TW201516872A - Method of balancing physical machine resource load in cluster environment - Google Patents

Abstract

The invention discloses a method of balancing a physical machine resource load in a cluster environment, including the steps of taking several resource usage rates of a virtual machine provided on a physical machine set in a cluster environment, and further calculating a discrete degree of the resource load of the physical machine set in a cluster environment and resource load arithmetic mean, via the usage rates of a virtual machine; and calculating a plurality of candidate virtual machines by a virtual machine removal algorithm for movement when the discrete degree of the resource load of the physical machine set in a cluster environment is higher than the discrete degree of a target load. The algorithm sets physical machine with resource load larger than the resource load arithmetic mean in a cluster environment in a group and others in another group, and selects virtual machines as movement candidates in a mode of selecting from each two physical machines with the same distance ranking with the resource load arithmetic mean to reduce the discrete degree of the resource load of the physical machine set in a cluster environment into the discrete degree of a target load after a removal.

Description

Method for balancing physical machine resource load in a cluster environment

The present invention relates to a dynamic resource configuration method, and more particularly to a method for balancing a physical machine resource load in a cluster environment.

The cluster environment is mainly used to divide the resources on the physical machine (generally called Host) into different equal parts through virtualization technology. Virtualization allows users or applications to easily obtain the computing resources they need without being limited to the initial installation, location, location, or combination of entities. Virtualization technology allows users to run multiple operating systems on a single physical machine. Virtualization technology operates through virtual machine management (hypervisor) (or generally referred to as virtual machine monitor (VMM)). (Virtual Machine) is separated from the physical machine hardware. In a physical machine, including processors, memory, storage devices, network I/O and other resources, several virtual machines are split on it, each virtual The machine allocates the required hardware resources according to the requirements of its application. The virtual machine management layer is responsible for communication between the virtual machine and the hardware. The virtual machine management layer is responsible for transmitting the instructions of the virtual machine to the required hardware resources for execution. The results of the operation are transmitted back to the virtual machine. Each virtual machine is independent of each other and does not know each other to share hardware resources. In other words, virtualization technology provides a logical classification of data, computing power, storage capacity, and other resources, rather than a physical boundary.

The cluster system is composed of a plurality of physical machines. In order to properly utilize the physical machine resources, each physical machine allocates resources to several virtual machines to improve resource utilization. The virtual machine's demand for resources fluctuates, so the resource usage rate on each physical machine is very different. The virtualization technology allows the virtual machine management layer to provide overcommit function. The virtual machine resource usage cannot be observed by observation. In order to improve the reliability of the server and the service efficiency, and avoid the service interruption caused by the excessive hardware load, the resources between the physical machines must start the load balancing mechanism to move one virtual machine from one physical machine with tight resources to another. A physical machine with remaining resources. The invention Objective To provide a method for dynamic resource allocation in a cluster environment, by obtaining virtual machine resource usage history data of devices on each physical machine as a basis for resource scheduling, so as to achieve a design method of physical machine resource load balancing.

The cluster environment designed by the present invention for the purposes stated in the prior art above The method for balancing the load of the physical machine resource includes the following steps: a virtualized resource management system monitors a plurality of virtual machines through the virtual machine management layer and calculates and expresses resource utilization indicators of the virtual machines according to the resource representation designed by the present invention. And when load balancing is performed, the settings of the virtual machine limit type, the preset discreteness threshold value preference server setting, and the like, which are required to be considered in different situations in the cluster operation, are required.

Further, in accordance with the purpose of the present invention, the proposed cluster environment is flat The method for balancing the load of the physical machine resource includes: A, the threshold value of the incoming cluster resource, the virtual machine configuration mutual exclusion and the group setting condition; B, the resource usage collected by each virtual machine history for a period of time C. Calculate the actual resource usage rate of the physical machine based on the resource usage of all the virtual machines in the physical machine, and further calculate the standard difference in the cluster environment by calculating the resource usage rate of all the physical machines to determine The load balancing state and the degree of dispersion between the physical machines; D, judging whether the standard deviation obtained is greater than the cluster resource threshold, and if so, selecting the physical machine to select at least one pair of physical machines, and Selecting a pair of physical machines to select a virtual machine for simulated mobile work to improve the standard deviation of the physical device resource usage, wherein one of the paired physical machines has a resource usage greater than another entity. E; Calculate the simulated moving virtual machine, determine whether the standard deviation of the physical machine resource usage rate is still higher than the cluster resource threshold, and if so, then Step D, and E, if not, step F; and F., Perform the move operation of the virtual machine, to move the two sets of each pair of the physical machine in the virtual machine selection.

Preferably, the degree of dispersion is calculated by the following mathematical formula: among them, Indicates the resource usage rate of a numbered j entity; X _i,j indicates whether a number i virtual machine is started on the physical machine of this number j, if it has a value of 1, if it is not enabled, its value is 0. ; The resource capabilities used by this virtual machine i; The arithmetic mean of the resource usage of the m physical machines in the cluster; σ is the standard deviation obtained from the above values to express the degree of dispersion of the resource load in the cluster.

Preferably, the step D further includes the following steps: calculating resource usage of each physical machine Average with arithmetic The difference D _j ; and the physical machine with tight resources in the cluster are classified into one group according to the calculated difference value D _j , the other physical machines are classified into another group, and the physical machines of the two groups are classified according to the difference D The size of _j is ranked to select the same ranking or a specific ranked physical machine between the two groups as the object of relocating the virtual machine, and then screening at least one pair of paired physical machines, wherein one of the paired is the source entity Machine, the other is the target physical machine.

Preferably, when the difference D _{j of the} above-mentioned physical machine is greater than zero, it is classified into a group of resource shortages in the cluster, and the physical machine whose difference D _{j is} less than zero belongs to another group.

Preferably, the above step D further comprises the following steps: After the standard physical machine is selected, the difference between the used resource and the mean value of the source physical machine and the target physical machine is calculated, wherein the difference U is the value of the difference D multiplied by the total capacity R of the physical machine target resource; and the source entity A virtual machine is selected in the machine. The virtual machine consumes more resources than the source physical machine and the mean value, and the difference is smaller than the predefined one according to the target resource. Values.

Preferably, the step D further includes the following steps: calculating the resource load median Thd according to the difference between the source resource and the target physical device and the mean value U _s , U _t ; and if the source physical machine cannot When selecting a virtual machine that matches the virtual machine, select a virtual machine from the source physical machine and select a virtual machine from the target physical machine, and move the virtual machine selected by the source physical machine to the target physical machine, and the target physical machine The selected virtual machine is moved to the source physical machine so that the difference between the resource capabilities V ^resource of the two virtual machines after moving is between | Ut | and | Us | and is closest to the Thd value.

Preferably, the resource load amount median Thd=( U _s - U _t )/2.

Preferably, the step D further includes the following steps: if the source entity and If the virtual machine cannot be selected in the target entity, a virtual machine is selected from the source entity. The resource capacity of the virtual machine Vresource needs to be close to the resource load median Thd.

According to the invention, it has one or more of the following advantages:

1. Balancing the resource load of a physical machine in a cluster environment is not only flexible, but also does not promptly propose a migration when the VM needs to work.

2. Under the premise that the physical machine hardware is compatible under the cluster environment, the load balancing configuration can be calculated for the physical machines with different capabilities without being limited to the prior art, and only the physical machine with the same hardware specifications can be balanced.

3. The invention dynamically modifies parameters according to different actual environments and requirements, and can flexibly make suitable balance results.

100‧‧‧Virtual Resource Management System

200‧‧‧Virtual Machine Management

300‧‧ ‧ cluster

400, 500‧‧‧ physical machine

S101~S106, S201~S205‧‧‧ steps

1 is a schematic diagram of an embodiment of an architecture employed in a method of balancing physical machine resource loads in a cluster environment of the present invention.

Figure 2 is a flow diagram of an embodiment of a method of balancing physical machine resource loads in a cluster environment of the present invention.

Fig. 3 is a detailed flow chart of step S104 in Fig. 2.

For the benefit of the examiner, the technical features, contents and advantages of the present invention and their The present invention will be described in detail with reference to the accompanying drawings, and the description of the embodiments of the present invention will be described in detail. True proportion and precise configuration, therefore, the scope and configuration relationship of the attached drawings should not be interpreted or limited, and the scope of rights of the present invention in actual implementation should be described first.

The following embodiment is a load balancing scheduling method for a physical machine in a cluster computing environment An improved resource representation is used to collect virtual machine resource usage and an advanced load balancing scheduling method to find one or more appropriate virtual machines and move to the appropriate physical machine so that the physical machine achieves efficient load balancing.

Please refer to FIG. 1 for a cluster environment for applying the preferred embodiment of the present invention. A system architecture diagram of a method for balancing the load of a physical machine resource. In an embodiment, there is a virtualized resource management system 100, which mainly functions to manage and monitor various physical devices in the equipment room, and utilizes the virtualization technology of the virtual machine management layer 200 to convert the physical machine resources of the cluster 300 into virtual resource pools. Use a consistent, common user interface and API to operate management resources and perform real-time virtual machine migration to achieve the ability to release load configuration commands after calculation.

Cluster 300 is usually composed of one or more physical machines, and each physical machine can execute one or more virtual machines. To simplify the description, the following is only taken as an example of two physical machines, one of which is called H1 and the other of which is called H2, where H1 is a CPU with a total of twelve cores, and each The core computing power is quantized to four hundred server hosts, on which five virtual machines (virtual machine 2, virtual machine 4, virtual machine 6, virtual machine 7, virtual machine 10) are executed, and another H2 has total Sixteen core CPUs, and each core computing power is quantized to 200 server hosts, which execute five virtual machines (virtual machine 1, virtual machine 3, virtual machine 5, virtual machine 8, virtual machine) 9), and all the memory resources of each of the physical machines 400 and 500 in this embodiment satisfy the memory requirement of simultaneously executing the 10 virtual machines in the same physical machine in the embodiment. The quantified values of the CPU computing power that the virtual machine has used are shown in Table 1 below. among them, Refers to the computing power that virtual machine i has used:

Please refer to FIG. 2, which is a cluster environment in accordance with a preferred embodiment of the present invention.

A flowchart of an embodiment of a method for balancing a physical machine resource load, the process steps are: Step S101, importing a cluster resource threshold value of 5% and a virtual machine configuration mutual exclusion and group setting and other restrictions. In step S102, the historical resource usage of each virtual machine is collected for a period of time, and the resource usage rate is calculated by the CPU usage rate as a load balancing consideration of the embodiment. Step S103: Calculate the actual resource usage rate of the physical machine according to the resource usage of all the virtual machines, and further calculate the standard deviation of the cluster environment by calculating the resource usage rate of all the physical machines to determine the load between the physical machines. Balanced state and degree of dispersion. Among them, the degree of dispersion is calculated by the following mathematical expression: among them Indicates the resource usage rate of a numbered j entity machine. In this embodiment, it is the computing power that virtual machine i has used. X _i,j indicates whether a number i virtual machine is started on the physical machine of this number j. , if it has a value of 1 and is not enabled, its value is 0. The resource capabilities used by this virtual machine i, The arithmetic mean of the resource usage of the m physical machines in the cluster is represented, and σ is the standard deviation obtained from the above values to express the degree of dispersion of the resource load in the cluster, in this embodiment σ=15.8854%. The calculated standard deviation is higher than the initial defined threshold of 5%. Therefore, step S104 is performed to select two suitable physical machines. The selection method is to perform the following steps: calculating the difference between the resource usage rate and the average value of each physical machine. D _j , whose formula is: D _j = - μ ^resource ; After obtaining the difference of each physical machine, the physical machines with tight resources in the cluster can be classified into a group and called SHs. The characteristics of this group of physical machines are: D _j >0; others The physical machine is another set of SHt. D _j ranked by magnitude of distance in the physical machine and SHt is SHs group, and two groups of the same rank within the selected target entity unit circle relocation of virtual machines with each other, or select only a pair of entities ranked first according to different needs The machine is the object of relocating virtual machines to each other, that is, finding more than one pair of physical machines in the cluster for load balancing. After the paired physical machine pairing is completed, the physical machine with the resource shortage is called the source physical machine Hs and the other physical machine is the target physical machine Ht. In this example, Hs is the physical machine 500, that is, H2, and Ht is the physical machine. 400, which is H1. Find the difference between the resource used and the mean value. The mathematical expression is expressed as follows: U _t = D _t * R _t ; U _s = D _s * R _s ; that is, the total value of the target resource of the physical machine is multiplied by the difference Capability, this example yields a result of D ₁ = -15.8854%; D ₂ = 15.854%, U _s = 508.33 and U _t = -762.5; then a resource load median Thd is obtained, the mathematical expression of which is expressed as follows: Thd=( U _s - U _t )/2; then select the virtual machine on the two physical machines to perform the simulated moving movement to maximize the standard deviation of the physical machine resource usage rate, that is, the closest or lower than the initial definition. The threshold value is further executed in step S105, and the standard deviation of the physical device resource usage rate is calculated to be higher than the critical value after the simulated moving virtual machine is simulated. If the threshold value is higher than the critical value, steps S104 and S105 are repeated, otherwise step S106 is performed, and the step is performed. The mobile machine selected by S104 is moved. Fig. 3 is a detailed flow chart of step S104. When step S104 is performed, it will be disassembled into steps S201 to S205. Step S201 is to calculate a difference in load balance of each of the physical machine targets. Step S202 selects two most suitable physical machines Hi and Hj based on the result calculated in step S201, and the resource utilization rate of Hi must be greater than the resource usage rate of Hj. Step S203 is to select method 1 for the virtual machine. By selecting a most suitable virtual machine VM on the Hs in the method of selecting the virtual machine, the VM is characterized by using more resources than the source physical machine resources and the mean value. The gap and the gap are smaller than a pre-defined value according to the target resource, so that the overall system load can be balanced after moving the VM. If the appropriate virtual machine cannot be selected in step S203, the virtual machine selection method 2 is used in step S204 for selection. Step S204, by finding one of the most suitable virtual machines VMi and VMi' on the physical machine Hs and the physical machine Ht, so that the virtual machine VMi is moved to the physical machine Ht and the VMi' is moved to the physical machine Hs, the whole can be made The load of the system is the most balanced. The characteristic is that the difference after moving is not only between | Ut | and | Us | but also closest to the Thd value. The mathematical expression is as follows: ( > If the two suitable virtual machines cannot be selected in step S204, the virtual machine selection method 3 is used in step S205 for selection. In step S205, by selecting a most suitable virtual machine VMh on the physical machine Hj, after the VMh is moved, the load of the overall system can be most balanced, and the feature is that the resource load usage in the Hs is closest to the resource load. The number of virtual machines whose mathematical expressions are expressed as follows: According to the above steps, the virtual machine 8 can be selected in the method 1 to relocate the virtual machine according to the above method, and then the virtual machine is used to virtualize the virtual machine from H1 to H2; after the virtual machine is moved, the cluster load is moved. The standard deviation is reduced to 2.6042%, which will make the whole physical machine resource load dispersion less than 15.7884% before the move and less than the critical value of 5%, that is, stop the transfer calculation, and submit the migration command to the virtual resource management system to perform the virtual operation. Machine migration to achieve the goal of load balancing of physical resources.

After selecting the virtual machine included in the relocation according to the above method, use the virtual machine. The technology of management virtualization moves the virtual machine from the physical machine with tight resources to the physical machine with residual resources; after moving the virtual machine, the load of the whole physical machine resource is less than that before the move, until it meets the preset load. The discreteness threshold or the degree of dispersion is smaller than the shift calculation before the move, and the virtual resource management and control system issues the migration command to perform the virtual machine migration to achieve the goal of load balancing of the physical machine resources.

In summary, the steps used in the present invention can guide the provision of cluster computing services. How do you start with the useful data from the collected virtual machine data, and then use the virtual machine that can make the standard value of the physical device resource usage lower than the critical value to move and move to achieve the physical machine in the cluster computing environment. Load balancing design.

The above is intended to be illustrative only and not limiting. Any without departing from the invention The spirit and scope of the invention, and equivalent modifications or alterations thereto, shall be included in the scope of the appended patent application.

S101~S106‧‧‧Steps

Claims (8)

A method for balancing a physical machine resource load in a cluster environment, comprising the following steps: A, a sink cluster resource threshold, a virtual machine configuration mutual exclusion, and a group setting condition; B, collecting a virtual machine history collected for a period of time Resource usage; C. Calculate the actual resource usage rate of the physical machine based on the resource usage of all virtual machines in the physical machine, and further calculate the standard deviation in the cluster environment by calculating the resource usage rate of all the physical machines. The value is used to determine the load balance state and the degree of dispersion between the physical machines; D. Determine whether the standard deviation obtained is greater than the cluster resource threshold, and if so, select the physical machine to select at least one pair of physical machines. And selecting a virtual machine on the two selected physical machines to perform a simulated moving operation to improve the standard deviation of the physical machine resource usage rate, wherein the resource usage rate of one of the paired physical machines is More than another physical machine; E. After calculating the simulated moving virtual machine, determine whether the standard deviation of the physical machine resource usage rate is still higher than the cluster resource. Value, if yes, repeat step D, and E, if not, step F; and F., Perform the move operation of the virtual machine to another physical machine move the two sets of pairs of the selected virtual machine. The method for balancing the load of a physical machine resource in a cluster environment as described in claim 1 of the patent scope, wherein the degree of dispersion is calculated by the following mathematical formula: among them, Indicates the resource usage rate of a numbered j entity; X _i,j indicates whether a number i virtual machine is started on the physical machine of this number j, if it has a value of 1, if it is not enabled, its value is 0. ; The resource capabilities used by this virtual machine i; The arithmetic mean of the resource usage of the m physical machines in the cluster; σ is the standard deviation obtained from the above values to express the degree of dispersion of the resource load in the cluster. The method for balancing the load of a physical machine resource in the cluster environment described in claim 2, wherein the step D further includes the following steps: calculating the resource usage rate of each physical machine. Average with arithmetic The difference D _j ; and the physical machine with tight resources in the cluster are classified into one group according to the calculated difference value D _j , the other physical machines are classified into another group, and the physical machines of the two groups are classified according to the difference D The size of _j is ranked to select the same ranking or a specific ranked physical machine between the two groups as the object of relocating the virtual machine, and then screening at least one pair of paired physical machines, wherein one of the paired is the source entity Machine, the other is the target physical machine. For example, in the cluster environment described in claim 3, the method for balancing the load of the physical machine resource, wherein the difference D _{j of the} physical machine is greater than zero is classified into a group of resource shortages in the cluster, and the difference D _{j is} less than zero. The physical machine belongs to another group. The method for balancing the load of a physical machine resource in the cluster environment described in claim 3, wherein the step D further comprises the following steps: after the source entity machine and the target entity machine are selected, the source entity machine and the target entity machine are calculated. The difference between the resource and the mean, where the difference U is the value of the difference D multiplied by the total capacity R of the target resource of the physical machine; A virtual machine is selected from the source entity machine, and the virtual machine consumes more resources than the source physical machine uses the resource and the mean value, and the difference is less than a predefined value according to the target resource. The method for balancing the load of the physical machine resource in the cluster environment described in claim 5, wherein the step D further comprises the following steps: according to the difference between the resource of the source entity and the target entity and the mean U _s , U _t , Calculate the resource load median Thd; and if the virtual machine cannot be selected in the source entity machine, select a virtual machine in the source entity machine and select a virtual machine in the target physical machine, and source the source The virtual machine selected by the physical machine is moved to the target physical machine, and the virtual machine selected by the target physical machine is moved to the source physical machine, so that the difference of the resource capacity V ^resource of the two virtual machines after moving is located at | Ut | | Us | and the closest to the Thd value. A method for balancing a physical machine resource load in a cluster environment as described in claim 6 wherein the resource load median Thd = ( U _s - U _t )/2. For example, in the cluster environment described in claim 6, the method for balancing the load of the physical machine resource, wherein the step D further comprises the following steps: if the virtual machine cannot be selected in the source entity machine and the target entity machine, then A virtual machine is selected from the source entity, and the resource capability V ^{resource of the} virtual machine needs to be close to the resource load median Thd.