The content of the invention
The technical problems to be solved by the invention are by tradition for above shortcomings in the prior art there is provided one kind
Probe technique in IP network is incorporated into SDN, and disposing probe as few as possible realizes the state acquisition to SDN,
Then network state is analyzed using Bayesian network, realizes the failure section of the SDN of the fault detect of SDN
Point detection method.
A kind of malfunctioning node detection method of SDN, including step:
Step 1:The part node sets the probe in a network, and passes through the node belonging to the probe
State set obtains the sparse matrix of the state of the node of the probe paths traversed;
Step 2:Calculate factor of influence of the probe to the node;
Step 3:According to every probe to described in belonging to the result of detection and the probe of the state of the node
The path of node, calculates the probability of malfunction of the node;
Step 4:The condition when probability of malfunction for finding out the node reaches maximum;
Step 5:The node set that all probability reach maximum is exported, obtained malfunctioning node collection is as calculated
Close.
In step 2, the calculation formula of the factor of influence of the node belonging to the probe is:
IF(v0)=α Fin (v0)+βFout(v0)
Wherein:V is some specific node in network, and Fin (v0) is the internal influence factor of the node, and α is inside
Influence coefficient;Fout (v0) is the external action factor of the node, and β is external action coefficient, and Sn is what the node influenceed
Business number, the numerical value that Li is quantified by the node to the influence degree of some business, K (v) is the interior subordinate of the node
Property value.
In step 3, the calculation formula of the probability of malfunction of the node is:
P(V1,V2,...Vi... Vn,T1,T2,...Tj,...Tm)=
P(T1|Pa(T1))P(T2|Pa(T2))...P(Tm|Pa(Tm))P(V1)P(V2)...P(Vn)
Wherein:P (V1, V2 ... Vi ... Vn, T1, T2 ... Tj ... Tm) be all nodes conditional probability,
Characterize the influence that the failure of the node is brought to whole network;
(Pa (Tj) represents all links that the probe Tj passes through, when having node failure in the probe Tj, p (Tj=
1 | Pa (Tj))=1, p (Tj=0 | Pa (Tj))=0, when the probe Tj is trouble-free, p (Tj=1 | Pa (Tj))=0, p
(Tj=0 | Pa (Tj))=1;Tj=1 represents normal, and Tj=0 represents failure;
Vi is node described in i-th, and Tj is probe described in j-th strip, and n is the node number, and m is the probe bar number (m
< n).
After calculating the factor of influence of the node belonging to the probe, before the probability of malfunction of the calculating node
Also include step:According to the sparse matrix, the complete of the state that includes all nodes is obtained using maximal margin matrix algorithm
Complete matrix.
In summary, the present invention has the following advantages that compared with prior art:
The present invention disposes probe using Bayesian network in SDN and realized to malfunctioning node detection, can effectively solve
In current SDN during failure node, due to the huge structure of network, it is difficult to find its trouble point, and trouble point is determined
The problem of position comes out.
The present invention carries out the fault-finding of SDN using Bayesian network:Realized by disposing probe in a network
The collection of node state, while it is general that status data is inputted into the node failure that Bayesian network calculates under traffic failure state
Rate, instant posterior probability is obtained using prior probability * factors of influence, realizes the prediction to malfunctioning node in SDN and fixed
Position.
Embodiment
To make those skilled in the art more fully understand technical scheme, below in conjunction with the accompanying drawings and specific embodiment party
Formula is described in further detail to the malfunctioning node detection method and its malfunctioning node detection system of SDN of the present invention.
A kind of diagnostic method of malfunctioning node in SDN, the probe technique in traditional IP network is incorporated into SDN nets by it
In network, disposing probe as few as possible realizes the state acquisition to SDN, then using Bayesian network to network state
Analyzed, realize the fault detect of SDN.
Below in conjunction with the flow chart of the malfunctioning node location algorithm based on probe shown in Fig. 1, illustrate allowed for influencing factors
SDN malfunctioning node detection method:
Step 1:Probe is set in node, and obtains state detection result of the probe to node.
In this step, in order to reduce the load of network, a small amount of probe is sent into SDN first, actually should
In, probe can be one section of program being deployed on the node of each in SDN, can obtain the status information of node;So
Go out the result of remaining probe using probe prediction technological prediction end to end afterwards, the device node of linking probe is selected at random.
Here, without sending probe to all nodes in network, first to random placement in SDN one during probe deployment
Divide probe, the deployed position of remaining probe is then detected using Predicting Technique end to end.Random selection device node
Probe is affixed one's name to, other nodes are attached with the probe being deployed to.
The step sends multiple probes by the Predicting Technique of probe, but not fully sends to all nodes and be
Wherein a little node is sent, as long as the probe of deployment, which can be connected, covers all network equipments, from
And can ensure to complete the detection to SDN with a small amount of probe.
Probe deployment can be obtained in a sparse matrix, sparse matrix on node by the probe set A sent
Each numerical value represent whether probe paths traversed normal.The mode of acquisition is:In this sparse matrix where element
Row correspond to the start node in the path that the probe that has sent passes through, row correspond to the end in the path that the probe sent passes through
Node, so that the access path between forming probe, returning result represents whether the link connected between two nodes connects.If
The result that this probe is returned is normal (being connected between two nodes where representing probe), then corresponds to element in sparse matrix
It is worth for -1;If the result that this probe is returned is failure, the value for corresponding to element in sparse matrix is 1;The probe not sent,
The value for then corresponding to element in sparse matrix is 0, that is to say, that the element in sparse matrix is constituted by 1,0, -1.
Step 2:Calculate the factor of influence of the affiliated node of probe.
There is a factor of influence to the whole network in whole SDN in the node belonging to probe.In this step, utilize
Formula (1)-formula (3) calculates the factor of influence of each node.
For SDN, its business trend is more fixed, and N-1 principles are followed simultaneously for important business, i.e., every
Individual important service can all have a more than one path, and how much the significance level of business can be characterized with optional number of paths, such as
Network element connection figure shown in Fig. 2 is to be used as network model using the sub-fraction node connection in network.
First, network model in fig. 2 does service definition:Because set scale of model is smaller, so limiting industry
The quantity of business, provides to have 5 business, i.e. S1-S5 in a model now.SDN business have itself the characteristics of and type, in order to
The bearing mode and content of these business are distinguished, and the emphasis of concern is embodied, is convenient for based on bearer service resource
The analysis of side, here to set probe to carry out example eight nodes, the form of common factor is taken with resource come table by business S1-S5
Show, user's mathematical formulae is described as follows:
S1:X1∩X2∩X5S2:X3∩X2∩X6S3:X6∩X7∩X8
S4:X4∩X6∩X7S5:X3∩X6∩X8
Next, the impacted degree of analysis business:Because business is moved towards to fix, various failures are to the influence degree of business
Different, for quantitative mark business degree of susceptibility, the standard that influence degree is provided is reported for service impact, with reference to actual motion
In possibility situation, according to assume the problem of model, business degree of susceptibility can be drawn as shown in Table 1:
The impacted degree of the business of table one
Node is represented with X, and L is identified as influence degree, using v as corresponding variable in calculating process, according to
Each node failure of Business Process Analysis for business influence degree, as a result as shown in Table 2:
The node disturbance degree of table two
|
V1 |
V2 |
V3 |
V4 |
V5 |
V6 |
V7 |
V8 |
S1 |
L4 |
L3 |
L1 |
L2 |
L4 |
L2 |
L1 |
L1 |
S2 |
L2 |
L3 |
L4 |
L2 |
L2 |
L4 |
L1 |
L1 |
S3 |
L1 |
L1 |
L1 |
L1 |
L1 |
L4 |
L3 |
L4 |
S4 |
L1 |
L2 |
L1 |
L4 |
L2 |
L3 |
L4 |
L1 |
S5 |
L2 |
L2 |
L4 |
L2 |
L2 |
L3 |
L3 |
L4 |
Then, the factor of influence of calculate node:In the case where network element device breaks down, can with objective method come
Measure influence of the SDN equipment to business.Node failure can not only directly affect business, can also indirectly influence associated section
Point, utilizes Complex Networks Theory, it is considered to the built-in attribute and external attribute of node, the comprehensive factor of influence to seek node.
In SDN, built-in attribute is influence of the node failure for other nodes in network topology, including node
Degree, betweenness, tight ness rating;External attribute is combined influence degree of the node failure to each business.By quantifying these indexs, obtain
To the numerical value of each node factor of influence.
In order to corresponding with built-in attribute, quantify external attribute first, for one numerical value of each grade regulation, such as table three
It is shown:
Combined influence degree of the node failure of table three for business
Influence degree level value in table three is example, in fact should can be arbitrarily defined as times between 0-1
What numerical value.
Node failure influence value:
Wherein:V is some specific node in SDN, and Sn is the business number that node influences, and Li is node to some
The numerical value (numerical value i.e. in table three) that the influence degree of business is quantified, according to the node of SDN and the relationship characteristic of business,
Node failure combined influence value is the average value for each service impact, is also the outside shadow of a node in SDN
Ring power.
SDN is designated as G, wherein there is the link E between node V and node, then has in G=(V, E), node v0 shadow
The factor is rung to be defined as follows:
IF(v0)=α Fin (v0)+βFout(v0) (2)
Wherein, Fin (v0) is the internal influence factor of node, and α is internal influence coefficient;Fout (v0) is the outside of node
Factor of influence, β is external action coefficient, and meets alpha+beta=1, wherein, α=, two values of β can be according to actual conditions and network
Structure is manually set, and the two, which is added, is equal to 1, and the larger then internal influence coefficient value of internal influence is a little high;External action compared with
Greatly, then external action coefficient value is a little high, can set at random without limiting.Consider to be node for outer due to more here
The influence that portion's business is brought, therefore the proportion of the external action factor is larger, and β is taken to relatively large value, preferable experience value is α
=0.2, β=0.8.
Fout (v0) is the external action factor, and as formula (1) calculates the fault impact value Cd (V0) of obtained node,
Fin (v0) is the internal influence factor, is calculated and obtained by following formula (3), it is possible thereby to calculate egress according to formula (2)
Factor of influence IF numerical value.
Here,
Wherein:K (v) is the built-in attribute value of node, can use the node self attributes such as the number of degrees, betweenness, tight ness rating, due to
α value is smaller in the regulation of formula (2), so specifically choosing influence of which built-in attribute value to result and little;max(K
(v) it is) normalization factor, in order to eliminate the influence of network size logarithm value so that it is interval interior that index is unified in [0,1].
Step 3:The path passed by according to every probe to the affiliated node of result of detection and probe of the state of node,
The probability of malfunction of calculate node.
Probe deployment can interconnect to after node, be related to multilink.In this step, every probe is inputted
,, will using Bayesian network model on the basis of the factor of influence of above-mentioned node to the result of detection of the state of affiliated node
The factor of influence of node applies in new probability formula to position the node broken down in network.I.e. according to the detection of every probe
As a result path and the probability of malfunction of node that, probe is passed by, are calculated at formula P (T1 | Pa (T1)) P (T2 | Pa (T2)) respectively
... P (Tm | Pa (Tm)) P (V1) P (V2) ... P (Vn) numerical value.
Step 4:The condition when probability of malfunction for finding out step 3 interior joint reaches maximum.
In this step, find out P (V1, V2 ... Vi ... Vn, T1, T2 ... Tj ... Tm) bar of the value in maximum
The specific node V of some in part, i.e. network value (being 0 or 1), wherein:0 represents node failure, and 1 represents that node is normal.
MMMF algorithms are maximal margin matrix algorithm (Maximum-Margin Ma-trix Factorization), such as
Fig. 3 show MMMF algorithm principle schematic diagrames.MMMF algorithms are applied to the malfunctioning node detection method of the SDN of the present embodiment
When middle, the sparse matrix (matrix is an ill-conditioned matrix) obtained according to the result of detection of probe obtains a complete square
All elements in above-mentioned sparse matrix namely are predicted by battle array, the calculating process for 0 probe, i.e., in predicted path not
The state of the node of probe is set.MMMF is a synchronous study characteristic vector v and coefficient vector u process, by original matrix Y
It is decomposed into the matrix of low norm.Low-rank constraint is substantially exactly the dimension in binding characteristic space, and it is predicted to Y each row
Real is exactly a prediction task on lower dimensional space.Using the nonindependence between element in matrix calculate a feature to
V and coefficient vector u is measured, then a perfect matrix is obtained using the two vectorial products, and this perfect matrix is
For the approximate evaluation of ill-conditioned matrix, the detection event of all probes is represented with this.It should be understood that the probe of deployment
State is also without all detecting and returning, as long as the state for the probe that return path is related to.
The positioning of malfunctioning node is carried out using Bayesian network model after matrix is predicted.What probe was detected
Information, actually failure occur after by various entities interact and the external manifestation of generation, so now network therefore
Barrier positioning just embodies its uncertain feature.And Bayesian network has under condition of uncertainty environment, to causality
The advantage of diagnosis.Bayesian network goes to express the joint probability distribution and its conditional independence of variable, energy in the form of figure simultaneously
Enough substantially reduce failure and determine Bayesian network model and connect the node in network and detection, forming one two layers directly has
Xiang Tu.
The relation between node and probe is illustrated in figure 4, the wherein device node in network is considered as father node, and probe is considered as
Child node, the result of probe depends on the result of its father node, the shape of probe when and if only if its all father node is correct
State is just correct, by calculating the node that the probability of malfunction of each node can be most possibly to break down in location path.
Calculate all according to parameter (prior probability and the result of all probes that include all-network node) to be positioned
The conditional probability of network node, that is, calculate P (V1, V2 ... Vi ... Vn, T1, T2 ... Tj ... Tm) probability, wherein Vi
For i-th of network node, Tj is j-th strip probe, and n is network node number, and m is probe bar number (m < n).P (V1, V2,
... Vi ... Vn, T1, T2 ... Tj ... Tm) calculation formula be:
P(V1,V2,...Vi... Vn,T1,T2,...Tj,...Tm)=
P(T1|Pa(T1))P(T2|Pa(T2))...P(Tm|Pa(Tm))P(V1)P(V2)...P(Vn)
Wherein:
(1) P (V1, V2 ... Vi ... Vn, T1, T2 ... Tj ... Tm) be all nodes conditional probability, characterize section
The influence that point failure is brought to whole network;
(2) Pa (Tj) represents all links that probe Tj passes through, when having node failure in probe Tj, and p (Tj=1 | Pa
(Tj))=1, p (Tj=0 | Pa (Tj))=0, p when trouble-free (Tj=1 | Pa (Tj))=0, p (Tj=0 | Pa (Tj))=
1;Tj=1 represents normal, and Tj=0 represents failure.
(3) prior probability is the out of order probability of network node obtained according to previous experiences and analysis, is an estimation
Value in value, different SDN systems is different.In the model of the present embodiment, the Prior Probability of node failures is
0.1, the Prior Probability p (Xi)=0.9 during node fault-free, therefore, when faulty the probability of malfunction of node be p (Xi)=
0.1*IF (Xi), that is, consider the failure probability of node and the product of factor of influence, because the characteristic of Bayesian network, using
The method of prior probability * factors of influence, obtained instant posterior probability, can more be accurately positioned malfunctioning node.Here, failure probability
As prior probability, probability of malfunction is posterior probability=prior probability * factors of influence.
(4) formula P (V are utilized1,V2..., Vn,T1,T2,...,Tm)=P (Vx)...P(Vy)*0.9n-k, wherein k is network
The number of faults of the node of middle appearance, Vx...Vy is malfunctioning node, the probability of malfunction of P (Vx) ... P (Vy) correspondence respective nodes, pin
To above-mentioned node, k is the number from 0-8, takes different values of the k from 0-8, can calculate different P (V1, V2 ... Vi,
... Vn, T1, T2 ... Tj ... Tm), the value under all situations is calculated, condition when taking its maximum judges V value.
The node that there may be failure for being not provided with probe is also predicted to come, i.e. the number of V=0 malfunctioning node, so
The probability in the case of each node V=0 and V=1 is calculated afterwards.Calculate under all situations P (V1, V2 ... Vi ... Vn,
T1, T2 ... Tj ... Tm) after value, obtain P (V1, V2 ... Vi ... Vn, T1, T2 ... Tj ... Tm) value is in maximum
Condition, i.e. V value (be 0 or 1).
Here it will be understood that to ensure computational efficiency, the present embodiment is only to calculate the shadow for the node that probe is related to
Ring exemplified by the factor, now efficiency should be highest in theory.But, it is not excluded that the factor of influence of all nodes is all calculated
Mode out, is not limited here.
Step 5:All V=0 node set is exported, obtained malfunctioning node set is as calculated.
Because nodes failure can cause service path to interrupt, using probe collector node information, shellfish is then utilized
This network calculations of leaf are out of order probability.In this step, all V=0 node set is malfunctioning node set.
The malfunctioning node detection system of SDN as shown in Figure 5, it includes acquiring unit 1, computing unit 2, positioning list
Member 3 and output unit 4, wherein:
Acquiring unit 1, the state for obtaining nodes, and by the state transfer of node to computing unit 2;
Computing unit 2, for the state of receiving node, the factor of influence and node of calculating network interior joint are in the paths
Faulty probability;
Positioning unit 3, for according to the interior joint of computing unit 2 faulty probability in the paths, finding out the failure of node
Probability reaches condition during maximum;
Output unit 4, the node set of maximum is reached for exporting all probability, obtained malfunctioning node is as calculated
Set.
In the malfunctioning node detection system of the SDN, acquiring unit 1, computing unit 2, positioning unit 3 are corresponded to respectively
The calculating of the information that the acquisition of detecting probe information in the malfunctioning node detection method of SDN, probe are obtained, malfunctioning node are determined
Three steps in position.Acquiring unit 1 occurs in the form of multiple probes, by multiple node deployment probes in SDN, uses
To obtain the state of nodes, while by the information transmission got to computing unit 2.
In acquiring unit 1:Part of nodes sets probe in a network, and is obtained by the state set of the affiliated node of probe
To the sparse matrix of the state of the node of probe paths traversed.
Computing unit 2 is calculated the information that probe is got, and calculates the node factor of influence in network, Yi Jijie
The probability that point breaks down in the paths.Computing unit 2 includes factor of influence computing module, in factor of influence computing module,
The calculation formula of the factor of influence of the affiliated node of probe is:
IF(v0)=α Fin (v0)+βFout(v0)
Wherein:V is some specific node in network, and Fin (v0) is the internal influence factor of node, and α is internal influence
Coefficient;Fout (v0) is the external action factor of node, and β is external action coefficient, and Sn is the business number that node influences, and Li is section
The numerical value that point is quantified to the influence degree of some business, K (v) is the built-in attribute value of node.
Computing unit 2 includes node probability evaluation entity, in node probability evaluation entity, the meter of the probability of malfunction of node
Calculating formula is:
P(V1,V2,...Vi... Vn,T1,T2,...Tj,...Tm)=
P(T1|Pa(T1))P(T2|Pa(T2))...P(Tm|Pa(Tm))P(V1)P(V2)...P(Vn)
Wherein:P (V1, V2 ... Vi ... Vn, T1, T2 ... Tj ... Tm) be all nodes conditional probability, characterize
The influence that the failure of node is brought to whole network;
(Pa (Tj) represents all links that probe Tj passes through, when having node failure in probe Tj, and p (Tj=1 | Pa
(Tj))=1, p (Tj=0 | Pa (Tj))=0, when probe Tj is trouble-free, and p (Tj=1 | Pa (Tj))=0, p (Tj=0 | Pa
(Tj))=1;Tj=1 represents normal, and Tj=0 represents failure;
Vi is i-th of node, and Tj is j-th strip probe, and n is node number, and m is probe bar number (m < n).
Preferably, computing unit 2 also includes completion module, and completion module is used to calculate the shadow of the affiliated node of probe
After the sound factor, also include step before the probability of malfunction of calculate node:According to sparse matrix, calculated using maximal margin matrix
Method obtains the perfect matrix for the state for including all nodes.
It is understood that the principle that embodiment of above is intended to be merely illustrative of the present and the exemplary implementation that uses
Mode, but the invention is not limited in this.For those skilled in the art, the essence of the present invention is not being departed from
In the case of refreshing and essence, various changes and modifications can be made therein, and these variations and modifications are also considered as protection scope of the present invention.