CN102263676A - Network bottleneck detection method - Google Patents

Abstract

The invention relates to the technical field of networks and discloses a network detection method. The method comprises the following steps: 101, establishing a mathematical model according to a network topology structure; 102, determining an objective function of a geometric programming problem according to a network utility maximization principle; 103, judging whether the input rates of all paths are fixed, and then executing a step 104, a step 107 and a step 108 in sequence if the input rates of all paths are fixed or executing a step 105, a step 106, the step 107 and the step 108 in sequence if the input rates of all paths are not fixed; 104, determining a constraint condition of the fixed rate; 105, reading a set insurance degree value epsilon; 106, determining a constraint condition of a random rate; 107, solving the geometric programming problem to obtain the packet loss rate of each link, wherein the geometric programming problem is composed of the objective function and the constraint condition of the fixed rate or the random rate; and 108, determining a network bottleneck according to the solved packet loss of each link. The detection method provided by the invention can be used for improving the accuracy of the network bottleneck detection, reducing the network load and improving the flexibility ratio of the network management.

Description

The network bottleneck detection method

Technical field

The present invention relates to networking technology area, be specifically related to a kind of network bottleneck detection method.

Background technology

Detect network bottleneck and the result is analyzed, can be well understood to each assembly operating situation in the network, simultaneously the operation conditions of unusual assembly the user be can be informed, foundation and reasonable suggestions provided for the user aspect improvement network performance, the raising network service quality.Therefore, network bottleneck is carried out correlation analysis playing the part of the role of ever more important with research in the exploration of network performance management, research and Innovation Networks bottleneck analysis algorithm hold the balance to the effect that improves overall performance of network.In numerous performance characteristics, bandwidth, delay and packet loss are three the most frequently used in network performance management important indicators, also are three principal elements that produce network bottleneck.So the research of the network bottleneck of present stage mainly concentrates on three aspects, be respectively the measurement that bottleneck measurement and link packet drop rate were measured, postponed to bandwidth bottleneck.Introduce respectively below.

1. bandwidth bottleneck detection technique

Bandwidth bottleneck is meant that specifically restricting data wraps in the amount of bandwidth of the ability of propagating in the network link, can be divided into measurement of link bandwidth bottleneck and available bandwidth bottleneck and measure, and usually the measurement to the link bandwidth bottleneck is called the measurement of bottleneck bandwidth.

Mainly adopt the method for bandwidth measurement at present for the measurement of link bandwidth bottleneck, mainly be divided into two classes: a class is single bag method of measurement, basic thought is that the method for measurement of the variant pack size that proposes of the linear relationship according to packet size and propagation delay time is measured, this method mainly utilizes the various minimum delays of the bag of different sizes to ask the bandwidth of link by link system to compare minimum bandwidth then, and its main survey tool has Pathchar, Pchar, Clink and Bing etc.; Another kind of method is that bag is to method of measurement, basic thought is to utilize the bag that proposes based on the right time delay spacing of back-to-back packet and bandwidth relationship to method of measurement (Packet Pair/Train Dispersion, PPTD) measure link bandwidth, this method sends two or more packets back-to-back, according to packet back-to-back because bottleneck bandwidth is estimated at the interval that queuing forms, main survey tool has Sprobe, Pathrate, TCPanaly and Nettimer.

The detection technique of available bandwidth bottleneck mainly is divided into two classes, i.e. packet rate method PRM and inter-packet gap method PGM.PRM is based on a measurement model of " introduce certainly and block " notion, and basic thought is an available bandwidth of deriving link by the linear relationship that packet became that sends different rates, and typical survey tool is Pathload.The basic thought of PGM is that maintenance data bag formation ratio of propagation time difference and data packet length in bottleneck link is inferred available bandwidth, and the measurement of correlation instrument has: Spruce, IGI and Delphi.Up-to-date available bandwidth bottleneck detection technique is Cprobe at present, and it estimates available bandwidth by the short ICMP data packet queue of transmission.

In the patented technology of bandwidth bottleneck detection technique aspect, application number is that 200310113676.3 Chinese patent application discloses a kind of end-to-end network bottleneck bandwidth measuring method, comprise the steps: three measurement parameters of 1. preresearch estimatess: the loopback delay RTT of tested network path, bandwidth measurement accuracy rating BIN and one are than actual value bottleneck bandwidth left side dividing value; 2. it is right that transmitting terminal sends the long measurement data bag of the different bags of many groups, receiving terminal receive the measurement data bag to after, record storage of measurement data, and send and confirm to reply; 3. receiving terminal divides into groups measurement data with different sequence numbers by the difference bag is long respectively, interweave in twos and carry out rectangle and present processing, obtain several and measure sample value distributed rectangular figure, then according to the position distribution situation at peak in all distribution maps, utilization is chosen peak and filter peak operation and is found out the relatively-stationary peak position that becomes, and draws the bottleneck bandwidth measurement result value.

2. postpone the bottleneck detection technique

Network delay is one of important attribute of network performance, and it comprises three parts such as queueing delay, transmission delay and propagation delay.In measuring end to end, packet is orientated as through the path at maximum delay place in the sequence of router node formation and is postponed bottleneck end to end, and the size of the delay of this link is orientated the size that postpones bottleneck as.

For postponing bottleneck, the domestic and international research achievement is all fewer, but along with the exigent application of real-time such as speech communication network, video stream media network popularize gradually and popular, the research association that postpones bottleneck obtains healing and comes many more concerns.Existing main research at present is as follows: J.C.Bolot utilizes the mode that periodically sends the UDP bag to measure round-trip delay, has analyzed the packet delay end to end among the Internet and has lost behavior.S.B.Moon was studying the correlation between the packet delay and packet loss in a continuous Media Stream in 1998.People such as G.Almes did detailed introduction to one-way latency and round-trip delay notion and method of measurement respectively in 1999.

Application number is the assay method that 200710074656.8 Chinese patent application discloses a kind of multimedia terminal audio frequency delay, can find the audio frequency delay bottleneck.Concrete steps comprise: 1. finish the multimedia terminal and with it behind the network design of the media device of intercommunication, the test point and the outside monitoring point of inside, described multimedia terminal is set; 2. described multimedia terminal mutual steadily after, extract each test point of inside, described multimedia terminal and the voice data of outside each monitoring point; 3. read each test point of inside, described multimedia terminal and the voice data of outside each monitoring point respectively, and make audio frequency delay according to described voice data respectively and estimate.

3. packet loss bottleneck detection technique

Packet loss is the key factor that embodies network link performance, and it is directly connected to the direct feel that the user uses network.Network bottleneck to packet loss has had certain research at present, and especially the research for the packet loss bottleneck context of detection in the UDP network has had very big progress.

A kind of leading technology that detects the packet loss bottleneck at present is to detect network bottleneck by the mathematic programming methods of utilizing that people such as N.Shetty proposed in 2008.Under the prerequisite of capacity, network topology structure and the path input rate of known every link, construct the maximization of utility problem of geometric programming form, bottleneck detection problem is converted into the overall planning model problem that to obtain unique solution by mathematical modeling, location packet loss bottleneck, this method also is applicable to the network condition of input rate at random.

Introduce the defective of above-mentioned prior art below.

1. bandwidth bottleneck and postpone the problem of bottleneck detection technique

(1) to a certain extent network is caused load

For the measurement of bandwidth and delay, prior art for example application number is that 200310113676.3 patent application need send detection packet in network, and by carry out the correlation detection data of just can obtaining a result at transmitting terminal and receiving terminal.This has just caused inevitable offered load to network, not only the network operation is impacted, and also can make measurement result produce deviation, reaction network bottleneck situation really.

(2) real-time is poor

The measurement of bandwidth bottleneck and delay bottleneck all needs to expend certain measurement duration, this time mainly comprises the time that detection packet sends and reception, the time of handling and data post analysis are handled, and for example application number is 200710074656.8 patent application.Measure the real-time that longer duration can directly have influence on measurement result.

2. the problem of packet loss bottleneck detection technique

(1) network environment is considered not comprehensive

Network is made of network node and network link two large divisions, and the applicable models of existing packet loss measuring technique is merely regarded network as the set of link, ignored the key effect of network node in network, unilateral like this model will inevitably cause the deviation of measurement result and truth, makes measurement result undesirable.

(2) lack practical application

Packet loss is as a newer bottleneck research field, and the technology of current this aspect just realizes on Mathematical Modeling and calculates, not practical application in real network.And the network in the reality often exists the uncertain factor outside many ideals.It is far from being enough that any technology only builds on the pure Fundamentals of Mathematics, only is applied to could really verify in the real network its correctness and reliability, could provide more foundation for method improvement.

Summary of the invention

(1) technical problem that will solve

The technical problem to be solved in the present invention is: how to improve the accuracy that network bottleneck detects.

(2) technical scheme

For solving the problems of the technologies described above, the invention provides a kind of network bottleneck detection method, may further comprise the steps:

101, set up Mathematical Modeling according to network topology structure;

102,, determine the target function of geometric programming problem according to network maximization of utility principle;

103, whether the input rate of judging all paths all fixes, if, then the order execution in step 104,107,108; Otherwise order execution in step 105,106,107,108;

104, according to the input rate in all paths all fixedly the time, the speed summation of all data of every the link of flowing through is not more than the principle of the queueing capacity sum in link capacity and the switching node, determines the constraints of fixed rate;

105, read set safety feature value ε;

106, determine the constraints of speed at random;

107, find the solution the geometric programming problem, to draw the packet loss of every link, wherein, described geometric programming problem by described target function and fixed rate or at random the constraints of speed constitute;

108, according to the packet loss of every link being obtained, determine network bottleneck.

Preferably, step 101 specifically comprises: all link packet drop rate set are designated as

Wherein, L is the link sum, and the packet loss of link i to be measured is l _i, corresponding link i percent of pass is k _iEvery link link corresponding capacity set is designated as

The capacity set of the network node of link initiating terminal correspondence is designated as

W _iThe capacity of the network node of expression link i initiating terminal correspondence; Obtain data flow in the network with traffic probe, with the one group link set of path representation data flow from the input node to the output node process; Make R represent the number in path, r ₁, r ₂..., r _RThe initial input speed of representing these paths; The data that get access to are expressed as link-path matrix A=[a _Ij] _{L * R}, a wherein _Ij=m represents that link i is that path j goes up the m bar link of counting from the input node.

Preferably, in the step 102, the target function of determined geometric programming problem is as follows:

$\mathrm{Maximize}{\mathrm{\Π}}_{i=1}^L{k}_i^{{\mathrm{\α}}_i}$

Wherein, k _iBe the percent of pass of link i, α _iBe the parameter corresponding to link i, α _iValue determine that according to network topology and path the position of link i in all paths is the closer to the then corresponding α of initiating terminal _iBig more.

Preferably, determine the parameter alpha of all link i _iThe step of value as follows: in the upstream link in all paths in the selected network at first, with the parameter alpha of this link _iBe made as link sum L, and remove this link, successively L-1, L-2 are decremented to 1 then always, the value of gained is given α successively _i

Preferably, in the step 104, the constraints of determined fixed rate is shown below:

$s.t.{\mathrm{\Σ}}_{j\∈i}{k}_i{r}_j^i\≤C_i+W_i,\∀i\∈\mathrm{1,2},...,L$

k _i≤1，k _i＞0

K wherein _iBe the percent of pass of link i, Speed when flowing into link i for the data flow on the j of path is defined as:

$r_j^i=r_j{k}_{{\left(1\right)}_j}{k}_{{\left(2\right)}_j}...k_{{(m-1)}_j}$

Wherein, (1) _j, (2) _j... (m-1) _jBe the go forward numbering of m-1 bar link and a of path j _Ij=m.

Preferably, described safety feature value ε value is 0～1.

Preferably, in the step 106, the constraints of determined speed at random is shown below:

$s.t.{\mathrm{\Σ}}_{j\∈i}{\mathrm{\μ}}_j{k}_i^j+\sqrt{{\mathrm{\Σ}}_{j\∈i}{\mathrm{\σ}}_j^2{k}_i^{j^2}}{Q}^{-1}(1-\mathrm{\ϵ})\≤C_i+W_i,\∀i\∈\mathrm{1,2},...,L$

0＜k _i≤1

Wherein, r _jNormal Distribution

μ _jBe r _jAverage,

Be variance, Q is μ for (1-ε) corresponding average that satisfies _j, variance is

The accumulated probability density value of normal distribution, $k_i^j=k_i{k}_{{\left(1\right)}_j}{k}_{{\left(2\right)}_j}...k_{{(m-1)}_j},{\left(1\right)}_j,{\left(2\right)}_j,...{(m-1)}_j$ Be the go forward numbering of m-1 bar link of path j,

The percent of pass that is all links from path j to the required process of link i is long-pending, C _iExpression link i link corresponding capacity, W _iThe capacity of the network node of expression link i initiating terminal correspondence.

Preferably, in the step 107, utilize mathematical optimization software to find the solution the geometric programming problem, to draw the packet loss of every link.

(3) beneficial effect

Beneficial effect of the present invention is:

1) offered load is low: after obtaining network parameter and topology, the mathematic programming methods that places one's entire reliance upon solves the packet loss computational problem, can not send extra detection packet in network, has greatly reduced offered load;

2) accuracy height: the designed Mathematical Modeling of the present invention is regarded network the set of link and node as, and link and node all have the capacity of oneself simultaneously, more near live network, thereby have guaranteed that result of calculation has higher accuracy; And by determining target function and constraints, the present invention is converted into the geometric programming problem with the problem of finding the solution of link packet drop rate, thereby can utilize existing mathematical optimization software efficiently to solve, and the result is more accurate and consuming time shorter;

3) network management flexibility ratio height: for the network environment of input rate random distribution, the safety feature of network management is taken into account finding the solution of geometric programming problem as parameter, thereby provide higher flexibility ratio for network manager.

Description of drawings

Fig. 1 is a method flow diagram of the present invention;

Fig. 2 determines parameter alpha in the step 102 of Fig. 1 _iFlow chart;

The network topological diagram that Fig. 3 is in the embodiment of the invention to be supposed.

Embodiment

Below in conjunction with drawings and Examples, the specific embodiment of the present invention is described in further detail.Following examples are used to illustrate the present invention, but are not used for limiting the scope of the invention.

The present invention determines network bottleneck by calculating link packet drop rate.As shown in Figure 1, method of the present invention comprises the steps:

101), set up corresponding Mathematical Modeling according to network topology structure at existing network.

All link packet drop rate set are designated as

Wherein, L is the link sum, and the packet loss of link i to be measured is l _i, corresponding link i percent of pass is k _iEvery the set of link link corresponding capacity is designated as In not having the network of annular data flow, the flow direction of all data is consistent on same link, so the capacity set of the network node of link initiating terminal correspondence is designated as

W _iThe capacity of the network node of expression link i initiating terminal correspondence.Obtain data flow in the network with traffic probe, with the one group link set of " path " expression data flow from the input node to the output node process.Make R represent the number in path, r ₁, r ₂..., r _RThe initial input speed of representing these paths.The data that get access to are expressed as link-path matrix A=[a _Ij] _{L * R}, a wherein _Ij=m represents that link i is that path j is upward counted m bar link from ingress, does not comprise link i as if in the set of path j, then a _Ij=0.

102), determine the target function of geometric programming problem according to network maximization of utility principle.

The present invention is converted into network maximization of utility (Maximize) problem with the packet loss computational problem, by finding the solution geometric programming problem detection network bottleneck.The target function of this geometric programming is as follows:

$\mathrm{Maximize}{\mathrm{\Π}}_{i=1}^L{k}_i^{{\mathrm{\α}}_i}$

This target function is considered as utility function with the power exponent form of link percent of pass, and percent of pass and the importance of this link in network of taking all factors into consideration every link obtain final network effectiveness.Wherein, k _iBe the percent of pass of link i, α _iBe the parameter corresponding to link i, α _iValue relevant with network topology and path, the position of link i in all paths is the closer to the then corresponding α of initiating terminal _iBig more.

Determine all α _iThe concrete steps of value as shown in Figure 2, at first in the selected network one in the upstream link in all paths (promptly for each paths that comprises this link, all other links that will pass through all are the downstreams of this link, and described upstream link has one at least), with the parameter alpha of this link _iBe made as link sum L, and remove this link, have at least one link to be in the upstream in all paths in the network link that can find to be left again, successively L-1, L-2 are decremented to 1 always, and give α successively _i

103) whether the input rate of judging all paths all fixes, if, then the order execution in step 104,107,108; Otherwise order execution in step 105,106,107,108.

104) according to the input rate in all paths all fixedly the time, the speed summation of all data of every the link of flowing through is not more than the principle of the queueing capacity sum in link capacity and the switching node, determines the constraints of fixed rate.

The input rate in all paths is all fixedly the time, the speed summation of all data of every link of flowing through should be not more than the queueing capacity sum in link capacity and the switching node, and the percent of pass span of link (0,1] in the interval, so the constraints of described fixed rate is shown below:

$s.t.{\mathrm{\Σ}}_{j\∈i}{k}_i{r}_j^i\≤C_i+W_i,\∀i\∈\mathrm{1,2},...,L$

k _i≤1，k _i＞0

K wherein _iBe the percent of pass of link i, Speed when flowing into link i for the data flow on the j of path is defined as:

$r_j^i=r_j{k}_{{\left(1\right)}_j}{k}_{{\left(2\right)}_j}...k_{{(m-1)}_j}$

Wherein (1) _j, (2) _j... (m-1) _jBe the go forward numbering of m-1 bar link and a of path j _IjA among the=m, the link-route matrix that is to say that and if only if _YjDuring=x, (x) _j=y, the numerical tabular in link-route matrix is understood the ordinal position x of link y in the j of path, promptly link y is the x bar link on the j of path.

105) read the default safety feature value ε of network manager.

If the input rate in all paths all is a random distribution, and during on-fixed, i.e. the initial input speed r of path j _jNormal Distribution

The time, for the input rate in a part of path fix, the situation of the input rate random distribution in a part of path, the fixing path of input rate can be considered as Normal Distribution N (μ _j, 0).At first read the ε of network manager input this moment, and ε is the safety feature of network management, and value is 0～1, and network manager can come the Control Network management to reach which type of safety feature by different ε is set.The situation of ε=0 is corresponding to being in the way to manage of insuring most that worst case is taked at network, ε=0.5 is in average case corresponding to network.

106) determine the constraints of speed at random.

For every link i, the constraints of speed is shown below at random:

$s.t.{\mathrm{\Σ}}_{j\∈i}{\mathrm{\μ}}_j{k}_i^j+\sqrt{{\mathrm{\Σ}}_{j\∈i}{\mathrm{\σ}}_j^2{k}_i^{j^2}}{Q}^{-1}(1-\mathrm{\ϵ})\≤C_i+W_i,\∀i\∈\mathrm{1,2},...,L$

0＜k _i≤1

μ wherein _jBe r _jAverage,

Be variance.Q is μ for (1-ε) corresponding average that satisfies _j, variance is

The accumulated probability density value of normal distribution.

Be the go forward numbering of m-1 bar link of path j,

The percent of pass that is all links from path j to the required process of link i is long-pending, C _iExpression link i link corresponding capacity, W _iThe capacity of the network node of expression link i initiating terminal correspondence.

107) find the solution the geometric programming problem.

Above by the constraints (104) of structure target function (102), fixed rate or the constraints of speed (106) at random, obtain a maximized geometric programming problem, next utilized mathematical optimization software can solve the packet loss of every link.Described mathematical optimization software is the mathematical optimization instrument of (for example) Matlab, bag is as GGPLAB, utilize this kit defined variable (being the percent of pass of link herein), and input target function and constraints can calculate percent of pass, also just calculate the packet loss of link.

108) according to the packet loss of every link obtaining, determine network bottleneck.

Embodiment

Part topological structure with an actual UDP network is the advantage that example shows network bottleneck detection method of the present invention below.Network topology structure as shown in Figure 3.This network does not have annular data flow, network node according to the difference of output port respectively with independently oval expression, link is represented with rectangle, the capacity of network node is designated as W, the capacity of link is designated as C, represent the path of data flow process 5 paths are arranged in the present embodiment with the solid line of band arrow points, the initial input speed in each path is designated as r _j

Detailed step is as follows:

101) set up Mathematical Modeling

Obtain network topology and link, routing iinformation by Network Measurement Technologies, all link packet drop rates set are designated as L={l ₁, l ₂, l ₃, l ₄, l ₅, every link link corresponding capacity set is designated as C={C ₁, C ₂, C ₃, C ₄, C ₅}={ 1.6,2,4.5,2,2}, the capacity set of the network node of link initiating terminal correspondence is designated as W={W ₁, W ₂, W ₃, W ₄, W ₅}={ 1,1,1,1,1}.Know the number R=5 in path, the initial input speed r in path by flow measurement ₁, r ₂, r ₃, r ₄, r ₅Equal Normal Distribution N (1,0.1), then link-path matrix is

$A={\left[a_{\mathrm{ij}}\right]}_{5\×5}=\left[\begin{array}{ccccc}1& 1& 0& 0& 0\\ 0& 0& 1& 1& 0\\ 2& 2& 2& 2& 1\\ 0& 3& 3& 0& 0\\ 3& 0& 0& 3& 2\end{array}\right]$

102) determine target function

Substitution target function formula, the target function that obtains geometric programming is as follows:

$\mathrm{Maximize}{\mathrm{\Π}}_{i=1}^5{k}_i^{{\mathrm{\α}}_i}$

According to definite parameter alpha _iMethod find suitable α _iValue.If N is 5, at first in the selected network wherein one of the upstream link in all paths, promptly choose link l ₁, with α ₁Being changed to N is 5, removes link l ₁And N subtracted 1, choose the upstream link in the remaining network link more successively, and with the α of correspondence _iBe changed to N.Finally obtain α _iValue set=5,4,3,2,1}.

103-106) determine constraints.Detailed step comprises:

(1) judges that input rate is fixed value or random value.In the present embodiment, suppose through judging that input rate as can be known is a random distribution, and Normal Distribution N (1,0.1), then enter the situation of input rate at random, determine the constraints of speed at random then in the step below.

(2) read the ε that network manager is imported, determine the constraints of speed at random

For the ease of observing the method performance under the various management safety feature, selected a plurality of ε, these ε values are to serve as 5 numerical value that the interval is chosen with 0.2 from 0.1 to 0.9 interval, i.e. the set of ε value be ε=0.1,0.3,0.5,0.7,0.9}.

Next determine the constraints of speed at random, obtain for the constraints of link i as follows according to formula:

${\mathrm{\Σ}}_{j\∈i}{k}_i^j+\sqrt{{\mathrm{\Σ}}_{j\∈i}{0.1k}_i^{j2}}{Q}^{-1}(1-\mathrm{\ϵ})\≤C_i+W_i$

Wherein Q is the accumulated probability density value of (1-ε) corresponding standardized normal distribution, i.e. F (1-ε)=Q. $k_i^j=k_l{k}_{{\left(1\right)}_j}{k}_{{\left(2\right)}_j}...k_{{(m-1)}_j},{\left(1\right)}_j,{\left(2\right)}_j,...{(m-1)}_j$ Be go forward m-1 bar link and a of path j _Ij=m, a that is to say that and if only if _YjDuring=x, (x) _j=y.

107) separate the geometric programming problem

By above-mentioned steps, it is as follows to obtain the geometric programming problem:

${\mathrm{Maximize\Π}}_{i=1}^5{k}_i^{{\mathrm{\α}}_i}$

$s.t.{\mathrm{\Σ}}_{j\∈i}{k}_i^j+\sqrt{{\mathrm{\Σ}}_{j\∈i}{0.1k}_i^{j2}}{Q}^{-1}(1-\mathrm{\ϵ})\≤C_i+W_i,i\∈L$

0＜k _i≤1

Utilize mathematical optimization software to separate this geometric programming problem, can obtain result data efficiently, corresponding to different safety feature values, the percent of pass result of calculation of each link is as shown in table 1 below:

Table 1

108) determine bottleneck.Calculated data by table 1 as can be known, link l ₅The percent of pass minimum, i.e. packet loss maximum is so the decidable testing result is: link l ₅Bottleneck for this network.

As can be seen from the above embodiments, the present invention can improve accuracy, the reduction offered load that network bottleneck detects, and improves the flexibility ratio of network management.

The above only is a preferred implementation of the present invention; should be pointed out that for those skilled in the art, under the prerequisite that does not break away 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 (8)

1. a network bottleneck detection method is characterized in that, may further comprise the steps:

101, set up Mathematical Modeling according to network topology structure;

102,, determine the target function of geometric programming problem according to network maximization of utility principle;

103, whether the input rate of judging all paths all fixes, if, then the order execution in step 104,107,108; Otherwise order execution in step 105,106,107,108;

104, according to the input rate in all paths all fixedly the time, the speed summation of all data of every the link of flowing through is not more than the principle of the queueing capacity sum in link capacity and the switching node, determines the constraints of fixed rate;

105, read set safety feature value ε;

106, determine the constraints of speed at random;

107, find the solution the geometric programming problem, to draw the packet loss of every link, wherein, described geometric programming problem by described target function and fixed rate or at random the constraints of speed constitute;

108, according to the packet loss of every link being obtained, determine network bottleneck.

2. the method for claim 1 is characterized in that, step 101 specifically comprises: all link packet drop rate set are designated as Wherein, L is the link sum, and the packet loss of link i to be measured is l _i, corresponding link i percent of pass is k _iEvery link link corresponding capacity set is designated as

The capacity set of the network node of link initiating terminal correspondence is designated as

W _iThe capacity of the network node of expression link i initiating terminal correspondence; Obtain data flow in the network with traffic probe, with the one group link set of path representation data flow from the input node to the output node process; Make R represent the number in path, r ₁, r ₂..., r _RThe initial input speed of representing these paths; The data that get access to are expressed as link-path matrix A=[a _Ij] _{L * R}, a wherein _Ij=m represents that link i is that path j goes up the m bar link of counting from the input node.

3. method as claimed in claim 2 is characterized in that, in the step 102, the target function of determined geometric programming problem is as follows:

$\mathrm{Maximize}{\mathrm{\Π}}_{i=1}^L{k}_i^{{\mathrm{\α}}_i}$

Wherein, k _iBe the percent of pass of link i, α _iBe the parameter corresponding to link i, α _iValue determine that according to network topology and path the position of link i in all paths is the closer to the then corresponding α of initiating terminal _iBig more.

4. method as claimed in claim 3 is characterized in that, determines the parameter alpha of all link i _iThe step of value as follows: in the upstream link in all paths in the selected network at first, with the parameter alpha of this link _iBe made as link sum L, and remove this link, successively L-1, L-2 are decremented to 1 then always, the value of gained is given α successively _i

5. as claim 3 or 4 described methods, it is characterized in that in the step 104, the constraints of determined fixed rate is shown below:

$s.t.{\mathrm{\Σ}}_{j\∈i}{k}_i{r}_j^i\≤C_i+W_i,\∀i\∈\mathrm{1,2},...,L$

k _i≤1，k _i＞0

K wherein _iBe the percent of pass of link i,

Speed when flowing into link i for the data flow on the j of path is defined as:

$r_j^i=r_j{k}_{{\left(1\right)}_j}{k}_{{\left(2\right)}_j}...k_{{(m-1)}_j}$

Wherein, (1) _j, (2) _j... (m-1) _jBe the go forward numbering of m-1 bar link and a of path j _Ij=m.

6. as claim 3 or 4 described methods, it is characterized in that described safety feature value ε value is 0～1.

7. method as claimed in claim 6 is characterized in that, in the step 106, the constraints of determined speed at random is shown below:

$s.t.{\mathrm{\Σ}}_{j\∈i}{\mathrm{\μ}}_j{k}_i^j+\sqrt{{\mathrm{\Σ}}_{j\∈i}{\mathrm{\σ}}_j^2{k}_i^{j^2}}{Q}^{-1}(1-\mathrm{\ϵ})\≤C_i+W_i,\∀i\∈\mathrm{1,2},...,L$

0＜k _i≤1

Wherein, r _jNormal Distribution

μ _jBe r _jAverage,

Be variance, Q is μ for (1-ε) corresponding average that satisfies _j, variance is

The accumulated probability density value of normal distribution, $k_i^j=k_i{k}_{{\left(1\right)}_j}{k}_{{\left(2\right)}_j}...k_{{(m-1)}_j},{\left(1\right)}_j,{\left(2\right)}_j,...{(m-1)}_j$ Be the go forward numbering of m-1 bar link of path j,

The percent of pass that is all links from path j to the required process of link i is long-pending, C _iExpression link i link corresponding capacity, W _iThe capacity of the network node of expression link i initiating terminal correspondence.

8. as each described method in the claim 1～4, it is characterized in that, in the step 107, utilize mathematical optimization software to find the solution the geometric programming problem, to draw the packet loss of every link.

Images