CN102684902B - Based on the network failure locating method of probe prediction - Google Patents
Based on the network failure locating method of probe prediction Download PDFInfo
- Publication number
- CN102684902B CN102684902B CN201110066944.5A CN201110066944A CN102684902B CN 102684902 B CN102684902 B CN 102684902B CN 201110066944 A CN201110066944 A CN 201110066944A CN 102684902 B CN102684902 B CN 102684902B
- Authority
- CN
- China
- Prior art keywords
- probe
- network
- matrix
- malfunctioning node
- sent
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a kind of network failure locating method based on probe prediction, comprise step: step 101: the detection set A that initialization is current and malfunctioning node set F are for empty; Wherein, detect set A and represent the issued probe selected from probe set M; Malfunctioning node set F represents the malfunctioning node set navigated to; Step 102: select K probe and send from probe set M; Step 103: obtain a sparse matrix by probe set A, and remove the probe sent from described probe set M; Step 104: predict obtained sparse matrix, obtains a perfect matrix; Step 105: select the most uncertain N number of probe and send from described perfect matrix; Step 106: utilize the result of the probe sent in step 105 to upgrade described perfect matrix; Step 107: the location of carrying out malfunctioning node, calculates malfunctioning node set.The method reduce the required number of probes sent of active probing technique.
Description
Technical field
The present invention relates to network detection technical field, particularly relate to a kind of network failure locating method based on probe prediction.
Background technology
Along with popularizing of internet information, the scale of computer system network constantly expands and constantly increases with complexity, and the difficulty of network management is also increasing, how to manage network timely and effectively and becomes our current problem demanding prompt solution.
At present, received Network Management Function must comprise following five aspects: fault management, accounting management, configuration management, performance management and safety management.Fault management is again one of of paramount importance function in network management.When certain component failure in network, network manager must find the source of trouble timely and accurately and get rid of, with the reliability of Logistics networks.When network size is smaller, network manager according to processing alarm for information about, can be fixed a breakdown.And when network size complexity is higher, network manager just should perform some diagnostic tests to distinguish failure cause.And along with the continuous expansion of network size and the raising day by day of network complexity, the difficulty of fault management is also just increasing.
Failure diagnosis, as a Core Feature of fault management, plays increasing effect in current network management.Fault diagnosis technology the most frequently used is at present active probing technique.First active probing technique sends detection from selected base station to other all network equipment, then by calculating the entropy of detection, is then applied in Bayesian network model, by further compute location to the source of trouble in network.But along with the continuous expansion of network size, the increasing of all kinds of service application amount, and fault is in the appearance of different agreement layer, also more and more higher to the requirement of network detection.
The method of a current solution network failure location difficult problem mainly utilizes the technology of initiatively probe.Probe is used for the end-to-end test affairs collection of gathering system network performance information specially, initiatively probe is then utilize probability inference technology and the active from probe sets of combining information opinion method selects a small amount of probe most with amount of information, to carry out Rapid Inference to the state of grid.
The state of development of prior art is as follows:
Technical scheme 1: application number is the Chinese patent application of 200510074911.X, relate to a kind of method of recognition network malfunctioning node, its core technology is: first the network node of system sends detection messages to network node associated with it, then the response message of described associated network node is monitored, when the response message of the network node be associated described in confirmation does not receive, then determine that this related network node is failing network-node.
Technical scheme 2: application number be 200710188015.5 Chinese patent application disclose a kind of service failure diagnosis system based on active probe and method thereof, use the service performance in the effective monitoring network of active probe, ensure the performance using service in each Access Network, when monitoring notes abnormalities, the symptom according to monitoring orients fault rootstock rapidly and accurately.This application for a patent for invention monitoring overhead is little, has good performance of fault diagnosis, and this application for a patent for invention simultaneously chooses probe based on uncertainty models, with the verification and measurement ratio of less probe expense guarantee to each fault, and can obtain good diagnosis performance.
Literature protocol 1: " Adaptive diagnosis in distributed systems ", the document proposes initiatively probe technique the earliest.Probe is used for the end-to-end test affairs collection of gathering system network performance information specially, initiatively probe technique is then utilize probability inference technology and the active from probe sets of combining information opinion method selects a small amount of probe most with amount of information, to carry out Rapid Inference to the state of grid.This algorithm not only in the number of probes sent than traditional algorithm few a lot, and also greatly to reduce on the time spent by fault location.
Literature protocol 2: " Fault diagnosis in IP Networks via Noiseless Multicast Probing ", this document proposes one based on the fault model of extracting of multicasting technology, solve a fault location difficult problem.Extract in fault model at this, certain ancestor node of and if only if a reception probe breaks down, and this receives probe just can record this fault measuring.Utilize this model and most probable malfunctioning node can be calculated by probe value.The model treatment method that the document proposes makes full use of the fabric feature of multicast tree, more effective compared with common probability inference.
Discuss the defect of prior art below.
At present, from this subject of computer science, derive diversified network monitoring method, generally speaking can be divided into two classes: the detection method of passive collection information and the network detection method of active probe.The network management system adopting passive collection information to carry out this method of network detection requires that networking component provides the function of inside story to network management system, with the message active reporting oneself collected to network management system.But along with the continuous increase of network size and the increase of complexity, the network management system of passive collection information, due to the large and lower operational efficiency of its difficulty realized, cannot meet the needs managing the network become increasingly complex.Meanwhile, such as not be there is management interface by the managed entity of third party's assembly, middleware or other main frame exploitation, the function of reporting message cannot be realized.This just requires that we provide better model to detect network, and under this background, active probing technique is carried out.Network management system based on active probe uses the method for active probe detect managed device and analyze result of detection, to complete fault location, the message collected by it is reported without the need to managed device, there is active, efficient and adaptive characteristic, and the symptoms such as network, the system failure, serv-fail and performance degradation can be known as soon as possible with very little cost, thus provide foundation for the positioning analysis of root fault.
Although the method for active probe tool compared with the method for passive collection information has great advantage, its configuration overhead extra for network brings and flow load.First, the method for active probe needs to configure in a network the detection base station of some to ensure that the probe sent from base station can cover whole network, and the fault that can effectively occur in fixer network.And the existence of such special joint will introduce configuration to node and maintenance costs.Secondly, the probe of transmission is in fact also a kind of network overhead, although excessive network detection set is good for result of detection, also brings larger load to network, causes the congested of some network node.Therefore, from candidate probe set, how to select that one efficient and offered load is less detection set just seems particularly important.The uncertainty of network state should be reduced to minimum by the probe set that namely probe selection algorithm is selected, and the scale of probe set is little as much as possible simultaneously.
Specifically, the defect of technique scheme 1 is: the program sends to its network node be associated the state that monitoring message judges its network node be associated by network node, this scheme needs to send in a network to monitor message in a large number, very large on the flow load impact of network, and when cannot receive monitoring message when network congestion, can not determine it is that the network node be associated necessarily is in malfunction.
The defect of technique scheme 2 for: the program first sends the monitoring probe of some in institute's monitoring network, whether there to be fault in supervising network produce, the diagnostic probe of some is sent again to carry out fault location when there being fault to produce in network, this scheme needs to send a large amount of probes in network, when network is more complicated, the required number of probes sent is more, which increases the flow in network, larger to the load effect of network.
The defect of above-mentioned literature protocol 1 is: the document proposes the technology of active probe the earliest, the advantage of this algorithm be make use of probability theory technology and combining information opinion method has carried out reasoning to network state, compared to passive collection message method offered load expense little, accurate positioning.But this algorithm needs to carry out complicated calculating to the entropy of probe, and calculation cost is very high, and is only applicable to the situation occurring single fault in network.So this algorithm be not suitable for the more situation of current this network environment more complicated, failure ratio.
The defect of above-mentioned literature protocol 2 is: in the consideration technically adding time factor of active probe, and make the obtainable mutual information of probe set institute large as much as possible, not only fault location prepares more, and the average overhead of network is also less.But this algorithm needs to send a large amount of probe in network, still larger to the occupancy of Internet resources.
Summary of the invention
(1) technical problem that will solve
For needing the problem sending a large amount of probe in network in existing network Detection Techniques, the technical problem to be solved in the present invention is: reduce the required number of probes sent of active probing technique, and quick position is to the fault in network.
(2) technical scheme
In order to solve the problems of the technologies described above, the invention provides a kind of network failure locating method based on probe prediction, comprising the following steps:
Step 101: the detection set A that initialization is current and malfunctioning node set F are empty; Wherein, detect set A and represent select, issued probe from probe set M; Malfunctioning node set F represents the malfunctioning node set navigated to; Probe set M is made up of the probe sent to the network equipment in network to be positioned;
Step 102: select K probe and send from probe set M;
Step 103: obtain a sparse matrix by detecting set A, and remove the probe sent from described probe set M;
Step 104: predict obtained sparse matrix, obtains a perfect matrix;
Step 105: select the most uncertain N number of probe and send from described perfect matrix;
Step 106: utilize the result of the probe sent in step 105 to upgrade described perfect matrix;
Step 107: the location of carrying out malfunctioning node, calculates malfunctioning node set.
Wherein, in step 104, maximal margin matrix decomposition MMMF is utilized to predict obtained sparse matrix.
Wherein, in step 107, Bayesian network model is utilized to carry out the location of malfunctioning node.
Wherein, also comprise after step 107: judge whether the up-to-date malfunctioning node set calculated by Bayesian network model is changed compared with current malfunctioning node set, if unchanged, the up-to-date malfunctioning node set obtained is the result that will obtain, otherwise enters the fault location that step 104 carries out next round.
Wherein, before the fault location of carrying out next round, first the up-to-date perfect matrix obtained is transformed into two sub matrixs, then performs the fault location that next round is carried out in step 104 ~ 107, until malfunctioning node set no longer includes change.
Wherein: 201: the characteristic vector u of study sparse matrix and coefficient vector v; 202: the product calculating u and v, obtains perfect matrix.
Wherein, step 107 specifically comprises:
Step 301, conditional probability according to calculation of parameter all-network node to be positioned;
Step 302: calculate send as an envoy to described conditional probability maximum time condition, all node set meeting this condition are the malfunctioning node set that will calculate.
Wherein, in the perfect matrix obtained when adopting MMMF Forecasting Methodology to predict, the real number in element is filled to be [-1,1] interval, absolute value of a real number is less, and the uncertainty of probe is higher.
Wherein, described parameter to be positioned comprises the prior probability of all-network node and the result of all probes.
(3) beneficial effect
The network modelling of reality is become the Bayesian network model of two points by the present invention, this model is conducive to determining the relation between network node and all the other nodes, according to the feature of model, select the set of the probe that will send, but also not exclusively send and just send wherein a little, then the algorithm of the performance prediction end to end MMMF proposed is utilized to predict the result that all the other do not send probe, and then select a small amount of detection and send, to improve the accuracy of prediction, not only make the required number of probes sent in network less, thus avoid network failure to locate the unnecessary network overhead caused, and the accuracy of fault location is also very high, realize the fault location of the low expense of high efficiency.Its key problem in technology point is:
1) bayesian theory real network is utilized to carry out modeling.By setting up Bayesian model, can allowing each network node that their respective states need not be reported just can to carry out fault location, decreasing the network traffics produced because reporting.
2) adopt the method for probe prediction to carry out fault location, and do not send all probes selected.Predict the result of all probes by sending a small amount of probe, thus carry out fault location, so just further mitigate the flow burden in network.
3) adopt and select the most uncertain probe that predicts the outcome to resend from the result of detection predicted, improve the accuracy of prediction.In the process of carrying out probe prediction, if do not select part probe to re-start transmission, then the accuracy of probe prediction is very low, and the result of fault location is just not necessarily accurate.If Stochastic choice part probe re-starts transmission, then may select those probes sent, the accuracy of actual probes prediction improves seldom.And select predict the outcome in the most uncertain probe resend, avoid select those probes sent, also farthest improve the accuracy of probe prediction, thus failing network-node can be navigated to faster.
Accompanying drawing explanation
Fig. 1 is the method flow diagram of the embodiment of the present invention;
Fig. 2 is the schematic diagram that in the method for the embodiment of the present invention, particle carries out location updating;
Fig. 3 is the example of a Bayesian network model;
Fig. 4 is the flow chart utilizing Bayesian network model to carry out fault location;
Fig. 5 shows experimental result.
Embodiment
Below in conjunction with drawings and Examples, the specific embodiment of the present invention is described in further detail.Following examples for illustration of the present invention, but are not used for limiting the scope of the invention.
Mentality of designing of the present invention is: first in network, send a small amount of probe, utilize probe prediction technological prediction end to end to go out the result of remaining probe after obtaining result of detection, finally utilize the node broken down in the Bayesian network model computing network in active probing technique.
First the parameter will used when implementing the localization method of the embodiment of the present invention is described:
Network topology: the i.e. topology of networks of pending fault location, comprise network-termination device, communication equipment and transmission medium, network-termination device and communication equipment can be referred to as the network equipment.Wherein, network-termination device can be computer, server etc.; Communication equipment can be router, switch etc.; Transmission medium can be optical fiber, twisted-pair feeder and wireless etc.;
Probe set M: select multiple (getting 10 in the present embodiment) computer at random as probe base station, ping probe is sent to detect the packet loss between each network equipment according to critical path method (CPM) from these probe base stations to other all-network terminal equipment and communication equipment, namely multiple (getting 100 in the present embodiment) bag is sent with ping probe to other all-network terminal equipment and communication equipment, then detect and have received how many bags, the result of probe is defined as the bag number that receives divided by the bag number sent, set to the probe of the network equipment transmission in network to be positioned constitutes probe set M, wherein, ping probe is the computer network program for measuring end to end connectivity, and it is by sending a request data package and wait-receiving mode response data packet is confirmed whether to be communicated with opposite end to opposite end.
The number of probes K of initial transmission: set the initial number of probes sent as 1/4th (being obtained by experience) of total number of probes in the present embodiment, or add 20 or deduct 20 for 1/4th of total number of probes for 1/4th of total number of probes;
The number of probes N that each selection next round will send: because the number of probes sent at first is fewer, cause the prediction accuracy of the result of probe set M very low, so need select from probe set M some, least determine that the probe predicted the outcome sends, in the present embodiment, the value of N is 5,10,20 or 30.
The method flow diagram of the foundation embodiment of the present invention as shown in Figure 1, comprises the steps:
Step 101: the detection set A that initialization is current and malfunctioning node set F are empty.Wherein, detect set A and represent the issued probe selected from probe set M; Malfunctioning node set F represents the malfunctioning node set utilizing active probing technique to navigate to.
Step 102: select K probe and send from probe set M.
The system of selection of probe: select from probe set M at random, ensures that in the matrix corresponding with probe, every a line and each row are all non-full zeros.
Step 103: obtain a sparse matrix by probe set A, and remove the probe sent from probe set M.
A sparse matrix can be obtained by probe set A, the mode obtained is: the row of this sparse matrix corresponds to the start node of the probe sent, row correspond to the terminal note of the probe sent, if the result that this probe returns is normal, then in corresponding sparse matrix, the value of element is-1, the element value then corresponding for fault is 1, and do not have the element value in the corresponding sparse matrix of probe sent to be 0, the element that is in sparse matrix forms by 1,0 ,-1.
Step 104: utilize MMMF (Max Margin Matrix Factorization, maximal margin matrix decomposition) to predict obtained sparse matrix, obtain a perfect matrix.
The sparse matrix (this matrix is an ill-conditioned matrix) obtained by MMMF calculation procedure 103, obtain a perfect matrix, the probe that all elements in above-mentioned sparse matrix is namely 0 by this computational process is all predicted out.MMMF is the process of a synchronous study characteristic vector v and coefficient vector u.Computational process utilizes the nonindependence in matrix between element to calculate an a characteristic vector v and coefficient vector u, then utilize these two vectorial products to obtain a perfect matrix B=uv, and this perfect matrix is the approximate evaluation of ill-conditioned matrix.
Detailed MMMF predicts the step of ill-conditioned matrix as shown in Figure 2, comprising: step 201: the characteristic vector u and the coefficient vector v that learn out ill-conditioned matrix (namely sparse matrix) to be predicted; Step 202: calculate u*v, obtain the approximate evaluation of this ill-conditioned matrix, i.e. perfect matrix, the element set in this perfect matrix becomes the real number in [-1,1] interval.
Step 105: predict from step 104 in the perfect matrix obtained that selecting the most uncertain N number of probe sends.
The most uncertain probe is defined as: adopt MMMF Forecasting Methodology to carry out in the perfect matrix obtained when predicting, element is filled to be [-1,1] real number in interval, from 0 more close to (namely absolute value of a real number is less), uncertainty is higher, selects the highest N number of probe of uncertainty to send.
The value of N: when carrying out raising prediction accuracy, needs the probe selecting some from the perfect matrix doped to carry out the transmission of next round, and the quantity of each detection selected is different, and it is just different that prediction accuracy improves degree.As can be seen from following table 1, the number of probes selected is less, the accuracy raising degree of prediction is higher, but the number of probes selected is less, the number of repetition of the approximate matrix calculated is more, the time consumed is more, so need to balance the number of probes selected and the relation calculated between approximate matrix number of times.In the present invention, the value of N is 5,10,20 or 30, corresponding with initial N value.
Step 106: the probe utilizing step 105 to send upgrades the perfect matrix doped.
The result of the probe sent by epicycle replaces element value corresponding in the perfect matrix doped.
Step 107: utilize Bayesian network model to carry out the location of malfunctioning node.
Bayesian network model: the node in network and detection connect by Bayesian network model, form two layers of direct directed graph, node wherein in network is considered as father node, probe is considered as child node, the result of probe depends on the result of its father node, when and if only if its all father node is correct, this probe is just correct, and Fig. 3 is expressed as the example of a Bayesian network model.The node most possibly broken down can be located by the probability of malfunction calculating each node.
Figure 3 shows that set up a Bayesian network model example.
Utilize Bayesian network model to carry out the flow process of fault location as shown in Figure 4, comprising:
Step 301, (comprise the prior probability of all-network node and the result of all probes according to parameter to be positioned, described prior probability is according to previous experiences and analyzes the out of order probability of network node obtained, it is an estimated value, value in different network systems is different, this value is 0.9 in the present embodiment) calculate the conditional probability of all-network node, namely calculate P (t
1, t
2..., t
n, X
1, X
2..., X
m) probability, wherein t
ibe i-th network node, X
jfor jth bar probe, n is network node number, and m is probe number.P (t
1, t
2..., t
n, X
1, X
2..., X
m) computing formula be
P(t
1，t
2，...，t
n，X
1，X
2，...，X
m)
＝P(X
1|Pa(X
1))P(X
2|Pa(X
2))...P(X
m|Pa(X
m))P(t
1)P(t
2)...P(t
n)，
In above formula, Pa (X
i) represent all links of probe Xi process.As probe X
iin when having a node failure, P (X
i=1|Pa (X
i))=1, P (X
i=0|Pa (X
i))=0, P (X time trouble-free
i=1|Pa (X
i))=0, P (X
i=0|Pa (X
i))=1.And P (t
1) P (t
2) ... P (t
n)=0.1
k* 0.9
n-k, wherein k corresponds to the number of faults of the node occurred in network, i.e. t
ithe number of=0.Then P (t is calculated
1, t
2..., t
n, X
1, X
2..., X
m) middle t
i=0 and t
iprobability in=1 situation.Prior probability above-mentioned specifically refers to obtain the out of order probability of network node according to previous experiences and analysis, is an estimated value.
Step 302: calculate the P (t under all situations
1, t
2..., t
n, X
1, X
2..., X
m) after value, find out P (t
1, t
2..., t
n, X
1, X
2..., X
m) condition of value when maximum, i.e. t
ivalue (being 0 or 1).All t
ithe node set of=0 is the malfunctioning node set calculated.
The present embodiment can also comprise the following steps:
Step 108: judge whether the up-to-date malfunctioning node set calculated by Bayesian network model is changed compared with current malfunctioning node set, if unchanged, this up-to-date malfunctioning node set obtained is the result that will obtain, otherwise enters the fault location that step 104 carries out next round.When carrying out the fault location of next round, first the up-to-date perfect matrix obtained is transformed into two sub matrixs, namely be greater than in matrix 0 be transformed into 1, be less than 0 be transformed into-1, run MMMF method again, and apply the calculating that Bayesian network model method carries out probability of node failure, repeat above step until malfunctioning node set no longer changes.
The following describes embodiments of the invention:
We test the present invention in this network simulation environment of NS2.First utilize this instrument stochastic generation of GI-ITM topological diagram, number of nodes is 100, and number of links obeys random distribution, and the number of links of topological diagram is respectively 99.In order to the fault in analog network, add error model in simulations, the packet loss of normal link is set as 0.001, and the packet loss of faulty link is set as 0.9.The each number of probes N selecting next round to send of setting is 10, and in the topology of 100 nodes, K is 220, and initialization has detected set A and malfunctioning node set F is sky.Then from probe set M, select 220 probes send, obtain a sparse matrix:
Then utilize MMMF method to predict this sparse matrix, obtain perfect matrix:
Utilize Bayesian network model to calculate the probability of malfunction of all-network node after doping the result of all probes, namely in step 301
P(t
1，t
2，...，t
n，X
1，X
2，...，X
m)
＝P(X
1|Pa(X
1))P(X
2|Pa(X
2))...P(X
m|Pa(X
m))P(t
1)P(t
2)...P(t
n) (3)，
Wherein X
j, the value of j=1,2...m is corresponding with the value in matrix (2), and in matrix, element is-1, then X
jvalue be 1, element is 1, then X
jbe 0, then t
i=0 and t
ithe situation of=1 substitutes into formula (3) respectively, can obtain t
ip (t when=0
i)=0.9, t
ip (t when=1
i)=0.1, then can calculate P (t
1, t
2..., t
n, X
1, X
2..., X
m) probability in varied situations, find P (t
1, t
2..., t
n, X
1, X
2..., X
m) maximum that situation, i.e. t
icorresponding value, t
i=0 be malfunctioning node, the malfunctioning node set F calculated like this is { 38, 52, 67}, judge whether this malfunctioning node set changes, because failure collection becomes { 38 by empty set, 52, 67}, just from matrix (2), the minimum element of 10 absolute values is found out, send the probe corresponding to them, and the result obtained is updated in matrix (2), then matrix (2) is transformed into two sub matrixs, namely be greater than in matrix 0 be transformed into 1, be less than 0 be transformed into-1, run MMMF method again, and apply the calculating that Bayesian network model method carries out probability of node failure, repeat above step until malfunctioning node set no longer changes, namely malfunctioning node set is { 38, 53, 64} just no longer changes, then malfunctioning node set { 38, 53, 64} is the result wanted.Fig. 5 lists the result of this experiment.
As seen from Figure 5, the probe that will send required for the present invention is also fewer than 1/3rd of total probe set, simultaneously also can be as seen from Table 1, after predicting, reselect a small amount of probe at every turn re-start the larger of the accuracy raising sending rear prediction, this also make required for the negligible amounts of probe that resends, thus make the required total number of probes sent seldom just can complete fault location.Table 1 is expressed as the relation between the value of N and prediction error rate.
Table 1
The above is only embodiments of the present invention; it should be pointed out that for those skilled in the art, under the prerequisite not departing from the technology of the present invention principle; can also make some improvement and modification, these improve and modification also should be considered as protection scope of the present invention.
Claims (2)
1. based on a network failure locating method for probe prediction, it is characterized in that, comprise the following steps:
Step 101: the detection set A that initialization is current and malfunctioning node set F are empty; Wherein, detect set A and represent select, issued probe from probe set M; Malfunctioning node set F represents the malfunctioning node set navigated to; Probe set M is made up of the probe sent to the network equipment in network to be positioned;
Step 102: select K probe and send from probe set M, the system of selection of described probe is: select K probe at random from probe set M, and in the matrix corresponding with a described K probe every a line and each to arrange are all non-full zeros;
Step 103: obtain a sparse matrix by detecting set A, and the probe sent is removed from described probe set M, the acquisition pattern of described sparse matrix is: the row of described sparse matrix corresponds to the start node of the probe sent, row correspond to the terminal note of the probe sent, if the result that this probe returns is normal, element value then in corresponding sparse matrix is-1, if the result that this probe returns is fault, element value then in corresponding sparse matrix is 1, does not have the element value in the corresponding sparse matrix of probe sent to be 0;
Step 104: predict obtained sparse matrix, obtains a perfect matrix, is specially: utilize maximal margin matrix decomposition MMMF to predict obtained sparse matrix;
Step 105: select the most uncertain N number of probe and send from described perfect matrix, wherein, in the perfect matrix obtained when adopting MMMF Forecasting Methodology to predict, element is filled to be [-1,1] real number in interval, absolute value of a real number is less, and the uncertainty of probe is higher;
Step 106: utilize the result of the probe sent in step 105 to upgrade described perfect matrix;
Step 107: the location of carrying out malfunctioning node, calculates malfunctioning node set, is specially: utilize Bayesian network model to carry out the location of malfunctioning node;
Described step 107 specifically comprises:
Step 301: according to the conditional probability of calculation of parameter all-network node to be positioned, described parameter to be positioned comprises the prior probability of all-network node and the result of all probes;
Step 302: calculate send as an envoy to described conditional probability maximum time condition, meet all t of this condition
ithe node set of=0 is the malfunctioning node set that will calculate, wherein, and t
ibe i-th network node;
Also comprise after step 107: judge whether the up-to-date malfunctioning node set calculated by Bayesian network model is changed compared with current malfunctioning node set, if unchanged, the up-to-date malfunctioning node set obtained is the result that will obtain, otherwise enters the fault location that step 104 carries out next round; Before the fault location of carrying out next round, first the up-to-date perfect matrix obtained is transformed into two sub matrixs, then performs the fault location that next round is carried out in step 104 ~ 107, until malfunctioning node set no longer includes change.
2. the method for claim 1, is characterized in that, step 104 specifically comprises:
Step 201: the characteristic vector u of study sparse matrix and coefficient vector v;
Step 202: the product calculating u and v, obtains perfect matrix.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110066944.5A CN102684902B (en) | 2011-03-18 | 2011-03-18 | Based on the network failure locating method of probe prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110066944.5A CN102684902B (en) | 2011-03-18 | 2011-03-18 | Based on the network failure locating method of probe prediction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102684902A CN102684902A (en) | 2012-09-19 |
| CN102684902B true CN102684902B (en) | 2015-10-14 |
Family
ID=46816303
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110066944.5A Expired - Fee Related CN102684902B (en) | 2011-03-18 | 2011-03-18 | Based on the network failure locating method of probe prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102684902B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103501257B (en) * | 2013-10-11 | 2016-10-19 | 北京邮电大学 | A kind of system of selection of IP network fault probe |
| CN105486976A (en) * | 2015-11-19 | 2016-04-13 | 云南电力调度控制中心 | Fault positioning detection selection method |
| CN105721209B (en) * | 2016-02-19 | 2018-09-04 | 国网河南省电力公司信息通信公司 | A kind of fault detection method having network of making an uproar |
| CN106452877B (en) * | 2016-10-19 | 2021-09-24 | 国网江苏省电力公司南京供电公司 | Fault positioning method for power information network |
| CN107171832A (en) * | 2017-05-02 | 2017-09-15 | 国网辽宁省电力有限公司 | A kind of malfunctioning node detection system of SDN |
| CN107135100A (en) * | 2017-05-02 | 2017-09-05 | 国网辽宁省电力有限公司 | A kind of malfunctioning node detection method of SDN |
| CN109886292B (en) * | 2019-01-09 | 2021-05-11 | 同济大学 | Abnormal reason diagnosis method based on abnormal association graph |
| CN112422322A (en) * | 2020-10-28 | 2021-02-26 | 北京邮电大学 | Multi-objective optimization method based on probe scheduling and related equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101170447A (en) * | 2007-11-22 | 2008-04-30 | 北京邮电大学 | Service failure diagnosis system based on active probe and its method |
| CN101394314A (en) * | 2008-10-20 | 2009-03-25 | 北京邮电大学 | Fault positioning method for Web application system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7640224B2 (en) * | 2007-03-26 | 2009-12-29 | International Business Machines Corporation | Active sampling collaborative prediction method for end-to-end performance prediction |
-
2011
- 2011-03-18 CN CN201110066944.5A patent/CN102684902B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101170447A (en) * | 2007-11-22 | 2008-04-30 | 北京邮电大学 | Service failure diagnosis system based on active probe and its method |
| CN101394314A (en) * | 2008-10-20 | 2009-03-25 | 北京邮电大学 | Fault positioning method for Web application system |
Non-Patent Citations (1)
| Title |
|---|
| 褚灵伟,邹仕洪,程时端,王文东.概率和噪声环境下基于主动探针的Internet服务故障管理.《中国科学 E辑:信息科学》.2008,第38卷(第10期),1733-1746. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102684902A (en) | 2012-09-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102684902B (en) | Based on the network failure locating method of probe prediction | |
| CN107394773B (en) | Distribution network information physical system reliability assessment method considering fault processing overall process | |
| Miao et al. | Agnostic diagnosis: Discovering silent failures in wireless sensor networks | |
| Dusia et al. | Recent advances in fault localization in computer networks | |
| US11348023B2 (en) | Identifying locations and causes of network faults | |
| Di Martino et al. | Automated generation of performance and dependability models for the assessment of wireless sensor networks | |
| CN101783749B (en) | Network fault positioning method and device | |
| CN104796273A (en) | Method and device for diagnosing root of network faults | |
| CN102802182B (en) | A kind of wireless sensor network fault diagnostic device and method | |
| CN101170447A (en) | Service failure diagnosis system based on active probe and its method | |
| Lu et al. | A new fault detection method for computer networks | |
| Liu et al. | Self-diagnosis for detecting system failures in large-scale wireless sensor networks | |
| CN102739527B (en) | Network packet loss rate detection method | |
| CN111600805B (en) | Bayes-based power data network congestion link inference method | |
| CN104486113A (en) | Fault link positioning method based on active greed and passive greed in sensor network | |
| CN104022907A (en) | Failure detection system and method of campus network | |
| JP3579834B2 (en) | Proactive online diagnostics in manageable networks | |
| CN102299829B (en) | Network failure probing and positioning method | |
| CN102883359A (en) | Method, device and system for measuring key nodes of wireless sensor network | |
| CN102281103A (en) | Optical network multi-fault recovering method based on fuzzy set calculation | |
| Yüksel et al. | Modelling and analysis of smart grid: A stochastic model checking case study | |
| Arjannikov et al. | Using markov chains to model sensor network reliability | |
| Alghuried et al. | Anomaly detection in large-scale networks: A state-space decision process | |
| Bhattacharyya et al. | A discrete event systems approach to network fault management: detection and diagnosis of faults | |
| CN107171868A (en) | The malfunctioning node detection method and its malfunctioning node detection system of SDN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20151014 Termination date: 20210318 |