CN103605578A - Load balance scheduling method based on virtual machine migration - Google Patents

Abstract

The invention relates to a load balance scheduling method based on virtual machine migration. The load balance scheduling method comprises the steps of performing hotspot detection on all nodes, and determining the minimum number of the nodes to be migrated according to a bin packing problem if hotspots emerge; and constructing a migration cost model, determining a migration scheme with the least cost, migrating virtual machines according to the migration scheme with the least cost, and eliminating the hotspots to implement load balance scheduling. According to the load balance scheduling method, the hotspots which cannot be eliminated by a heuristic method can be eliminated; the global virtual machines are rearranged through dynamic migration, and the hotspots are eliminated; furthermore, the global optimization is achieved.

Description

Load equilibration scheduling method based on virtual machine (vm) migration

Technical field

The present invention relates to the resource management in cloud computing, refer to particularly a kind of load equilibration scheduling method based on virtual machine (vm) migration.

Background technology

Cloud computing (cloud computing) is increase, use and the delivery mode of the related service based on internet, and being usually directed to is provided dynamically easily expansion and be often virtualized resource by internet.Cloud is a kind of metaphor saying of network, internet.Past often represents telecommunications network with cloud in the drawings, is also used for afterwards representing the abstract of internet and underlying basis facility.Narrow sense cloud computing refers to payment and the use pattern of IT infrastructure, refer to by network with as required, the mode of easily expansion obtains resource requirement; Broad sense cloud computing refers to payment and the use pattern of service, refer to by network with as required, the mode of easily expansion obtains required service.It is relevant with software, internet that this service can be IT, also other services.It means that computing power also can be used as a kind of commodity and circulates by internet.Understand for convenience the thought of above cloud, cloud computing can be analogous to electric system and set forth and express." server zone " is similar to " generator " " electric power " resource is provided; Virtual technology is similar to " potential device " and voltage is multiplied or reduces, thereby realize elasticity, calculates; Resource Scheduler is similar to " fairing ", and the electric power that can integrate each " power house " carries out centrally connected power supply; Service managerZ-HU transmits cloud service, is similar to " power transmitting device ", can unify to provide a series of cloud services with managing I aaS (infrastructure serve), PaaS (platform serves) and SaaS (software serve) etc.

Cloud computing requires can computational resource dynamically be cut and be distributed, and this is a very very difficult thing.Intel Virtualization Technology is the answer of this problem.Virtual can realization from two levels of hardware and software in cloud computing.Some hardware, have asked to install the server of IBM AIX or (SuSE) Linux OS as IBM System PTM allows user.The cpu resource of server is managed by load manager.By to the monitoring of cpu resource and business diagnosis, load manager final decision distributes how many cpu resources to server.Differential by resource is joined (Micropartitioning) and dynamic load leveling, and load manager provides a set of powerful virtual infrastructure to user.At the beginning of the initiation of a project, its cpu resource using often seldom.Because now project is in the development phase, user's visit capacity is very little.Meanwhile, the project of some other comparative maturity may have larger visit capacity.Accordingly, the server certain hour that these projects are used has larger load.At this moment, load manager will dynamic assignment be given the more cpu resource of these servers.

The appearance of cloud computing has solved the problem of a lot of fields complexity, yet still exist many gordian techniquies in cloud computing, needs further research.Resource management in cloud computing is the important component part of cloud computing, and from providing convenience for user, the angle of service efficiently, it and traditional distributed computing environment or the resource management under cluster computing environment are similar.For cloud computing realizes, above-mentioned characteristic provides solution to the continuous maturation of Intel Virtualization Technology.Computational resource virtual makes to remap and become possibility at virtual machine and physical resource, solved well the problem of dynamic resource allocation, so the management of virtual machine becomes under cloud computing and important problem.

Intel Virtualization Technology provides a kind of effective solution for the resource management in cloud computing model.By by service encapsulates in virtual machine and be mapped on each physical server, Intel Virtualization Technology can carry out remapping of virtual machine and physical resource according to the variation of load, thereby dynamically realizes the load balancing of whole system.Remapping of virtual machine and physical resource can realize by virtual machine dynamic migration technology.At present, the dynamic migration strategy of virtual machine is based on didactic algorithm, in part, to reach optimum and the Cost Problems of considering migration mostly.

Summary of the invention

In view of above-mentioned the deficiencies in the prior art, the object of the invention is to provide a kind of load equilibration scheduling method based on virtual machine (vm) migration, and the method to overall virtual machine disposing again, is eliminated focus to realize the effect of global optimum by dynamic migration.

The technical scheme that realizes the object of the invention employing is a kind of load equilibration scheduling method based on virtual machine (vm) migration, and the method comprises:

Each node is carried out to Hot spots detection, as occurs focus, by the overweight scheduling virtual machine of load in focus to the lighter node of load, to eliminate focus, realize load balancing.

In technique scheme, described Hot spots detection comprises: detection node virtual machine desired value Rp, Rm and Rn to CPU, internal memory and bandwidth on each, when detected value exceeds preset value Cp, the Cm of this node cpu, internal memory and bandwidth and Cn, time monitor starts timing, after elapsed time interval T 0, detects again described each node, periodically repeat above-mentioned steps, if detected value falls back under preset value after n time, this node is not focus, otherwise is focus.

Further, the described load equilibration scheduling method based on virtual machine (vm) migration, also comprises:

As occurred, focus utilizes bin packing to determine the minimum node number of needs migration;

Set up moving costs model and determine the migration scheme of cost minimization, according to the migration scheme migration virtual machine of described cost minimization, eliminate described focus, realize load balancing.

The inventive method is the feature for cloud computing platform, and a kind of combination dynamic programming algorithm of proposition and the virtual machine (vm) migration strategy of moving costs, to reach overall optimum.The inventive method has had the following advantages:

1, can monitor in real time the service condition of resource in cloud computing environment, determine the generation of focus.

2, the problem of moving costs is included in to the factor of consideration, saved to a certain extent the consumption of resource.

3, can solve the indelible focus of heuritic approach, by dynamic migration, to overall virtual machine disposing again, eliminate focus and reached overall optimum.

Accompanying drawing explanation

Fig. 1 is the process flow diagram that the present invention is based on the load equilibration scheduling method of virtual machine (vm) migration.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.

As shown in Figure 1, the load equilibration scheduling method that the present invention is based on virtual machine (vm) migration comprises the following steps:

S100, each node is carried out to Hot spots detection, as there is focus in node, by the overweight scheduling virtual machine of load in focus to the lighter node of load, to eliminate the existing load balancing of focus.

Node meets certain constraint condition, and this constraint condition is embodied as:

$\begin{array}{cc}\mathrm{Rp}*\mathrm{Hi}\≤\mathrm{Cp}\left(\mathrm{ni}\right)& \∀\mathrm{Hi}\∈N\end{array}$

$\begin{array}{cc}\mathrm{RM}*\mathrm{Hi}\≤\mathrm{Cm}\left(\mathrm{ni}\right)& \∀\mathrm{Hi}\∈N\\ \mathrm{Rn}*\mathrm{Hi}\≤\mathrm{Cn}\left(\mathrm{ni}\right)& \∀\mathrm{Hi}\∈N\end{array}$

In above formula, Rp, Rm and Rn are the desired value of virtual machine to CPU, internal memory and bandwidth; Cp, Cm and Cn are the preset value of node cpu, internal memory and bandwidth; Hi=(hi1 ..., hij), if hij=1 represents there is virtual machine Vj on i node ni.According to above-mentioned three formulas, in Rp, Rm and Rn, any one value surpasses respectively Cp, Cm and Cn, thinks and occurs focus.

Because Rp, Rm and Rn are of short duration while surpassing preset value, the cost consumption moving is done by institute will be far over not doing the cost consumption moving.Therefore, the present embodiment provides the hot spot detecting method that a kind of accuracy rate is high, the method comprises: Rp, Rm and the Rn of detection node virtual machine on each, and when detected value exceeds this node Cp, Cm and Cn, monitor starts timing the control time, after elapsed time interval T 0, detect again each node, so periodically repeat above-mentioned steps n time in Hot spots detection process, if detected value falls back under preset value after n time, this node is not focus, otherwise is focus.

S200, to there is the node of focus, utilize bin packing determine to need the minimum node number of migration.Utilize the definite minimum node that meets migration constraint condition of vanning to count X, X=Σ Ui, Ui represents the virtual machine quantity on node ni.

S300, set up the migration scheme that moving costs model is determined cost minimization.

Wherein moving costs model is:

$f\left(p\right)=\underset{v_j\∈v_{\mathrm{migrate}}}{\mathrm{\Σ}}f\left(V_j\right),f\left(V_j\right)=\mathrm{Rm}\left(V_j\right)+\underset{s\∈\mathrm{prevs}\left(v_j\right)}{\mathrm{\Σ}}f\left(s\right),f\left(s\right)=\max \left(\mathrm{Rm}\left(V_j\right)\right),V_j\∈s$

In formula, f (p) represents the moving costs of p kind scheme, f(V _j) represent the moving costs of virtual machine Vj, Rm (V _j) representing to distribute to the memory size of virtual machine, f (s) is illustrated in the maximal value that s step is distributed to the internal memory of virtual machine.

S400, on cost minimum migration scheme are moved virtual machine, eliminate described focus, realize load balance scheduling.The minimum value that calculates f (p) obtains the migration scheme of cost minimization, and the present embodiment is realized the migration of virtual machine by optimal adaptation algorithm, specifically comprise following:

(1) configuration of establishing each main frame is the same, and Cp, Cm and Cn represent respectively the CPU of main frame, internal memory and bandwidth.First check the main frame of all non-NULLs, meet the main frame of constraint condition if find, find out the most suitable main frame that can put into the overweight virtual machine of load in the main frame satisfying condition, most suitable according to cost minimization judgement, cost minimization is most suitable.By the overweight virtual machine of load one, next packs main frame successively into, until the overweight virtual machine of all loads is all loaded into.

(2) if do not find the main frame that meets constraint condition in the main frame of non-NULL, enable a main frame that does not pack virtual machine into, by the overweight virtual machine of load one, next packs main frame according to the order of sequence into, until the overweight virtual machine of all loads is all loaded into.

(3) the overweight virtual machine of load eliminates focus after being completed by migration, thereby reaches overall optimum.

Claims (6)

1. the load equilibration scheduling method based on virtual machine (vm) migration, is characterized in that, comprising:

Each node is carried out to Hot spots detection, as there is focus in node, by the overweight scheduling virtual machine of load in focus to the lighter node of load, to eliminate focus, realize load balancing.

2. the load equilibration scheduling method based on virtual machine (vm) migration according to claim 1, it is characterized in that, described Hot spots detection comprises: detection node virtual machine desired value Rp, Rm and Rn to CPU, internal memory and bandwidth on each, when detected value exceeds preset value Cp, the Cm of this node cpu, internal memory and bandwidth and Cn, time monitor starts timing, after elapsed time interval T 0, detect again described each node, periodically repeat above-mentioned steps, if detected value falls back under preset value after n time, this node is not focus, otherwise is focus.

3. the load equilibration scheduling method based on virtual machine (vm) migration according to claim 2, is characterized in that:

As occurred, focus utilizes bin packing to determine the minimum node number of needs migration;

Set up moving costs model and determine the migration scheme of cost minimization, according to the migration scheme migration virtual machine of described cost minimization, eliminate described focus, realize load balance scheduling.

4. the load equilibration scheduling method based on virtual machine (vm) migration according to claim 3, is characterized in that, utilizes vanning to determine that the minimum node that meets transition condition counts X, X=Σ Ui, and Ui represents the virtual machine quantity on node ni.

5. the load equilibration scheduling method based on virtual machine (vm) migration according to claim 3, is characterized in that, described moving costs model is:

$f\left(p\right)=\underset{v_j\∈v_{\mathrm{migrate}}}{\mathrm{\Σ}}f\left(V_j\right),f\left(V_j\right)=\mathrm{Rm}\left(V_j\right)+\underset{s\∈\mathrm{prevs}\left(v_j\right)}{\mathrm{\Σ}}f\left(s\right),f\left(s\right)=\max \left(\mathrm{Rm}\left(V_j\right)\right),V_j\∈s$

In formula, f (p) represents the moving costs of p kind scheme, f(V _j) represent the moving costs of virtual machine Vj, Rm (V _j) representing to distribute to the memory size of virtual machine, f (s) is illustrated in the maximal value that s step is distributed to the internal memory of virtual machine.

6. the load equilibration scheduling method based on virtual machine (vm) migration according to claim 3, is characterized in that, the migration scheme of described cost minimization realizes described migration virtual machine by optimal adaptation algorithm, comprising:

Check the main frame of all non-NULLs, meet the main frame of constraint condition if find, the virtual machine that load is overweight packs into according to the order of sequence;

If do not find the main frame that meets constraint condition in the main frame of non-NULL, enable a main frame that does not pack virtual machine into, the virtual machine that its load is overweight is put into according to the order of sequence;

Described constraint condition is that on main frame, virtual machine is less than or equal to respectively preset value Cp, Cm and the Cn of this host CPU, internal memory and bandwidth to desired value Rp, the Rm of CPU, internal memory and bandwidth and Rn.