CN102625198A - Intelligent light resource configuration method and system - Google Patents

Abstract

The invention discloses an intelligent light resource configuration method and an intelligent light resource configuration system. The method comprises the following steps of: dividing an entire network to be searched into un-overlapped sub-networks; searching for a preset number of shortest paths between inter-network nodes in each sub-network respectively; and performing entire-network searching according to the preset number of shortest paths between the inter-network nodes in each sub-network and the inter-network paths to obtain a second preset number of entire-network shortest paths as alternative intelligent light resource configuration schemes. According to the scheme, the entire network is divided into the sub-networks for path searching, so that computational efficiency can be improved; and the second preset number of finally obtained entire-network shortest paths serve as the alternative intelligent light resource configuration schemes which can be selected by a resource management expert to determine a final scheme, so that the final intelligent light source configuration scheme is accurate, and an optimal intelligent light source configuration scheme can be obtained.

Description

Light resources intelligence collocation method and system

Technical field

The present invention relates to communication technical field, particularly a kind of light resources intelligence collocation method, a kind of light resources intelligent configuration system.

Background technology

In general, the light resources of telecommunications is meant the essential communication resources of composition optical communication network such as OLT (optical line terminal, optical fiber cable termination) equipment, ONU (Optical Network Unit, optical node) equipment, optical fiber, optical cable, light joint.Along with the increase of user's access demand, rate requirement is also increasingly high, and therefore, the development optical communication could be satisfied the demand of growing access rate.

In the development of optical communication; The configuration of light resources intelligence is a routine work of common carrier; The configuration of opening, all related in the process such as engineering construction transformation the light path resource in customer service with reconfigure; It is unpractical depending merely on the path that manual work goes to seek between the Origin And Destination, even if be unit to adopt the office direction light path after the engineering association, data volume is also very huge.

When light resources is configured, require usually to accomplish to reduce resources costs as much as possible, O&M efficient also will be lacked, improved to the response time that simultaneity factor is handled.Correspondingly, for the configuration of light resources, also be to adopt multiple tolerance usually, for example: device resource cost, circuit construction cost, delay, stability or the like.When in the existing scheme light resources being configured, general way is that variant metric parameter is single metric parameter according to different weight conversions, thereby realizes the configuration of light resources in view of the above; And in order to ensure the resources effective utilization; Also can regard the shortest path of searching from origin-to-destination as, but in practical operation, because the data volume of telecommunication resources is very big; And need carry out correspondingly in real time, therefore need between resource utilization and response speed, do balance.

Because in the telecommunication technology solution; The resource data amount of telecommunication administration is huge; Differentiated control is adopted in traditional resource management, therefore, and the resource searching scheme of existing inquiry shortest path; Be to search the path of from the user node to the office, standing at Access Layer earlier, again from the path between convergence-level, the core layer office of searching station.Though sectioning search has improved the response of support system, also increased manually-operated complexity, and this sectioning search method neither the optimum scheme of the whole network resource utilization.In addition; Because the development of optical-fiber network has increased plurality of access modes; For example: FTTB, Fiber to the home, many, the outdoor nodes of outdoor equipment node possibly be connected to the station a plurality of innings; Adopt the method at the traditional office of being integrated into station can't search for the light path that is connected to other innings station automatically, need manual work to specify.Thereby for optical-fiber network, traditional multi-zone supervision mode also is not suitable for actual application.

Summary of the invention

To the problem that exists in the above-mentioned prior art; The object of the present invention is to provide a kind of light resources intelligence collocation method, a kind of light resources intelligent configuration system; It can improve the efficient of light resources intelligence configuration, and can search out optimum light resources intelligence allocation plan as far as possible accurately.

For achieving the above object, the present invention adopts following technical scheme:

A kind of light resources intelligence collocation method comprises step:

The whole network to be searched is divided into nonoverlapping subnet;

Search for the first preset number bar shortest path between the gateway node in each subnet respectively;

Carry out the whole network search according to path between the first preset number bar shortest path between the gateway node in each subnet, net and obtain second preset number bar the whole network shortest path as light resources intelligence allocation plan to be selected.

A kind of light resources intelligent configuration system comprises:

Cutting unit is used for the whole network to be searched is divided into nonoverlapping subnet;

The subnet search unit is used for searching for respectively the first preset number bar shortest path between the gateway node in each subnet;

The whole network route searching unit is used for carrying out the whole network search according to path between the first preset number bar shortest path between the gateway node in each subnet, net and obtains second preset number bar the whole network shortest path as light resources intelligence allocation plan to be selected.

According to the invention described above scheme; It is that the whole network to be searched is divided into a plurality of subnets; On the basis of these a plurality of subnets, calculate the first preset number bar shortest path between all gateway nodes in each subnet then respectively, and carry out the whole network search acquisition second preset number bar the whole network shortest path as light resources intelligence allocation plan to be selected, after being divided into a plurality of subnets according to path between the first corresponding preset number bar shortest path of this each subnet, net; Can parallel computation to the path between the gateway node in each subnet; Can utilize modes such as multinuclear, multiprocessor or multiple servers to carry out distributed parallel processing, raise the efficiency, and after network cut apart; The node number can reduce by 2 to 3 one magnitude usually, and then can further improve computational efficiency.Moreover; What carry out at last that the whole network search obtains is that second preset number bar the whole network shortest path is as light resources intelligence allocation plan to be selected; Can come from these light resources intelligence allocation plans to be selected, to select to confirm final scheme by the resource management expert; Thereby can make final light resources intelligence allocation plan more accurate, obtain optimum light resources intelligence allocation plan.

Description of drawings

Fig. 1 is the schematic flow sheet of light resources intelligence collocation method embodiment of the present invention;

Fig. 2 is the schematic flow sheet that the whole network to be searched is divided into nonoverlapping subnet embodiment;

Fig. 3 is the schematic flow sheet that the whole network is carried out the embodiment of multistage roughened;

Fig. 4 carries out the schematic flow sheet that collection of illustrative plates is cut apart to the whole network figure after the afterbody roughened in the multistage roughened;

Fig. 5 is the schematic flow sheet of the embodiment of cluster segmentation;

Fig. 6 is the structural representation of light resources intelligent configuration system embodiment of the present invention.

Embodiment

Preferred embodiment below in conjunction with is wherein set forth the present invention program in detail.

The schematic flow sheet of light resources intelligence collocation method embodiment of the present invention has been shown among Fig. 1.As shown in Figure 1, the light resources intelligence collocation method among this embodiment of the present invention comprises step:

Step S101: the whole network to be searched is divided into nonoverlapping subnet, gets into step S102;

Step S102: search for the first preset number bar shortest path between the gateway node in each subnet respectively, get into step S103;

Step S103: carry out the whole network search according to path between the first preset number bar shortest path between the gateway node in each subnet, net and obtain second preset number bar the whole network shortest path as light resources intelligence allocation plan to be selected.

According to aforesaid the present invention program; It is that the whole network to be searched is divided into a plurality of subnets; On the basis of these a plurality of subnets, calculate the first preset number bar shortest path between all gateway nodes in each subnet then respectively, and carry out the whole network search acquisition second preset number bar the whole network shortest path as light resources intelligence allocation plan to be selected, after being divided into a plurality of subnets according to path between the first corresponding preset number bar shortest path of this each subnet, net; Can parallel computation to the path between the gateway node in each subnet; Can utilize modes such as multinuclear, multiprocessor or multiple servers to carry out distributed parallel processing, raise the efficiency, and after network cut apart; The node number can reduce by 2 to 3 one magnitude usually, and then can further improve computational efficiency.Moreover; What carry out at last that the whole network search obtains is that second preset number bar the whole network shortest path is as light resources intelligence allocation plan to be selected; Can come from these light resources intelligence allocation plans to be selected, to select to confirm final scheme by the resource management expert; Thereby can make final light resources intelligence allocation plan more accurate, obtain optimum light resources intelligence allocation plan.

Below in conjunction with one of them specific embodiment, the light resources intelligence allocation plan of the invention described above is set forth in detail.

The present invention program for the whole network to be searched, at first is divided into nonoverlapping subnet with it when implementing.For ease of understanding, below earlier figure is cut apart the explanation that makes an explanation.

Suppose to have the heavy figure G of cum rights (V, E), the fixed-point set of V presentation graphs wherein, E representes the set on limit, (i, j) there is one in expression from v _iTo v _jThe limit, so, to cut apart be exactly to seek a kind of segmentation strategy to figure, to satisfy V=V ₁∪ V ₂∪ ... ∪ V _p, with E _CutThe limit removed when cutting apart of expression, so, figure is cut apart to seek makes sum (E a kind of cutting apart _Cut) value minimum, make simultaneously | V _i| equate that as far as possible seeking optimum algorithm is NP-hard.

The most classical algorithm that the figure of no coordinate is cut apart is the Kernighan/Lin algorithm, and it is the node through continuous exchange subgraph, and the effect that makes collection of illustrative plates cut apart is more excellent.If figure N is divided into A, B, N=A ∪ B, and | A|=|B|, the limit E of calculating A and B _Cut, seek subset X so, Y belongs to A, B, and | A|=|B|, so X, Y are exchanged, the exchange back is A ', B ', the limit E ' between A ' and the B ' _Cut, E ' _Cut＜E _CutIn the practical application, require the computational efficiency of evaluation function fast as far as possible, and in fact evaluation function calculating is very consuming time.

It is an important content that belongs to atlas analysis that figure is cut apart, in atlas analysis, will scheme G and use Laplce's matrix notation, and L (G)=D-A, wherein A is the adjacency matrix of cum rights, that is, and A _Ij=w (i, j), if (i, j)＞0, D is a diagonal matrix to w, D _Ij=0,

If vectorial X representes that a kind of two cut apart and are divided into S, S ', the X component is formed by 0,1,1 expression v _iBelong to S set, 0 expression v _iBelong to S set '; $X\·L\left(G\right)\·X^t=\underset{(i.j)\∈E}{\mathrm{\Σ}}w(i,j){(x_i-x_j)}^2,$ Can find out XL (G) X ^t=E _CutIf, v _iWith v _jThen corresponding of same set is 0; Theoretical proof λ ₂(the second little characteristic value of expression Laplce matrix) is 0, and then presentation graphs is not communicated with; The value of X is relaxed to real number by (0,1), and works as through finding the solution λ ₂Characteristic vector, what can obtain being similar to cuts apart, and it is higher to cut apart mass ratio.But, characteristic value to find the solution for the super large matrix be very slow, be not suitable for for the network topology of telecommunications.

Based on this; In the present invention program; When the whole network to be searched is divided into nonoverlapping subnet, can adopt collection of illustrative plates partitioning scheme based on collection of illustrative plates and cluster, the schematic flow sheet of the embodiment when the whole network to be searched is divided into nonoverlapping subnet has been shown among Fig. 2.Need to prove; In the explanation of following embodiment; Average in order to ensure each set sizes, can increase constraints, suppose to be divided into p sub-net SubG to network; Then set constraints

i=1...p, e is the parameter of control set sizes.

As shown in Figure 2, the concrete segmentation procedure that the whole network to be searched is divided into nonoverlapping subnet can comprise:

Step S201: the whole network to be searched is carried out multistage roughened, get into step S202;

Step S202: the whole network figure after the afterbody roughened in the said multistage roughened is carried out collection of illustrative plates cut apart, get into step S203;

Step S203: the figure after cutting apart according to the whole network figure after the roughened at different levels in the said multistage roughened and said collection of illustrative plates carries out cluster segmentation, obtains said nonoverlapping subnet.

In above-mentioned steps S201, when the whole network to be searched was carried out multistage roughened, concrete mode can be:

Figure G during for any one-level roughened ⁿ, choose figure G ⁿNode to (i, j),

A ⁿBe figure G ⁿAdjacency matrix;

Confirm new node, this new node comprises that node is to (i, node that j) is merged into and figure G ⁿIn be not chosen for the right node of node;

Whether the node number of the figure that judgement is confirmed by said new node less than the 3rd preset number threshold value, if, finish multistage roughened process, if not, carry out next stage roughened process to the figure that confirms by said new node.

Based on above-mentioned multistage roughening mode, illustrated among Fig. 3 the whole network to be searched has been carried out the schematic flow sheet among the embodiment of multistage roughened.As shown in Figure 3, the process of multistage roughened specifically can be:

The figure of note G for being made up of node, limit promptly is the pairing figure of the whole network to be searched;

In first time during iteration, note n=0, i.e. G=G ⁰

According to G ⁰Structure adjacency matrix A ⁰,

(all the other are 0 for i, j) ∈ E;

Choose G ⁰Node to (i, j),

Up to what can not choose, selected node is to (i j) does not have repetition, if do not choose node right this moment, then withdraws from, and returns the process of above-mentioned first time of roughening, the promptly above-mentioned G=G that makes again ⁰, according to G ⁰Structure adjacency matrix A ⁰Etc. process;

With node to (i j) merges into new node, simultaneously with all can not choose also as new node;

The adjustment weight, suppose with node to (i, j) merge into i ', with node to (k l) merges into j ', then has $A_{i^{\′}{j}^{\′}}^1=A_{\mathrm{Ik}}^0+A_{\mathrm{Il}}^0+A_{\mathrm{Jk}}^0+A_{\mathrm{Jl}}^0;$

Make n=n++, promptly n=n+1 makes n=1;

Judge

Corresponding figure G ¹The node number whether less than preset number (being above-mentioned the 3rd preset number threshold value), if not, the figure G of gained then ¹Be the whole network figure behind the multistage roughening, otherwise continue to carry out following process;

Choose G ¹Node to (i, j),

Up to what can not choose, selected node is to (i j) to there not being repetition, if do not choose node right this moment, then withdraws from again;

With node to (i j) merges into new node, simultaneously with all can not choose also as new node;

The adjustment weight, suppose with node to (i, j) merge into i ', with node to (k l) merges into j ', then has $A_{i^{\&prime;}{j}^{\&prime;}}^2=A_{\mathrm{<figure-callout\; id="1"\; label="Ik"\; filenames="HDA0000141989630000031.png,HDA0000141989630000041.png"\; state="\{\{state\}\}">Ik</figure-callout>}}^1+A_{\mathrm{<figure-callout\; id="1"\; label="Il"\; filenames="HDA0000141989630000031.png,HDA0000141989630000041.png"\; state="\{\{state\}\}">Il</figure-callout>}}^1+A_{\mathrm{<figure-callout\; id="1"\; label="Jk"\; filenames="HDA0000141989630000031.png,HDA0000141989630000041.png"\; state="\{\{state\}\}">Jk</figure-callout>}}^1+A_{\mathrm{<figure-callout\; id="1"\; label="Jl"\; filenames="HDA0000141989630000031.png,HDA0000141989630000041.png"\; state="\{\{state\}\}">Jl</figure-callout>}}^1;$

Make n=n++, promptly n=n+1 makes n=2;

Judge

Corresponding figure G ²The node number whether less than preset number (being above-mentioned the 3rd preset number threshold value), if not, the figure G of gained then ²Be the whole network figure behind the multistage roughening, otherwise continue to carry out G to figure ²Above-mentioned choose node to, node be combined be processes such as new node;

For any figure G ⁿ, choose figure G ⁿNode to (i, j),

Up to what can not choose, selected node is to (i j) does not have repetition, if do not choose node right this moment, then withdraws from again;

With node to (i j) merges into new node, simultaneously with all can not choose also as new node;

The adjustment weight, suppose with node to (i, j) merge into i ', with node to (k l) merges into j ', then has $A_{i^{\&prime;}{j}^{\&prime;}}^{n+1}=A_k^n+A_{\mathrm{Il}}^n+A_{\mathrm{Jk}}^n+A_{\mathrm{Jl}}^n;$

Make n=n++, i.e. n=n+1;

Continue said process, until obtaining the figure G of node number less than preset number (being above-mentioned the 3rd preset number threshold value) ^N+1, should be less than the figure G of preset number ^N+1Be the figure G that has realized behind the multistage roughening ^N+1

The preset number here can be carried out synthetic setting according to aspects such as Practical Calculation demand, operation times, and in the present invention program's a specific embodiment, above-mentioned preset number can be made as 3000.

Based on the schematic flow sheet shown in Fig. 3, the examples of program code in concrete implementation procedure can be described below:

graph_partition(G(V，E)，p，e，tmax)

The figure of //G for constituting by node, limit, p is for being divided into a few sub-graphs quantity, and e is the cluster size factor

//tmax is a maximum iteration time

The first step is carried out roughening, generates figure G ^N+1

According to G ⁰Structure adjacency matrix A ⁰, (all the other are 0 for i, j) ∈ E;

Do

Choose figure G ⁿNode to (i, j),

Up to what can not choose, selected (i is not j) to there being repetition again;

Were it not for and choose node right, then withdraw from;

Will (i j) is combined and is new node, simultaneously with all can not choose also as new node;

The adjustment weight; Node i; J merges into i '; Node k, l merges into j ', then

n++

While schemes G ^N+1The node number less than 3000

After accomplishing above-mentioned roughened process, can get among the above-mentioned steps S102 to the whole network figure G after the afterbody roughened in the multistage roughened ^N+1Carry out the process that collection of illustrative plates is cut apart, its concrete collection of illustrative plates cutting procedure can comprise:

To the whole network figure G after the afterbody roughened in the multistage roughened ^N+1Carry out collection of illustrative plates and cut apart, generate the S set ubG of p sub-net;

In the secondary splitting process of any one-level, obtain the figure tmpG and the corresponding matrix tmpA of divided least number of times among the S set ubG; Find the solution the second little characteristic value and the characteristic of correspondence vector X of tmpA; The X component is greater than or equal to 0 corresponding nodes puts into first set, will put into second set, and the limit is divided in first set, pairing two sub-graphs of second set, generate G less than 0 corresponding nodes _Sub', G _Sub";

Make SubG=(SubG ∪ { G _Sub', G _Sub" })-{ tmpG};

Whether the progression of judging secondary splitting is less than or equal to the first preset iterations, if, return the secondary splitting process that pair set SubG carries out next stage, if not, finish the secondary splitting process.

Based on aforesaid way, the whole network figure G in obtaining above-mentioned multistage roughened after the afterbody roughened ^N+1After, to this figure G ^N+1Carry out collection of illustrative plates and cut apart, generate the S set ubG of p sub-net, i.e. SubG={G _Sub, can carry out secondary splitting repeatedly to the S set ubG of this subnet then, the schematic flow sheet that antithetical phrase net collective SubG carries out secondary splitting repeatedly has been shown among Fig. 4, as shown in Figure 4, its detailed process can be:

Obtain the figure tmpG and the corresponding matrix tmpA of divided least number of times among the SubG;

Find the solution the second little characteristic value and the characteristic of correspondence vector X of tmpA;

The X component is greater than or equal to 0 corresponding nodes puts into a set (can be designated as first set), will less than 0 put into another set (can be designated as second set), and the limit is also divided in these two pairing two sub-graphs of set, generate two sub-graphs G _Sub ', G _Sub ";

Make SubG=(SubG ∪ { G _Sub', G _Sub" })-{ tmpG} after promptly the figure tmpG of divided least number of times removes among the subnet S set ubG, adds subnet S set ubG with two that generate new subgraphs from subnet S set ubG;

Make i=i++, i.e. i=i+1;

Judge whether i is less than or equal to preset iterations p; If return processes such as the figure tmpG that continue to carry out divided least number of times among the above-mentioned SubG of obtaining and corresponding matrix tmpA, until reaching preset iterations p; What need be careful is; The number of the subnet that the preset iterations p here is divided into when cutting apart with above-mentioned collection of illustrative plates is identical, and this is because above-mentioned collection of illustrative plates is cut apart the p sub-net that obtains; Be set to p and can guarantee as far as possible that this p sub-graphs all might accept secondary splitting through presetting iterations; Otherwise the current SubG that obtains is the S set ubG of the subgraph that obtains after collection of illustrative plates is cut apart, and adopts this current SubG that obtains to get into the process of follow-up cluster segmentation.

Based on the schematic flow sheet shown in Fig. 4, the examples of program code in concrete implementation procedure can be described below:

Because cluster segmentation is after roughened and collection of illustrative plates are cut apart, to carry out, therefore, also can be referred to as is the stage that refinement is cut apart.When carrying out cluster segmentation, can adopt various available clustering algorithms to cut apart, k-mean algorithm for example is with the calculated performance of utilizing this algorithm advantage with the scale linear growth.

Cluster is that a group is had manifold object, is divided into several types according to certain similarity, and the object in each type is more close; Difference between type is bigger, and this characteristic is similar with the figure ration of division, if a node is many with the related limit of some subnets; Just can think that it follows the node of this sub-net all more similar, can utilize this characteristic at same subnet; In the present invention program, can cut apart network with clustering algorithm.

When carrying out cluster segmentation; Need be applied to the whole network figure (the whole network figure of the roughened at different levels here of roughened at different levels in the above-mentioned multistage roughened; Do not comprise the whole network figure that obtains after the afterbody roughened) and the figure of above-mentioned collection of illustrative plates after cutting apart; When using the whole network figure of roughened at different levels; Can use all the whole network figure of roughened at different levels, be in the purpose that reduces amount of calculation, improves computational efficiency, can only choose the process that a part among the whole network figure after these roughened at different levels is carried out cluster segmentation.Carry out at selected part under the situation of cluster segmentation, the process of cluster segmentation specifically can be:

In said multistage roughened, among the whole network figure of roughened at different levels, choose the 4th a preset number figure;

Cluster centre of each figure after calculating said collection of illustrative plates and cutting apart;

In any one-level iterative process; To any any node that the figure corresponding nodes is concentrated among said the 4th preset number figure; Judge that this node which in each said cluster centre is nearest; And when the size of this nearest cluster centre place cluster is being preset in the magnitude range, this node is added the cluster at this nearest cluster centre place;

Recomputate cluster centre; And calculating is when the error of previous stage cluster centre and upper level cluster centre; Said the 4th a preset number figure iteration is not finished and error in preset error range the time; Next one figure among said the 4th preset number figure gets into the next stage iterative process, otherwise constructs said nonoverlapping subnet according to each cluster that obtains when the previous stage iterative process.

Cluster segmentation mode according to as implied above has illustrated the schematic flow sheet that carries out the embodiment of cluster segmentation among Fig. 5.In this embodiment, suppose from multistage roughened, to choose 10 figure among the whole network figure after the roughened at different levels, as shown in Figure 5, the process of cluster segmentation can be:

In above-mentioned multistage roughened, among the whole network figure of roughened at different levels, choose the 4th a preset number figure, from consideration to arithmetic speed and computational efficiency; The 4th preset number here can be provided with according to actual needs, with the needs that satisfy the refinement level simultaneously and the demand of arithmetic speed, therein in concrete example; The 4th preset number here can be chosen to be 10; When selecting, can adopt various possible modes to select, for example can average and choose by step-length; Consider and to be divided exactly by 10; During actual choosing, can choose with the mode of approximate average step length, to take all factors into consideration the roughened result in each stage in the multistage roughening process;

Calculate among the SubG after above-mentioned collection of illustrative plates is cut apart each figure cluster centre;

Then; To the 1st figure among 10 selected figure; To any node of concentrating for the 1st figure corresponding nodes, judge that this node which in each cluster centre is nearest, and the size of judging this cluster centre place cluster or set is whether in preset magnitude range;

If in preset magnitude range, then do not return this node judged again, to seek the nearer cluster centre of another one except above-mentioned cluster centre;

If in preset magnitude range, then this node is added the cluster or the set at this nearest cluster centre place;

After the 1st each corresponding node of figure be finished; After promptly each node among the 1st figure has all added the cluster or set at certain cluster centre place; Recomputate the cluster centre of this cluster or polymerization, and calculate the error of this cluster centre and cluster centre before;

If above-mentioned 10 are schemed not that the cluster iteration finishes and error in preset error range the time, carry out said process to the 2nd selected figure, that is:

In any one-level iterative process; To the concentrated any node of certain figure corresponding nodes from these 10 figure; Judge that this node which in each current cluster centre is nearest; And when the size of this nearest cluster centre place cluster is being preset in the magnitude range, this node is added the cluster at this nearest cluster centre place;

Recomputate cluster centre; And calculating is when the error of previous stage cluster centre and upper level cluster centre; Said second a preset number figure iteration is not finished and error in preset error range the time; Next one figure to not carrying out the cluster iterative process among these 10 figure as yet gets into the next stage iterative process, otherwise constructs said nonoverlapping subnet according to each cluster that obtains when the previous stage iterative process.

Based on the schematic flow sheet shown in Fig. 5, the examples of program code in concrete implementation procedure can be described below:

Order $c_i^0=\left({\mathrm{SubG}}_i\right),$ I=1,2,3 ..., p;

At G ⁰, G ¹... G ⁿIn, choose 10 figure, be assumed to be tmpG, its adjacency matrix is tmpA, set of node is tmpV;

According to Construct subgraph SubG _i

$V\left(\mathrm{Sub}{G}_i\right)=c_i^0,E\left({\mathrm{SubG}}_i\right)=\left\{(v_i,v_j)\right|v_i\&Element;c_i^0,v_j\&Element;c_i^0,(v_i,v_j)\&Element;E\},i=\mathrm{1,2,3}...p;$

Can obtain through the subgraph SubG after the cluster segmentation refinement through above-mentioned steps _i

At the subgraph SubG of cutting apart through above-mentioned collection of illustrative plates after obtaining the cluster segmentation refinement _iAfter, search each the subgraph SubG after these refinements more respectively _iMany shortest paths between all interior gateway nodes.

Searching each subgraph SubG respectively _iIn all gateway nodes between many shortest paths the time, the present invention program adopts the thought of Dynamic Programming, converts into and finds the solution a problem that scale is littler finding the solution a problem, up to end, problem can directly be solved.Through Dynamic Programming, problem can be resolved through the mode of iteration.

Suppose that we finally need find out k bar (being equivalent to the second preset number bar) shortest path; So, in each subnet, we can search u*k bar (being equivalent to the first preset number bar) path; Because u too senior general increases memory space; Therefore the concrete value of u can be set according to actual conditions, and in specific embodiment, the value of u can be 1＜u＜1.3 therein.

Can be described below through searching the mathematical principle that u*k bar path finds out k bar shortest path: supposing that n is a subnet node number, is the shortest path P of i if require the limit number _St(i), min (P _Sr(i-1)+and w (r, t)), r=1,2,3 ..., n.If P representes gateway node and arrives the distance of all nodes that A representes the adjacency matrix of cum rights, P ' _St=min (P _Sr+ A (r, t)), r=1,2,3 ..., n with the continuous iteration of P, will restrain during iteration to a certain step, i.e. P '=P.

Based on above-mentioned mathematical principle, in the present invention program, can adopt various possible modes to find out the k*u bar shortest path of gateway node in each subnet.Therein among embodiment; Can it be generalized to the shortest algorithm of k,

path of expression from s to r, m=1; 2; 3...k, when in the present invention program, using, can follow following rule:

Only search the mulitpath between the gateway node;

Only handle, promptly newly adding paths $P_{\mathrm{St}}^m.\mathrm{Status}=\mathrm{Lastmew};$

In algorithm, judge and the loop do not occur;

If $P_{\mathrm{Sr}}^m+A(r,t)<d,d=\mathrm{Max}\left(P_{\mathrm{St}}^m\right),$ M=1,2,3 ..., there is the new route from s to t in k;

Then explain and found all k shortest paths when can not find new path

; Need not carry out iteration again, but premature termination.

Cut apart through collection of illustrative plates the whole network to be searched is divided into subnet after; Because node number sparse, subnet compares less between intensive in the subnet, net; And the gateway node of subnet is than great; And therefore the limit can adopt based on the Dynamic Programming iterative algorithm and find out many shortest paths usually than comparatively dense.

When adopting the Dynamic Programming iterative algorithm to find out many shortest paths, concrete step can comprise:

After confirming a paths, search whether there is a shorter path, if exist, behind the renewal path status, judge whether to exist new route, if new route is arranged, continue to search whether have shorter path, until can not find shorter path.

Based on above-mentioned four, in a specific embodiment, the program implementation of searching many shortest paths between the gateway node in the subnet can be described below:

subnet_search(SubG，G(V，E，w)，k，u)

//SubG is certain sub-graphs after the whole network is cut apart, and G (V E) is the figure of the whole network, and p, k are the path to be selected that system promises to undertake, u is the Redundant Control factor, k ' expression redundant path quantity, and k '=k*u, p represent subgraph quantity, V _BorderExpression subgraph SubG _iBoundary point;

//status:lastnew representes last iteration new node, and new representes the path that increases newly to be defaulted as old;

The m paths of expression from s to t, $P_{\mathrm{Sr}}^m.\mathrm{Length}=\&infin;,$

$P_{\mathrm{St}}^m.\mathrm{Status}=\mathrm{Old},$ A representes the adjacency matrix of SubG, A _St=w (s, t), the length of expression from node s to node t.

The boundary point s that initialization is all, t, $P_{\mathrm{St}}^m.\mathrm{Length}=0,$ $P_{\mathrm{St}}^m.\mathrm{Status}=\mathrm{Lastnew},$

Above-mentioned

that obtains at last is exactly the SubG path; Wherein, m=1,2,3; ...; K ', s, t are boundary point.

Above-mentioned obtain the SubG path of the gateway node in each subnet after, can carry out the whole network search according to path between this SubG path, net then and obtain K bar shortest path as light resources intelligence allocation plan to be selected.

When carrying out the whole network route searching, be based on the path of the gateway node in above-mentioned each subnet that has obtained, search for the whole network path of the gateway node between each subnet.During concrete the realization, can carry out the whole network route searching based on the Dijstra algorithm.The principle of dijkstra's algorithm is: if P _StBe the shortest path from s to t, so for other node r on the path, P _SrIt also is the shortest path from s to r.

In the present invention program's a specific embodiment, can be on the basis of dijkstra's algorithm, the search simple extension in the whole network path has been arrived the k shortest path, specifically can follow following rule:

At first, the topology of the whole network search is made up of three parts: circuit between net; The path of the gateway node in each subnet of the aforementioned calculation of the gateway node of subnet, system's storage; The topology of the subnet at starting point, terminal point place;

Secondly; If

is the k bar shortest path from s to t; R is certain node that removes starting point, terminating point on the path;

also is the k ' bar shortest path from s to r so; K '≤k; Traditional dijkstra's algorithm is only to keep a paths to each node among the figure, and in the scheme after the present invention's expansion, needs to keep the k paths;

Its three, the loop does not appear in the path that requires to search;

Its four, for the k shortest path, if the limit of definite object node has been traveled through k time, then can premature termination search.

Comprehensive foregoing, in a concrete embodiment, its corresponding software realizes that program can be described below:

global_search(s，t，k，SubG，G(V，E，w)，Path _SubG)

//s, t are respectively starting point, terminal point, and SubGi is the i subgraph, and (V E) is the figure of the whole network to G, and (u v) representes the length from u to v to w

//Path _SubGPath between the gateway node in the expression subnet, path (u v) representes limit or subnet path from u to v, if the subnet path, then corresponding many limits.

The figure G of structure the whole network _Global, n is the node number, SubG _sBe starting point place subnet, SubG _tBe terminal point place subnet, V _EtBe the node that points to terminal point eventually, V _Et={ i| (i, t) ∈ E (G _Global);

The number of times visit that the limit of record sensing terminal point is visited _Et=0;

All paths of initialization at first;

i=1; 2; 3; ..., n; J=1,2,3 ..., k;

The path of initialization starting point;

j=1,2,3; ..., k;

Starting point is put into pending pile structure tmpV.v=s; TmpV.path=" "; TmpV.length=0; Heap.put (tmpV);

Resulting is exactly to treat routing path; Wherein,

j=1; 2; 3 ..., k.

Based on the mode in the above-mentioned concrete example; Can obtain between k bar (the second preset number bar) net the path as resource allocation proposal to be selected; Can come from these light resources intelligence allocation plans to be selected, to select to confirm final scheme by the resource management expert; Thereby can make final light resources intelligence allocation plan more accurate, obtain optimum light resources intelligence allocation plan.

Yet, because communication network is very big, and need constantly carry out engineering upgrading and transformation, new resource data can be entered into system at any time, causes the network topology of the whole network constantly to change, thereby has influence on the search to the whole network.In view of the above, in the present invention program, give of the variation of relevant resource change strategy with the reply network topology.

The change of resource generally includes following several types: the line out of service of the equipment failure of subnet, the increase of the circuit of subnet, subnet, the equipment increase of the whole network, equipment failure of the whole network or the like.

To this, in the present invention program, the resource change processing policy of being made can be to be described below:

Exist under the situation of line resource change when detecting server idle running and subnet, carry out the process of many shortest paths between all gateway nodes of searching for again in the subnet immediately;

When circuit change quantity reaches certain quantity in detecting the busy and subnet of server, search for many shortest paths between all gateway nodes in the subnet again;

Circuit that the whole network increases is a lot of and when reaching some when detecting, carry out again that collection of illustrative plates is cut apart and the subnet of counterweight new search collection of illustrative plates after cutting apart in all gateway nodes between many shortest paths;

Increased node when detecting the whole network, and starting point, terminal point are also unallocated during to certain sub-net, this moment can be on the basis of original division, and the mode through cluster is assigned to certain sub-net with this node that increases newly;

If the subnet equipment failure, and be gateway node, then the path of respective stored also is set to lose efficacy;

When the line out of service of subnet, then the path of respective stored also is set to lose efficacy.

Based on aforesaid light resources of the present invention intelligence collocation method, below describe with regard to one of them concrete application case.

In this concrete application case, to choose certain and economize inside the province the network of a maximum branch company at present, nearly 70,000 nodes of this network have the circuit about 100,000.

At first network is carried out dividing processing; Choose and be divided into the p=128 sub-net; The subnet size is controlled to be

e and gets 1.5, is segmented in 1 second to accomplish, and cuts apart the maximum subnet node several more than 700 in back; Minimum subnet node is several more than 400, and splitting speed is than very fast.Actual test through to this branch company's network finds, is no more than under 1000 the situation the cutting apart all in 15 seconds of full figure at the subnet number of cutting apart.

On the above-mentioned basis of cutting apart, increased a node newly, adopt k-mean to carry out cluster, because original division is relatively good, thereby can restrain in second at 1-2.

Set resource distribution path k=200 to be selected, redundancy factor u=1.1 is if only need upgrade each subnet; All in 2 seconds, in this real case, adopt the CPU multithreading of 4 nuclears to carry out the operation time of each subnet; Be evenly distributed workload, each nuclear distributes 32 sub-net parallel computations, because the subnet size is average relatively; Basically simultaneously accomplish, the time is within 20 seconds altogether.

The sum of all-ones subnet gateway node is at 3600; Add subnet initial, terminal node; The node sum has dwindled 10 times than original figure in 5000, inquiry k shortest path only needs the 2-3 time of second; Confirmed that the present invention program can improve the efficient of light resources intelligence configuration, and can search out the conclusion of optimum light resources intelligence allocation plan as far as possible accurately.

The invention described above scheme; Also have no similar technology and system to release in the telecommunications industry; Belong to and propose first and realize; It has proposed a kind of distributed, parallel light resources search plan, through cutting apart based on collection of illustrative plates and the algorithm of cluster is divided into a plurality of subnets with the whole network fast, and every pair of gateway node path in can the parallel computation subnet on subnet; On the basis of subnet path, search the treat routing path of the k bar path of the whole network during user inquiring as resource distribution; When network change and system when busy, the present invention program also proposes to adopt interim partitioning algorithm, will increase resource node on the basis rapidly newly and join the subnet of having cut apart in existing cutting apart, and improves response speed.Possess than high scalability, can be when performance is not enough simply through increasing the expansion of CPU or server.

Light resources intelligence collocation method according to the invention described above; The present invention also provides a kind of light resources intelligent configuration system; The structural representation of light resources intelligent configuration system embodiment of the present invention has been shown among Fig. 6; As shown in Figure 6, in this embodiment, light resources intelligent configuration system of the present invention comprises:

Cutting unit 601 is used for the whole network to be searched is divided into nonoverlapping subnet;

Subnet search unit 602 is used for searching for respectively the first preset number bar shortest path between the gateway node in each subnet;

The whole network route searching unit 603 is used for carrying out the whole network search according to path between the first preset number bar shortest path between the gateway node in each subnet, net and obtains second preset number bar the whole network shortest path as light resources intelligence allocation plan to be selected.

Wherein, in a specific embodiment, above-mentioned cutting unit 601 specifically can comprise:

Roughened unit 6011 is used for the whole network to be searched is carried out multistage roughened;

Collection of illustrative plates cutting unit 6012 is used for that the whole network figure after the said multistage roughened afterbody roughened is carried out collection of illustrative plates and cuts apart;

Cluster segmentation unit 6013, the figure after being used for cutting apart according to the whole network figure after the said multistage roughened roughened at different levels and said collection of illustrative plates carries out cluster segmentation, obtains said nonoverlapping subnet.

In the another one preferred embodiment, above-mentioned subnet search unit 602 can include more than two.Thereby, can utilize these a plurality of subnet search units 602 to carry out parallel computation to the path between the gateway node in the variant subnet, realize distributed treatment, improve treatment effeciency.Wherein, the subnet search unit 602 here can be realized through modes such as multinuclear, multiprocessor or multiple servers, does not repeat them here.

In addition, as shown in Figure 6, in the another one specific embodiment, can also include detecting unit 604, be used for the detection system state;

At this moment; Above-mentioned subnet search unit 602; Also be used for detecting unit 604 detect server idle running and subnet the line resource change, or detect in the busy and subnet of server circuit change quantity and reach when setting amount threshold, carry out the process of the first preset number bar shortest path between the above-mentioned gateway node of searching for respectively in each subnet again;

Above-mentioned cutting unit 601 also is used for reaching when setting amount threshold in the number of, lines that detecting unit 604 detects the whole network and increases, and carries out the said process that the whole network to be searched is divided into nonoverlapping subnet again;

Above-mentioned cluster segmentation unit 6013, also being used for detecting the whole network at detecting unit 604, to have increased new node and starting point, terminal point still unallocated during to certain sub-net, through cluster the node that this increases newly is assigned to subnet.

Wherein, in the light resources intelligent configuration system of the invention described above, the concrete implementation of each unit can with the light resources of the invention described above intelligence collocation method in identical, do not repeat them here.

The above embodiment has only expressed several kinds of execution modes of the present invention, and it describes comparatively concrete and detailed, but can not therefore be interpreted as the restriction to claim of the present invention.Should be pointed out that for the person of ordinary skill of the art under the prerequisite that does not break away from the present invention's design, can also make some distortion and improvement, these all belong to protection scope of the present invention.Therefore, the protection range of patent of the present invention should be as the criterion with accompanying claims.

Claims (10)

1. a light resources intelligence collocation method is characterized in that, comprises step:

The whole network to be searched is divided into nonoverlapping subnet;

Search for the first preset number bar shortest path between the gateway node in each subnet respectively;

Carry out the whole network search according to path between the first preset number bar shortest path between the gateway node in each subnet, net and obtain second preset number bar the whole network shortest path as light resources intelligence allocation plan to be selected.

2. light resources intelligence collocation method according to claim 1 is characterized in that the step that the whole network to be searched is divided into nonoverlapping subnet comprises:

The whole network to be searched is carried out multistage roughened;

The whole network figure after the afterbody roughened in the said multistage roughened is carried out collection of illustrative plates to be cut apart;

Figure after cutting apart according to the whole network figure after the roughened at different levels in the said multistage roughened and said collection of illustrative plates carries out cluster segmentation, obtains said nonoverlapping subnet.

3. light resources intelligence collocation method according to claim 2 is characterized in that the process of the whole network to be searched being carried out multistage roughened comprises:

Figure G during for any one-level roughened ⁿ, choose figure G ⁿNode to (i, j),

A ⁿBe figure G ⁿAdjacency matrix;

Confirm new node, this new node comprises that node is to (i, node that j) is merged into and figure G ⁿIn be not chosen for the right node of node;

Whether the node number of the figure that judgement is confirmed by said new node less than the 3rd preset number threshold value, if, finish multistage roughened process, if not, carry out next stage roughened process to the figure that confirms by said new node.

4. light resources according to claim 3 intelligence collocation method is characterized in that, the whole network figure after the afterbody roughened in the multistage roughened is carried out the step that collection of illustrative plates cuts apart comprise:

To the whole network figure G after the afterbody roughened in the multistage roughened ^N+1Carry out collection of illustrative plates and cut apart, generate the S set ubG of P sub-net;

In the secondary splitting process of any one-level, obtain the figure tmpG and the corresponding matrix tmpA of divided least number of times among the S set ubG; Find the solution the second little characteristic value and the characteristic of correspondence vector X of tmpA; The X component is greater than or equal to 0 corresponding nodes puts into first set, will put into second set, and the limit is divided in first set, pairing two sub-graphs of second set, generate G less than 0 corresponding nodes _Sub', G _Sub";

Make SubG=(SubG ∪ { Gsub ', Gsub " })-{ tmpG};

Whether the progression of judging secondary splitting is less than or equal to the first preset iterations, if, return the secondary splitting process that pair set SubG carries out next stage, if not, finish the secondary splitting process.

5. light resources intelligence collocation method according to claim 4; It is characterized in that the step that the figure after cutting apart according to the whole network figure of roughened at different levels in the said multistage roughened and said collection of illustrative plates carries out cluster segmentation, obtain said nonoverlapping subnet comprises:

In said multistage roughened, among the whole network figure of roughened at different levels, choose the 4th a preset number figure;

Each figure after calculating said collection of illustrative plates and cutting apart cluster centre;

In any one-level iterative process; To any any node that the figure corresponding nodes is concentrated among said the 4th preset number figure; Judge that this node which in each said cluster centre is nearest; And when the size of this nearest cluster centre place cluster is being preset the situation in the magnitude range, this node is added the cluster at this nearest cluster centre place;

Recomputate cluster centre; And calculating is when the error of previous stage cluster centre and upper level cluster centre; Said the 4th a preset number figure iteration is not finished and error in preset error range the time; Next one figure among said the 4th preset number figure gets into the next stage iterative process, otherwise constructs said nonoverlapping subnet according to each cluster that obtains when the previous stage iterative process.

6. according to any described light resources intelligence collocation method of claim 1 to 5, it is characterized in that, also comprise any one or combination in any in following each item:

When the line resource change that detects server idle running and subnet, carry out the process of the first preset number bar shortest path between the said gateway node of searching for respectively in each subnet again;

Circuit change quantity reaches when setting amount threshold in detecting the busy and subnet of server, carries out the process of the first preset number bar shortest path between the said gateway node of searching for respectively in each subnet again;

When the number of, lines of the whole network increase reaches the setting amount threshold, carry out the said process that the whole network to be searched is divided into nonoverlapping subnet again;

It is still unallocated during to certain sub-net to have increased new node and starting point, terminal point at the whole network, through cluster the node that this increases newly is assigned to subnet.

7. a light resources intelligent configuration system is characterized in that, comprising:

Cutting unit is used for the whole network to be searched is divided into nonoverlapping subnet;

The subnet search unit is used for searching for respectively the first preset number bar shortest path between the gateway node in each subnet;

The whole network route searching unit is used for carrying out the whole network search according to path between the first preset number bar shortest path between the gateway node in each subnet, net and obtains second preset number bar the whole network shortest path as light resources intelligence allocation plan to be selected.

8. light resources intelligent configuration system according to claim 7 is characterized in that, said cutting unit comprises:

The roughened unit is used for the whole network to be searched is carried out multistage roughened;

The collection of illustrative plates cutting unit is used for that the whole network figure after the said multistage roughened afterbody roughened is carried out collection of illustrative plates and cuts apart;

The cluster segmentation unit, the figure after being used for cutting apart according to the whole network figure after the said multistage roughened roughened at different levels and said collection of illustrative plates carries out cluster segmentation, obtains said nonoverlapping subnet.

9. light resources intelligent configuration system according to claim 8 is characterized in that the subnet search unit of telling comprises more than two.

10. according to any described light resources intelligent configuration system of claim 7 to 9, it is characterized in that also comprise: detecting unit is used for the detection system state;

Said subnet search unit; Also be used for said detection to the line resource change of server idle running and subnet, or detect in the busy and subnet of server circuit change quantity and reach when setting amount threshold, carry out the process of the first preset number bar shortest path between the said gateway node of searching for respectively in each subnet again;

Said cutting unit also is used in said detection the number of, lines that you detect the whole network and increase and reaches when setting amount threshold, carries out the said process that the whole network to be searched is divided into nonoverlapping subnet again;

Said cluster segmentation unit, it is still unallocated during to certain sub-net also to be used for having increased new node and starting point, terminal point in said detection to the whole network, through cluster the node that this increases newly is assigned to subnet.

Images