CN105430049A - Virtual streaming cluster cooperative migration method based on DCN - Google Patents

Abstract

The invention provides a virtual streaming cluster cooperative migration method based on a DCN, comprising: first, building a flow communication topological graph among virtual streaming servers; performing small cluster topological division specific to physical machines and large cluster topological division specific to partitions according to the flow communication topological graph among virtual streaming servers; designating the corresponding relation between a partition large cluster and a partition, and the corresponding relation between a physical machine small cluster and a physical machine according to positions where the virtual streaming servers are located and two level topological division results; and finally, migrating all virtual machines to be migrated to target physical machines through a virtual streaming cluster cooperative migration method.

Description

A kind of virtual streaming media cluster based on DCN works in coordination with moving method

Technical field

The invention belongs to cloud computing DCN technical field, relate to the division of streaming media cluster under DCN environment, placement and migration field, particularly a kind of virtual streaming media cluster based on DCN works in coordination with moving method.

Background technology

The development of nowadays cloud computing is subject to people and more and more pays close attention to, related application based on cloud computing also incorporates among the daily life of people gradually, therefore in order to meet the various demands of user, cloud service supplier constantly affixes one's name to large-scale application service at cloud platform upper.Along with user is to the increase of the resource quantity of data center and resource category demand, expanding data center scale has not been a kind of effective solution simply, because this can increase the construction cost of data center greatly, also no way simultaneously reduces application service quality.Therefore, in order to address this problem, researcher proposes the solution of multi-dummy machine migration under cloud computing environment.The following several sections of related to the present invention patents belonging to cloud computing DCN field retrieved, they respectively:

1. Chinese patent 201510036992.8, the data migration method of a kind of data center multi-dummy machine;

2. Chinese patent 201210551631.3, a kind of application cluster moving method and device;

In above-mentioned patent 1, provide the data migration method of a kind of data center multi-dummy machine, belong to field of cloud calculation.The present invention is based on the parallel multi-dummy machine migration strategy based on precloning that FrancoCallegati and WalterCerroni proposes; and propose in the present invention based on after based on the serial multi-dummy machine migration strategy that copies; introduce parameter m and modifying factor α; propose a kind of meet the constraints of the maximum downtime that service provider and user consult under; the data migration method of gross migration minimal time; the method is applied widely, can reduce the transit time cost of service provider further.

In above-mentioned patent 2, disclose a kind of application cluster moving method and device, the method comprises: in cloud computing platform, receives the application cluster migration request that user sends according to the agreement of migration framework; Described application cluster migration request is resolved, determines the migration attribute information of asking to carry out the application cluster moved; According to the migration attribute information determined, in idling-resource pond, determine the node resource meeting application cluster migration demand, according to the node resource determined and migration framework, the application cluster of request migration is moved in the node resource determined.Adopt such scheme, the transport efficiency of application cluster in cloud computing platform can be improved preferably.

Look into newly according to above-mentioned, problem existing for prior art is, all do not consider the feature of streaming media server height resource consumption and tight traffic communication, this can cause the heavy congestion of whole system for cloud computing, the internal bandwidth of a large amount of consumption network, and then the performance having a strong impact on Stream Media Application service.

Summary of the invention

In order to overcome the shortcoming of above-mentioned prior art, a kind of virtual streaming media cluster based on DCN is the object of the present invention is to provide to work in coordination with moving method, under the prerequisite ensureing the external service performance of Stream Media Application, significantly reduce the internal bandwidth consumption of DCN network, reduce the bulk migration time cost of streaming media cluster simultaneously.

To achieve these goals, the technical solution used in the present invention is:

Virtual streaming media cluster based on DCN works in coordination with a moving method,

First, the traffic communication topological diagram between virtual stream media server is built;

The little cluster topology division towards physical machine and the large cluster topology division towards subregion is carried out according to the traffic communication topological diagram between virtual stream media server;

According to the Topology partition result of virtual stream media server position and 2 levels, assign towards the large cluster of subregion and subregion corresponding relation and assign the corresponding relation of little cluster towards physical machine and physical machine;

Finally, work in coordination with moving method by virtual streaming media cluster, all virtual machine (vm) migrations to be migrated in object physical machine.

Traffic communication topological diagram between described structure virtual stream media server is expressed as Gvm=(V, E), wherein V={Vm ₁, Vm ₂... Vm _n, namely each virtual machine is a point in traffic communication topological diagram, and n is virtual machine number, and E is limit collection, represents between virtual machine whether there is communication; If there is communication between virtual machine i with virtual machine j, then E _ij=1, otherwise E _ij=0; Limit weights W _ijfor the communication size between virtual machine i and virtual machine j; Virtual machine i is expressed as Vm _i=(c, b), wherein c represents the cpu resource needed for virtual machine i, and b represents the bandwidth resources needed for virtual machine i.

It is as follows that the described little cluster topology towards physical machine divides concrete steps:

Step1: each virtual machine of initialization is a little cluster towards physical machine;

Step2: calculate limit gain, the percentage of all limits weights sum of the institute's tie point that square is directly proportional, accounts for limit weights of limit gain and limit weights is directly proportional, the resource of limit tie point and being inversely proportional to;

Step3: if there is limit gain to be greater than 0, then 2 points merging the maximum limit connection of limit gain are a little cluster towards physical machine;

Step4: repeat Step2, until the gain of all limits is all not more than 0;

Step5: merging remaining isolated node is a little cluster towards physical machine.

Described basis is as follows towards the concrete steps of the large cluster topology division of subregion:

Step1: initialization is a large cluster towards subregion towards the little cluster of physical machine;

Step2: calculate limit gain, the percentage of all limits weights sum of the institute's tie point that square is directly proportional, accounts for limit weights of limit gain and limit weights is directly proportional, the resource of limit tie point and being inversely proportional to;

Step3: if there is limit gain to be greater than 0, then 2 the little clusters merging the maximum limit connection of limit gain are a large cluster towards subregion;

Step4: repeat Step2, until the gain of all limits is all not more than 0;

Step5: merging remaining isolated little cluster is a large cluster towards subregion.

Described appointment is as follows towards the concrete steps of the corresponding relation of the large cluster of subregion and subregion:

Step1: internal storage data total amount formula Data (i) copied in virtual machine (vm) migration process=B*Vm _i(c)/(B-Vm _i(b)) represent, in formula: B represents transmission rate, Vm _ic () represents virutal machine memory size, Vm _ib () represents the dirty page generation rate of virutal machine memory;

Step2: the distance of virtual machine and subregion copies internal storage data total amount and its product representation to the communication switchboard number of section post process with virtual machine (vm) migration;

Step3: the distance towards the large cluster of subregion and subregion is all virtual machines of comprising towards the large cluster of subregion and subregion distance sum;

Step4: according to the large cluster towards subregion to the distance of subregion, use Hungary's dispatching algorithm to try to achieve optimal distributing scheme.

Described appointment is as follows towards the concrete steps of the corresponding relation of the little cluster of physical machine and physical machine:

Step1: if deploying virtual machine is in physical machine, then virtual machine and physical machine distance are 0, otherwise are 1;

Step2: the distance towards the little cluster of physical machine and physical machine is all virtual machines of comprising towards the little cluster of physical machine and physical machine distance sum;

Step3: according to the little cluster towards physical machine to the distance of physical machine, use Hungary's dispatching algorithm to try to achieve optimal distributing scheme.

Describedly work in coordination with moving method by virtual streaming media cluster, as follows to the concrete steps in object physical machine for all virtual machine (vm) migrations to be migrated:

Step1: judge whether the destination host resource of virtual machine to be migrated meets, virtual machine to be migrated is divided into transportable queue and etc. queue to be migrated;

Step2: calculate the network gain that the virtual machine (vm) migration in transportable queue brings, moves by it the network gain descending that brings, the inner stream flow size of DCN of virtual machine (vm) migration gain for reducing after virtual machine (vm) migration;

Step3: if transportable queue is empty, go to Step6; Otherwise from getting virtual machine in order transportable queue, calculate the migration path of virtual machine, migration path bandwidth is expressed as B;

Step4: if the dirty page generation rate of virutal machine memory that migration path bandwidth B is greater than 1.5 times, then do not meet transition condition, go to Step3; Otherwise, start its migration task;

Step5: virtual machine completes migration, upgrades the migration gain of the virtual machine be attached thereto, upgrades the available resources of source physical host, upgrades transportable queue and waits for resource queue, upgrades DCN network topology matrix available bandwidth; Go to Step3;

Step6: upgrading object physical host can resource, upgrade transportable and etc. queue to be migrated, upgrade DCN network topology matrix available bandwidth; Go to Step3;

Step7: whether the queues to be migrated such as judgement are empty, and if it is empty, then algorithm terminates; Otherwise, by the virtual machine (vm) migration in physical machine minimum for virtual machine number to be moved out to free physical machine.

Compared with prior art, contemplated by the invention the feature of streaming media server height resource consumption and tight traffic communication, carry out the collaborative migration of virtual streaming media cluster by the ideal migration scheme of trying to achieve virtual streaming media cluster.

Accompanying drawing explanation

Fig. 1 is the flow chart that a kind of virtual streaming media cluster based on DCN of the present invention works in coordination with moving method.

Embodiment

Below in conjunction with drawings and Examples, the present invention is described in further details.

As shown in Figure 1, the present invention is that a kind of virtual streaming media cluster based on DCN works in coordination with moving method, first, builds the traffic communication topological diagram between virtual stream media server; The little cluster topology division towards physical machine and the large cluster topology division towards subregion is carried out according to the traffic communication topological diagram between virtual stream media server; According to the Topology partition result of virtual stream media server position and 2 levels, assign towards the corresponding relation of the large cluster of subregion and subregion with towards the little cluster of physical machine and the corresponding relation of physical machine respectively; Finally, work in coordination with moving method by virtual streaming media cluster, all virtual machine (vm) migrations to be migrated in object physical machine.

Step by step technical scheme of the present invention is described in detail below.

1, traffic communication topological diagram between virtual stream media server is built

The traffic communication topological diagram built between virtual stream media server is expressed as Gvm=(V, E), wherein V={Vm ₁, Vm ₂... Vm _n, namely each virtual machine is a point in traffic communication topological diagram, and n is virtual machine number, and E is limit collection, represents between virtual machine whether there is communication; If there is communication between virtual machine i with virtual machine j, then E _ij=1, otherwise E _ij=0; Limit weights W _ijfor the communication size between virtual machine i and virtual machine j; Virtual machine i is expressed as Vm _i=(c, b), wherein c represents the cpu resource needed for virtual machine i, and b represents the bandwidth resources needed for virtual machine i.

2, the little cluster topology towards physical machine divides

According to the thinking that the traffic communication topological diagram between virtual stream media server carries out dividing towards the little cluster topology of physical machine be:

Initialized little cluster network topological representation is: Gsc=(Vsc, E), and its mid point is the little cluster that a virtual machine is formed, and namely first each virtual machine is initialized as a little cluster as the point of in figure, is expressed as Vsc _i=(c, b, d), now c represents again cpu resource needed for little cluster simultaneously, and b represents again bandwidth resources needed for little cluster simultaneously, and d represents little cluster internal all limits weights sum.Virtual machine i is initialized as little cluster i, and as the some i in figure, virtual machine j is initialized as little cluster j, and as the some j in figure, if there is a limit e between little cluster i and little cluster j, then weights are expressed as w _ij=Evsc _ij+ Evsc _ji, if there is not limit between little cluster i and little cluster j, then w in i.e. uninterrupted sum between little cluster _ij=0.

Little assemblage classification towards physical machine is to arrive maximum.

E _ijand a _icomputing formula be:

$a_i=\underset{j}{\Σ}{e}_{ij}$

E in formula _ijrepresent that the limit weights of tie point i and some j put a spot i and some j internal edges weights sum.A _irepresent the weights sum on the point all limits of i.Because needs consider node size, the i.e. demand of resources of virtual machine, the computing formula of limit gain delta Q:

M＝max{(Vsc _i(c)+Vsc _j(c))/C,(Vsc _i(b)+Vsc _j(b))/B}

$\mathrm{\&Delta;}Q=\left\{\begin{array}{c}\frac{e_{ij}*w_{ij}^2}{M*a_i*a_j};M<=1\\ 0;M>1\end{array}\right.$

In formula, w _ijrepresent the weights on the limit of tie point i and some j; C represents the internal memory of physical machine; B represents the bandwidth of physical machine; M represents the percentage taking physical machine resource after an i and some j merges.If physical machine resource does not meet its resource requirement altogether after two little clusters merge, then Δ Q is minimum value 0.

Concrete steps are:

Step1: input data: virtual machine discharge relation figure (matrix) Gvm=(V, E), physical machine memory size C, physical machine amount of bandwidth B.

Step2: using each for Gvm=(V, E) virtual machine as a little cluster, be initialized as Gsc=(Vsc, E), in Gsc, each point is a little cluster.

Step3: calculate e _ijand a _imeet:

$a_i=\underset{j}{\&Sigma;}{e}_{ij}$

Step4: calculate the limit gain delta Q after each little cluster merging having limit to be connected successively:

M＝max{(Vsc _i(c)+Vsc _j(c))/C,(Vsc _i(b)+Vsc _j(b))/B}

$\mathrm{\&Delta;}Q=\left\{\begin{array}{c}\frac{e_{ij}*w_{ij}^2}{M*a_i*a_j};M<=1\\ 0;M>1\end{array}\right.$

Step5: if there is not Δ Q > 0, then terminate algorithm; Otherwise, merge the maximum some i corresponding to limit of Δ Q and some j.After merging, point is Vsc _new:

Vsc _new(c)＝Vsc _i(c)+Vsc _j(c)

Vsc _new(b)＝Vsc _i(b)+Vsc _j(b)

Vsc _new(d)＝Vsc _i(d)+Vsc _j(d)+w _ij

Step6: upgrade the limit weights between all points be connected with a j with an i, w _{new, k}=w _ik+ w _jk, go to Step3.

3, the large cluster topology towards subregion divides

Gvm=(V, E) is divided into the Gsc=(Vsc, E) obtained after the little cluster of physical machine, Vsc _irepresent little cluster i, Wsc _ijrepresent the weights on the limit between little cluster i and little cluster j, Vpc _i=(h, d) represents large cluster i, Vpc _ih () represents the number of the little cluster comprised in large cluster i, Vpc _id () represents the limit weights between the little cluster that namely large cluster i internal edges weights comprise, Wpc _ijrepresent the weights between large cluster i and large cluster j, N represents the number of the physical machine contained in a subregion.

Employ community detecting algorithm equally towards the large assemblage classification algorithm of subregion, first provide e ' _ijwith a ' _icomputing formula:

$a_i^{\&prime;}=\underset{j}{\&Sigma;}{e}_{ij}^{\&prime;}$

E ' in formula _ijrepresent that the limit weights of tie point i and some j put a spot i and some j internal edges weights sum.A ' _irepresent the weights sum on the point all limits of i.Provide the computing formula of limit gain delta Q:

M＝(Vpc _i(h)+Vsc _j(h))/N

$\mathrm{\&Delta;}Q=\left\{\begin{array}{c}\frac{e_{ij}^{\&prime;}*{\mathrm{Wpc}}_{ij}^2}{M*a_i^{\&prime;}*a_j^{\&prime;}};M<=1\\ 0;M>1\end{array}\right.$

In formula, Wpc _ijrepresent the weights on the limit of tie point i and some j; N represents the number of physical machine in subregion; M represents that an i and some j takies physical machine number percentage in subregion after merging.If physical machine number is greater than physical machine number in subregion after two large clusters merge, then Δ Q is minimum value 0.

Concrete steps are as follows:

Step1: input data: towards the little cluster topology graph Gsc'=(Vsc, E) of physical machine, physical machine number N in subregion.

In Step2:Gsc'=(Vsc, E), each little cluster is as a large cluster, is expressed as, and in Gpc, each point is a little cluster.

Step3: calculate e ' _ijwith a ' _imeet:

$a_i^{\&prime;}=\underset{j}{\&Sigma;}{e}_{ij}^{\&prime;}$

Step4: calculate the limit gain delta Q after each large cluster merging having limit to be connected successively:

M＝(Vpc _i(h)+Vsc _j(h))/N

$\mathrm{\&Delta;}Q=\left\{\begin{array}{c}\frac{e_{ij}^{\&prime;}*{\mathrm{Wpc}}_{ij}^2}{M*a_i^{\&prime;}*a_j^{\&prime;}};M<=1\\ 0;M>1\end{array}\right.$

Step5: if there is not Δ Q > 0, then terminate algorithm.Otherwise, merge some i, j corresponding to limit that Δ Q is maximum.After merging, point is Vpc _new:

Vpc _new(h)＝Vpc _i(h)+Vpc _j(h)

Vpc _new(d)＝Vpc _i(d)+Vpc _j(d)+w _ij

Step6: upgrade limit weights between all points be connected with a j with an i, w _{new, k}=w _ik+ w _jk, go to Step3.

4, the corresponding relation of large cluster towards subregion and subregion is assigned

Gpc=(Vpc, E) large cluster topology graph, wherein Vpc _iinclude multiple little cluster, little cluster has multiple virtual machine; D ^partion(i, j) represents the switch number in DCN between subregion, and different DCN frameworks is different, is the Tree framework of p mouth, can be formulated for Guinier-Preston zone switch ports themselves:

In formula: i, j represent subregion label, from left to right number consecutively; represent that subregion is under same aggregation switch.Only pass through time access switch with a point intra-area communication, under same aggregation switch, different subregion communication is through 3 switches, and other situations will through 5 switches.

CostVm (i, j) represents the matrix between virtual machine and each subregion; CostCluster (i, j) represents the distance matrix between large cluster and each subregion; Assigned (i, j) represents appointment matrix, and this is the solution required by this algorithm, and assigned (i, j)=0 represents that large cluster i is assigned to subregion j.PartionNow (i) represents the partition functions at the present place of virtual machine, and this is known quantity.PartionTo (i) represents the partition functions to be migrated of virtual machine, if assigned (i, j)=1, PartionTo (i)=j; Data (i) represents virtual machine (vm) migration transmitted data amount; M represents large cluster number; N represents subregion number.

1) large cluster and subregion distance carry out defining and calculating

Virtual machine (vm) migration process need copies whole internal memory, and virtual machine (vm) migration to be estimated the migration data amount that produces at the DCN distance as virtual machine and subregion to the transition process of subregion.Virtual machine (vm) migration process will be transmitted whole internal memory, the constantly dirty page of copy and shut down the dirty page of copy residue.If virtual machine Vm _iinside save as Vm _ic (), it is Vm that dirty page produces speed _i(r), link transmission rate B, then virtual machine Vm _imigration needs transmitted data amount to be altogether:

$Data\left(i\right)={\&Sigma;}_{k=1}^n{\mathrm{Vm}}_i\left(c\right)*{\left({\mathrm{Vm}}_i\left(b\right)/B\right)}^{k-1}$

Virtual machine (vm) migration first time iteration will send whole internal memory, and consuming time is Vm _ic ()/B, this process creates (Vm _i(c) * Vm _i(b)) data of/B, second time iteration sends these data.Third time iteration by transmission by Vm consuming time _i(c) * (Vm _i(b)/B) ²data.If do not consider background traffic, namely give tacit consent to link bandwidth and be B and available.Stopping criterion for iteration is for reaching maximum iteration time n _maxor need the data of transmission to be less than presetting threshold value.

Conveniently calculate, propose the simplified model of total transmission data.The data transfer model simplified is formulated:

Data ^simple(i)＝B*Vm _i(c)/(B-Vm _i(b))

In formula: Vm _i(c)/(B-Vm _i(b)) illustrate the transmission time total in simplified model, namely simply think that (dirty page produces and adds internal memory under dirty page constantly produces and is constantly transmitted double action with internal memory, internal memory is transmitted and decreases internal memory), internal memory is all transmitted and namely completes the memory copying stage.The time that the memory copying stage uses is multiplied with the speed of transmission and is total data volume transmitted.

The distance of virtual machine and subregion is expressed as:

CostVm(i,j)＝Data(i)*D ^partion(PartionNow(i),j)

Namely virtual machine i moves the communication switchboard number needing the data volume of transmission to be multiplied by its place subregion and j.

The distance of large cluster and subregion is with being expressed as:

$CostCluster(i,j)=\underset{{\mathrm{Vm}}_k\&Element;{\mathrm{Vpc}}_i}{\&Sigma;}CostVm(k,j)$

I.e. all virtual machines of comprising of large cluster and subregion distance sum.

2) according to the large cluster of calculating and the distance of each subregion, ask always apart from minimum allocative decision

Need to distribute large cluster and subregion after trying to achieve CostCluster (i, j), a large cluster can only put into a subregion, and a subregion can only put a large cluster.Optimization aim is:

$\min Z=\underset{i=1}{\overset{m}{\&Sigma;}}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{CostCluster}}_{ij}*{\mathrm{assigned}}_{ij}$

$s.t\left\{\begin{array}{c}\underset{i=1}{\overset{m}{\&Sigma;}}{\mathrm{assigned}}_{ij}=1(j=1,......n)\\ \underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{assigned}}_{ij}=1(i=1,......m)\\ {\mathrm{assigned}}_{ij}=0,1\end{array}\right.$

In formula: represent that each subregion can only place a large cluster; represent that each large cluster can only put into a subregion.Assigned _ijbe that the large cluster i of 1 expression is assigned to subregion j.

5, the corresponding relation of little cluster towards physical machine and physical machine is assigned

Gsc=(Vsc, E) represent that chapters and sections 3.3 generate towards the little cluster topology of physical machine; CostVm (i, j) represents the distance of virtual machine i and physical machine j; CostCluster (i, j) represents the distance of little cluster i and physical machine j, and its value equals all virtual machines and physical machine j distance sum in little cluster; Assigned (i, j) represents the assignment matrix of virtual machine and physical machine, assigned _ij=1 represents that virtual machine i has been assigned to physical machine j.The physical machine that HostNow (i) virtual machine is now disposed; M represents the little cluster number in large cluster, generally m==n; N represents physical machine number in subregion, and this value is a constant; ListVsc represents all little cluster contained of specific large cluster; Hosts represents all physical machine in particular zones.

1) little cluster and physical machine distance

Virtual machine must move to the migration cost of physical machine to be selected and there is relation virtual machine present position, if there is certain physical machine, does not need migration, and virtual machine therewith physical machine migration cost is 0.On virtual machine (vm) migration to its non-existent physical machine, cost is the same, because these physical machine are positioned at same subregion, is equal to wherein any one distance.Therefore virtual machine and physical machine distance are:

The little cluster that can arrive and the range formula of physical machine are:

$CostCluster(i,j)=\underset{{\mathrm{Vm}}_k\&Element;{\mathrm{Vsc}}_i}{\&Sigma;}CostVm(k,j)$

2) according to little cluster and physical machine distance, the allocative decision that distance summation is minimum is found

CostCluster (i, j) is the coefficient matrix of input, and use Hungary Algorithm just optimum distribution solves.

6, Hungary's dispatching algorithm

(1) if from coefficient matrix c _ijrow (or row) each element in deduct the least member of this row (or row) respectively after obtain a new matrix b _ij, then with b _ijfor the Assignment Problems of coefficient matrix has identical optimal solution with former problem.After above-mentioned conversion, b _ijin often row and often arrange all at least containing 0 element, 0 element claiming to be positioned at different rows different lines is independently 0 element.

(2) if b _ijhave n independently 0 element, can obtain a dematrix thus, method is that order corresponds to b in X _ijthe element of 0 element position be 1, the element of other position is 0, then X is the optimal solution of Assignment Problems.

7, virtual streaming media cluster works in coordination with moving method

Moving method is worked in coordination with, as follows to the concrete steps in object physical machine for all virtual machine (vm) migrations to be migrated by virtual streaming media cluster:

Step1: judge whether the destination host resource of virtual machine to be migrated meets, virtual machine to be migrated is divided into transportable queue and etc. queue to be migrated;

Step2: calculate the network gain that the virtual machine (vm) migration in transportable queue brings, moves by it the network gain descending that brings, the inner stream flow size of DCN of virtual machine (vm) migration gain for reducing after virtual machine (vm) migration;

Step3: if transportable queue is empty, go to Step6; Otherwise from getting virtual machine in order transportable queue, calculate the migration path of virtual machine, migration path bandwidth is expressed as B;

Step4: if the dirty page generation rate of virutal machine memory that migration path bandwidth B is greater than 1.5 times, then do not meet transition condition, go to Step3; Otherwise, start its migration task;

Step5: virtual machine completes migration, upgrades the migration gain of the virtual machine be attached thereto, upgrades the available resources of source physical host, upgrades transportable queue and waits for resource queue, upgrades DCN network topology matrix available bandwidth; Go to Step3;

Step6: upgrading object physical host can resource, upgrade transportable and etc. queue to be migrated, upgrade DCN network topology matrix available bandwidth; Go to Step3;

Step7: whether the queues to be migrated such as judgement are empty, and if it is empty, then algorithm terminates; Otherwise, by the virtual machine (vm) migration in physical machine minimum for virtual machine number to be moved out to free physical machine.

Claims (7)

1. the virtual streaming media cluster based on DCN works in coordination with a moving method, it is characterized in that, comprises the steps:

First, the traffic communication topological diagram between virtual stream media server is built;

Then, the little cluster topology division towards physical machine and the large cluster topology division towards subregion is carried out according to the traffic communication topological diagram between virtual stream media server;

Then, according to the Topology partition result of virtual stream media server position and 2 levels, assign towards the large cluster of subregion and subregion corresponding relation and assign the corresponding relation of little cluster towards physical machine and physical machine;

Finally, work in coordination with moving method by virtual streaming media cluster, all virtual machine (vm) migrations to be migrated in object physical machine.

2. work in coordination with moving method based on the virtual streaming media cluster of DCN according to claim 1, it is characterized in that, the traffic communication topological diagram between described virtual stream media server is expressed as Gvm=(V, E), wherein V={Vm ₁, Vm ₂... Vm _n, namely each virtual machine is a point in traffic communication topological diagram, and n is virtual machine number, and E is limit collection, represents between virtual machine whether there is communication; If there is communication between virtual machine i with virtual machine j, then E _ij=1, otherwise E _ij=0; Limit weights W _ijfor the communication size between virtual machine i and virtual machine j; Virtual machine i is expressed as Vm _i=(c, b), wherein c represents the cpu resource needed for virtual machine i, and b represents the bandwidth resources needed for virtual machine i.

3. work in coordination with moving method based on the virtual streaming media cluster of DCN according to claim 1, it is characterized in that, it is as follows that the described little cluster topology towards physical machine divides concrete steps:

Step1: each virtual machine of initialization is a little cluster towards physical machine;

Step2: calculate limit gain, the percentage of all limits weights sum of the institute's tie point that square is directly proportional, accounts for limit weights of limit gain and limit weights is directly proportional, the resource of limit tie point and being inversely proportional to;

Step3: if there is limit gain to be greater than 0, then 2 points merging the maximum limit connection of limit gain are a little cluster towards physical machine;

Step4: repeat Step2, until the gain of all limits is all not more than 0;

Step5: merging remaining isolated node is a little cluster towards physical machine.

4. work in coordination with moving method based on the virtual streaming media cluster of DCN according to claim 1, it is characterized in that, the concrete steps of the described large cluster topology division towards subregion are as follows:

Step1: initialization is a large cluster towards subregion towards the little cluster of physical machine;

Step2: calculate limit gain, the percentage of all limits weights sum of the institute's tie point that square is directly proportional, accounts for limit weights of limit gain and limit weights is directly proportional, the resource of limit tie point and being inversely proportional to;

Step3: if there is limit gain to be greater than 0, then 2 the little clusters merging the maximum limit connection of limit gain are a large cluster towards subregion;

Step4: repeat Step2, until the gain of all limits is all not more than 0;

Step5: merging remaining isolated little cluster is a large cluster towards subregion.

5. work in coordination with moving method based on the virtual streaming media cluster of DCN according to claim 1, it is characterized in that, described appointment is as follows towards the concrete steps of the corresponding relation of the large cluster of subregion and subregion:

Step1: internal storage data total amount formula Data (i) copied in virtual machine (vm) migration process=B*Vm _i(c)/(B-Vm _i(b)) represent, in formula: B represents transmission rate, Vm _ic () represents virutal machine memory size, Vm _ib () represents the dirty page generation rate of virutal machine memory;

Step2: the distance of virtual machine and subregion copies internal storage data total amount and its product representation to the communication switchboard number of section post process with virtual machine (vm) migration;

Step3: the distance towards the large cluster of subregion and subregion is all virtual machines of comprising towards the large cluster of subregion and subregion distance sum;

Step4: according to the large cluster towards subregion to the distance of subregion, use Hungary's dispatching algorithm to try to achieve optimal distributing scheme.

6. the virtual streaming media cluster based on DCN according to claim 1 works in coordination with moving method, it is characterized in that, described appointment is as follows towards the concrete steps of the corresponding relation of the little cluster of physical machine and physical machine:

Step1: if deploying virtual machine is in physical machine, then virtual machine and physical machine distance are 0, otherwise are 1;

Step2: the distance towards the little cluster of physical machine and physical machine is all virtual machines of comprising towards the little cluster of physical machine and physical machine distance sum;

Step3: according to the little cluster towards physical machine to the distance of physical machine, use Hungary's dispatching algorithm to try to achieve optimal distributing scheme.

7. the virtual streaming media cluster based on DCN according to claim 1 works in coordination with moving method, it is characterized in that, describedly works in coordination with moving method by virtual streaming media cluster, as follows to the concrete steps in object physical machine for all virtual machine (vm) migrations to be migrated:

Step1: judge whether the destination host resource of virtual machine to be migrated meets, virtual machine to be migrated is divided into transportable queue and etc. queue to be migrated;

Step2: calculate the network gain that the virtual machine (vm) migration in transportable queue brings, moves by it the network gain descending that brings, the inner stream flow size of DCN of virtual machine (vm) migration gain for reducing after virtual machine (vm) migration;

Step3: if transportable queue is empty, go to Step6; Otherwise from getting virtual machine in order transportable queue, calculate the migration path of virtual machine, migration path bandwidth is expressed as B;

Step4: if the dirty page generation rate of virutal machine memory that migration path bandwidth B is greater than 1.5 times, then do not meet transition condition, go to Step3; Otherwise, start its migration task;

Step5: virtual machine completes migration, upgrades the migration gain of the virtual machine be attached thereto, upgrades the available resources of source physical host, upgrades transportable queue and waits for resource queue, upgrades DCN network topology matrix available bandwidth; Go to Step3;

Step6: upgrading object physical host can resource, upgrade transportable and etc. queue to be migrated, upgrade DCN network topology matrix available bandwidth; Go to Step3;

Step7: whether the queues to be migrated such as judgement are empty, and if it is empty, then algorithm terminates; Otherwise, by the virtual machine (vm) migration in physical machine minimum for virtual machine number to be moved out to free physical machine.