Virtual machine management method based on prediction of virtual machine workload prediction for virtual machines deployed on servers and virtual machine management system implementing the same
A virtual machine (VM) management method may involve simulating a change in deployment of VMs deployed on physical servers including a first physical server and a second physical server physically separated from the first physical server. Server workload prediction based on possible VM deployment may be used for scheduling deployment of VMs to obtain a satisfactory deployment of the VMs on the servers. This may involve classification of storage loads of the VMs, forming a prediction model for predicting prediction load based on storage loads in a classification, and selecting a prediction model for use in determining a target prediction load on a physical server as a result of target VM being analyzed.
1. A virtual machine (VM) management method based on a virtual machine workload prediction method implemented by a prediction device and a virtual machine deployment device, which are computing devices, to calculate a prediction load, which is a load on a physical server predicted to be imposed by a plurality of VMs installed on the physical server, the method comprising:
collecting storage load information which is information related to a storage load corresponding to a load on the physical server caused by the virtual machines installed on the physical server;
classifying storage loads of the virtual machines into groups by a predetermined classification method;
forming, when the storage loads are arranged on the basis of an arrangement reference, a prediction model for calculating the prediction load in units of the storage loads classified into a same group;
selecting at least one of a plurality of prediction models by a predetermined selection method to increase prediction accuracy of a target prediction load which is a load on the physical server predicted to be caused by a target virtual machine to be analyzed;
calculating target prediction load information which is information related to the target prediction load according to a predetermined calculation method on the basis of target storage load information, which is information related to a target storage load corresponding to a load on the physical server caused by the target virtual machine, and a selection model which is the at least one selected prediction model; and
deploying at least one of the virtual machines on the physical server based upon the target prediction load information,
wherein the predetermined selection method includes a predetermined first selection method of selecting, when the target storage load is located in a group of storage loads after the target storage load is arranged on the basis of the arrangement reference, a prediction model produced on the basis of the storage loads classified into the group in which the target storage load is located as the selection model,
wherein the predetermined selection method includes a predetermined second selection method of selecting, when the target storage load is not located in any group of storage loads among all groups of storage loads after the target storage load is arranged on the basis of the arrangement reference, prediction models corresponding to groups of storage loads satisfying a predetermined condition and apart from the position of the target storage load as the selection model,
wherein the predetermined calculation method is a method of calculating, when the selection model is singular, the target prediction load information by inputting the target storage load information to the selection model,
wherein the predetermined calculation method is a method of calculating, when the selection model has a plurality of selection models, the target prediction load information by merging a plurality of target preliminary loads on a basis of weights calculated on a basis of a ratio of separation distances from the target storage load to storage load groups corresponding to the selection models, and
wherein a plurality of pieces of target preliminary load information related to a plurality of a target preliminary loads is calculated by inputting the target storage load information to the selection models.
2. The virtual machine management method of claim 1 , wherein the storage load is a degree of operation of a CPU of a physical server resulting from running virtual machines.
3. The virtual machine management method of claim 1 , wherein the arrangement reference is a reference related to length of a task time and change rates of storage loads during the task time.
4. The virtual machine management method of claim 1 , wherein the predetermined condition requires the group to be located within a certain distance from the target storage load when the target storage load is arranged on the basis of the arrangement reference.
5. The virtual machine management method of claim 1 , the predetermined condition requires the group to be located within a reference range from a reference distance which is a shortest one of distances from the target storage load to the groups of storage loads when the target storage load is arranged on the basis of the arrangement reference.
6. The virtual machine management method of claim 1 , the predetermined calculation method is a method in which a given one of the weights is decreased as the distance between the target storage load and a storage load group corresponding to the selection model associated with the given weight increases.
7. The virtual machine management method of claim 6 , the predetermined calculation method is a method in which the given one of the weights is increased as the distance between the target storage load and the storage load group corresponding to the selection model associated with the given weight decreases.
8. The virtual machine management method of claim 1 , the predetermined calculation method is a method of calculating the target prediction load information for a certain time period.
9. A virtual machine management system for deploying at least one virtual machine on at least one physical server based upon a load on a physical server predicted to be imposed by a plurality of virtual machines installed on the physical server, the virtual machine management system comprising:
a prediction device, which is a computing device, comprising a prediction storage storing storage load information which is information related to a storage load corresponding to a load on the physical server caused by the virtual machines installed on the physical server;
a prediction controller performing computation including processing information required for implementing the virtual machine workload prediction method; and
a virtual machine deployment device, which is a computing device, comprising a deployment processor deploying at least one of the virtual machines on at least one of the physical server based upon target prediction load information;
wherein the virtual machine workload prediction method comprises:
collecting storage load information which is information related to a storage load corresponding to a load on the physical server caused by the virtual machines installed on the physical server;
classifying storage loads of the virtual machines into groups by a predetermined classification method;
forming, when the storage loads are arranged on the basis of an arrangement reference, a prediction model for calculating the prediction load in units of the storage loads classified into a same group;
selecting at least one of a plurality of prediction models by a predetermined selection method to increase prediction accuracy of a target prediction load which is a load on the physical server predicted to be caused by a target virtual machine to be analyzed; and
calculating target prediction load information which is information related to the target prediction load according to a predetermined calculation method on the basis of target storage load information, which is information related to a target storage load corresponding to a load on the physical server caused by the target virtual machine, and a selection model which is the at least one selected prediction model,
wherein the predetermined selection method includes a predetermined first selection method of selecting, when the target storage load is located in a group of storage loads after the target storage load is arranged on the basis of the arrangement reference, a prediction model produced on the basis of the storage loads classified into the group in which the target storage load is located as the selection model,
wherein the predetermined selection method includes a predetermined second selection method of selecting, when the target storage load is not located in any group of storage loads among all groups of storage loads after the target storage load is arranged on the basis of the arrangement reference, prediction models corresponding to groups of storage loads satisfying a predetermined condition and apart from the position of the target storage load as the selection model,
wherein the predetermined calculation method is a method of calculating, when the selection model is singular, the target prediction load information by inputting the target storage load information to the selection model,
wherein the predetermined calculation method is a method of calculating, when the selection model has a plurality of selection models, the target prediction load information by merging a plurality of target preliminary loads on a basis of weights calculated on a basis of a ratio of separation distances from the target storage load to storage load groups corresponding to the selection models, and
wherein a plurality of pieces of target preliminary load information related to a plurality of a target preliminary loads is calculated by inputting the target storage load information to the selection models.