6. The virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform according to claim 1, characterized in that: In step 6, the CPU resource utilization fluctuation coefficient index of the physical machine after the virtual machine is deployed is calculated in the following manner: Among them, v ij Represents virtual machine v j After deployment, the physical machine p i The CPU resource utilization fluctuation coefficient, N0 represents the set of all non-negative integers, f ij '(l) indicates f ij The slope at point l, m represents H ij The length, f ij To H ij The CPU resource utilization curve is fitted by the open source library Apache Commons Math in Java. ij For virtual machine v j Historical CPU utilization record vh j and candidate physical machine p i Historical CPU utilization records i Combined historical CPU utilization records Where a is vh j The length of , b is ph i length, n = min(a,b), Represents virtual machine v j Total CPU capacity, Represents the physical machine p i The total CPU capacity.
7. The virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform according to claim 1, characterized in that: In the step 6, each index obtained in the step 6 is standardized according to the maximum and minimum standardization.
8. The virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform according to claim 1, characterized in that: In step 6, the comprehensive score of the physical machine after the virtual machine is deployed is calculated according to the following formula: Among them, score is the comprehensive score of the physical machine after the virtual machine is deployed, d', s', and v' respectively represent the internal resource utilization balance deviation index, safety factor index, and CPU resource utilization fluctuation coefficient index after standardization, and a is the virtual machine v j The length of the historical CPU utilization record, b is the candidate physical machine p i The length of the historical CPU utilization record.
9. The virtual machine placement strategy based on efficient energy awareness and service quality improvement in a cloud computing platform according to claim 1, characterized in that: In step 7, a physical machine is selected as a target physical machine for the virtual machine to be migrated in the following manner: Maximize:score Maximize:(e) The constraints are: Among them, score is the comprehensive score of the physical machine after the virtual machine is deployed, e is the energy efficiency of the physical machine, N is the number of physical machines in the cloud data center, and r represents CPU, memory or bandwidth resources.
