CN101599870B - Network link performance measurement method - Google Patents

Abstract

The invention discloses a network link performance measurement method. A uniform performance analyzing linear model is established by analyzing the characteristics of network link performance; in the linear model, when rank of a route matrix is not equal to that of an augmented matrix and optimal solution of nonhomogeneous linear equation system can not be obtained, the problem is transformed to a multiobjective optimal problem by mathematic transformation, and genetic algorithm is used for solving sub-optimal solution of the nonhomogeneous linear equation system; and finally the network link performance is obtained by statistic probability distribution of sub-optimal solution in discretization interval. The invention uniforms the network link performance analyzing model as a linear analyzing model, transforms the link performance measurement problem to the multiobjective optimal problem, utilizes the genetic algorithm to solve the sub-optimal solution of the interior network link performance, and obtains link performance parameters by analyzing, such as link delay time and loss rate.

Description

Network link performance measurement method

Technical field

The present invention relates to a kind of network for formance measuring method, particularly network link performance measurement method.

Background technology

The network-external measuring technique is to propose a kind of new Network Measurement Technologies in recent years in the world, CT thought medically is incorporated in the network measure, analyzes and performance parameter and topological structures such as the message dropping rate of Measurement Network inner link and time-delay according to the measurement of network-external boundary node.Document " Network loss tomography using striped unicast probes; IEEE/ACM Transactions on Networking; 2006, Vol.14 (4), p697-710 " discloses a kind of loss of link rate estimation method of measuring based on end to end network.This method at first adopts a tree network topological structure, utilizes root node initiatively to send " bag string " as measured message; At leaf node measured message is received then; The partial ordering relation that receives according to measured message is inferred the network internal link performance at last.In fact, in the document, loss of link rate analytical model is the partial ordering relation that measured message receives, but this relation and be not suitable for network link supposition time of delay and analysis.So infer that based on the networking inner link performance of end-to-end measurement the problem that exists is now: the performance analysis models that neither one is unified is analyzed and Measurement Network inner link performance simultaneously, as link delay time and Loss Rate.

Summary of the invention

Can't utilize a unified performance analysis models to come the deficiency of Measurement Network inner link performance in order to overcome prior art, the invention provides a kind of network link performance measurement method, by phase-split network link performance characteristics, set up unified performance evaluation line style model; In this line style model, when the order of route matrix is not equal to the order of its augmented matrix, under the situation that can't obtain the Linear Equations optimal solution, change by mathematics, problem is converted into a multi-objective optimization question, and utilizes genetic algorithm to ask the suboptimal solution of Linear Equations; At last, distribute in the discretization interval probability and obtain network link performance by the statistics suboptimal solution.

The technical solution adopted for the present invention to solve the technical problems: a kind of network link performance measurement method is characterized in may further comprise the steps:

(a) on phase-split network loss of link rate and time of delay and path Loss Rate and time of delay basis, set up the unified linear model that network link performance is analyzed, Y=AX;

In the formula, Y is the end to end performance parameter, is a vectorial transposition of row; A is a route matrix; X is the link performance parameters column vector that will find the solution;

This linear model represents that it is R (A)=R (A|Y) that nonhomogeneous equation has the precondition of separating, and promptly the order of route matrix A must equal the order of its augmented matrix (A|Y);

(b) if R (A) ≠ R (A|Y), promptly the order of route matrix A is not equal to the order of its augmented matrix (A|Y), then

min?f ₁＝|a _1，1x ₁+a _1，2x ₂+...+a _1，mx _m-Y ₁?|

min?f ₂＝|y _2，1x ₁+y _2，2x ₂+...+y _2，mx _m-Y ₂|

……

min?f _n＝|y _n，1x ₁+y _n，2x ₂+...+y _n，mx _m-Y _n|

x ₁∈[Z _1，1，Z _1，2]，x ₂∈[Z _2，1，Z _2，2]，...，x _m∈[Z _m，1，Z _m，2]

Utilize genetic algorithm to try to achieve the suboptimal solution of multiple-objection optimization;

(c) in order to overcome " hamming steep cliff (Hamming Cliff) " problem that traditional binary coding is brought, adopt Gray code, fitness function is defined as:

$F\left(x\right)=\frac{1}{f\left(x\right)}=\frac{1}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_i{f}_i\left(x\right)};$

(d) by to the solution space discretization, add up the situation that suboptimal solution drops on the discretization solution space, obtain the link performance probability distribution graph; Select the center numerical value in the corresponding interval of maximum institute in the probability distribution graph, i.e. loss of link rate or link delay time.

The invention has the beneficial effects as follows: with the unification of network link performance analytical model is a linear analysis model, transform by mathematics, link performance measurement problem is changed into a multi-objective optimization question, utilize genetic algorithm to try to achieve the suboptimal solution of network internal link performance, and obtain link performance parameters by statistical analysis, as link delay time and Loss Rate.

Below in conjunction with drawings and Examples the present invention is elaborated.

Description of drawings

Fig. 1 is the network measure model that network link performance measurement method of the present invention is set up.

Fig. 2 is the flow chart of network link performance measurement method of the present invention.

Embodiment

With reference to Fig. 1～2, concrete steps are as follows: at first, set up a unified network link performance analytical model.

In the network measure model, free routing P _{I, j}Between message transmissions postpone each link delay time sum equal to be formed, this relation is called network link delay time analysis model, can be expressed as formula (1).

D _i＝d _i，1+d _i，2+...+d _i，j (1)

In the network measure engineering, free routing P _{I, j}The product of each link success transfer rate that message success transfer rate equals to be formed can be expressed as:

s _i＝η _i，1×η _i，2×...×η _i，j (2)

S wherein _iBe i path success transfer rate, η _{I, k}(1≤k≤j) for forming the link success transfer rate in this path.Taken the logarithm respectively (Lg) in formula (2) both sides, then have:

Lg(s _i)＝Lg(η _i，1)+Lg(η _i，2)+...+Lg(η _i，j) (3)

From formula (3) as can be seen, by after the variation of logarithm mathematics, also there is linear relationship between path and the link success transfer rate; As long as can calculate link success transfer rate s, loss of link rate α can obtain by α=1-s.

There are linear relationship in aggregative formula (1) and (3) between network link and the path performance, unified being expressed as:

Y _i＝y _i，1+y _i，2+...+y _i，j (4)

In conjunction with Fig. 1 network measure model, then have again:

Y＝AX (5)

In (5) formula, Y is the end to end performance parameter, is a vectorial transposition of row, and A is a route matrix, the link performance parameters column vector of X for finding the solution, and the present invention is called network link performance with (5) formula and analyzes linear model.It not only is suitable for and link delay time series analysis, and is applicable to the analysis of loss of link rate.It should be noted that for Loss Rate on linear model based, to have only, just can obtain final goal and separate by after separating again of being asked transformed through index.

Secondly, Linear Equations is found the solution changed into multi-objective optimization question.

(5) precondition of separating is arranged is R (A)=R (A|Y) to formula, and promptly the order of route matrix A must equal the order of its augmented matrix (A|Y).If R (A) ≠ R (A|Y) at this moment utilizes the linear algebra method to find the solution (5).If (4) formula is rewritten as following form: f _i=| y _{I, 1}+ y _{I, 2}+ ...+y _{I, j}-Y _i|, then (5) formula can be rewritten as (6) formula.

min?f ₁＝|a _1，1x ₁+a _1，2x ₂+...+a _1，mx _m-Y ₁|

min?f ₂＝|y _2，1x ₁+y _2，2x ₂+...+y _2，mx _m-Y ₂|

……

min?f _n＝|y _n，1x ₁+y _n，2x ₂+...+y _n，mx _m-Y _n|

x ₁∈[Z _1，1，Z _1，2]，x ₂∈[Z _2，1，Z _2，2]，...，x _m∈[Z _m，1，Z _m，2] (6)

In (6) formula, [Z _{I, 1}, Z _{I, 2}] (1≤i≤m) represents each decision variable (link performance that will measure) constant interval.So just will be converted into multi-objective optimization question, can try to achieve the suboptimal solution of multi-objective problem by genetic algorithm based on the link performance measurement problem of end-to-end measurement.

In the 3rd step, utilize genetic algorithm for solving linear analysis model suboptimal solution.

Coding: the present invention adopts Gray code (Gray Code), because the hamming distance is 1 between the adjacent integer, can overcome " hamming steep cliff " (Hamming Cliff) problem that traditional binary coding is brought, very big hamming distance is promptly arranged between the binary code of some adjacent integer.

Fitness function: the weight coefficient converter technique is adopted in the establishment of fitness function, promptly according to the sub-goal function f _i(x) (i=1,2 ..., n) significance level in multi-objective optimization question is given weights omega _i, then the target function of multi-objective optimization question is each sub-goal function f _i(x) linear weighted function and, can be expressed as shown in the formula (7).

$f\left(x\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_i{f}_i\left(x\right)---\left(7\right)$

In link performance measurement problem and since each sub-goal function to measurement result to influence significance level identical, so get ω ₁=ω ₂=...=ω _n=1/n.For this reason, fitness function may be defined as formula (8):

$F\left(x\right)=\frac{1}{f\left(x\right)}=\frac{1}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_i{f}_i\left(x\right)}---\left(8\right)$

If with the evaluation function of F (x) as multi-objective optimization question, then multi-objective optimization question can be converted into the single goal optimization problem, then can utilize the single goal Genetic Algorithms for Optimization to find the solution multi-objective optimization question.

Initial population: the present invention is by adjusting the initial population that generates at random under the little population scale, reject and repeat individuality, adjust the frequency that gene occurs on each locus, individuality is evenly distributed on the whole solution space, as far as possible to reach the diversified purpose of initial population gene.In the link performance method of measurement based on end-to-end measurement, the initial population that evenly distributes generating algorithm is as follows.

Step1: generate initial population at random;

Step2: calculate in the initial population between any two individualities hamming apart from γ _iIf, γ _i＜θ then gets and removes wherein any one; Otherwise two individualities all keep.

Step3: add in the population Goto Step2 if individual amount, produces several body at random less than the demand of population.

Step4: the composition that detects each locus in the population; If 0 at the locus proportion greater than 50%, then select part 0 genetic mutation to become 1 at random, making 0 proportion is 50%, otherwise selects part 1 genetic mutation to become 0 at random, making 0 proportion is 50%;

Step5: initial population generates and finishes.

It is 1000 that the present invention selects the initial population number for use, and stopping algebraic degree is 80.

Genetic manipulation.

Select operator: in the multi-objective optimization question solution procedure of measuring based on link performance, come the individuality in the population is carried out the natural selection operation, make the higher individuality of fitness to be genetic to colony of future generation with bigger probability by selecting operator.Select the concrete algorithm of operator to be achieved as follows:

Step1: initial population is divided into the x group in proper order according to front and back, and x is the number of sub-goal function.

Step2: as follows to the individual processing of each group:

(1) is the individuality of i for the group number, utilizes i sub-target function to calculate its functional value;

(2) with certain probability P _gChoose several individualities according to the sub-goal functional value from i group individuality, the sub-goal functional value is more little, illustrates that this individual adaptability is strong more, and the possibility of then choosing is big more.

(3) i++, if i≤x, Goto (1), otherwise Goto (Step3).

Step3: the x that chooses is organized according to sequential combination before and after the group number, constitute the new population of this iteration.

Crossover operator: the present invention adopts single-point to intersect (One-point Crossover), promptly in population according to certain probability P _cSelect the chromosome of two pairings at random, and produce a crosspoint at random, exchange two portion gene groups that pairing is individual mutually at this point.When the operation crossover operator,, require to have only when Hamming distance surpasses certain threshold value between the individuality that participates in intersection, just allow to carry out between the two crossing operation in order to prevent inbreeding.Initial threshold value can adopt the mean value of initial population hamming distance, and it reduces along with the increase of iterations.

Mutation operator: mutation operator replaces original genic value with new genic value, changes the individual chromosome structure, produces new chromosome, to improve the diversity of population, precocity is had certain inhibitory action.The variation probability that the present invention chooses population is: P _m=0.005.Each individual coding according to the variation probability, is specified at random that value morphs on some locus, and promptly " 0 " becomes " 1 " or " 1 " becomes " 0 ".

The 4th step, statistical analysis link performance probability distribution.

In the link delay measuring process, the decision variable minimum value is 0, and maximum is the minimum value that the path network delay performance is arranged under this link in the one-shot measurement, is labeled as α; Then the solution vector space of this link delay time can be expressed as [0, α].And in the Loss Rate measuring process, the present invention supposes the successful transfer rate of every link between 10% and 100%, owing in measuring process, to take down percentage sign %, and to carry out logarithm operation one time, so the solution vector space of each bar link success transfer rate can be expressed as [1,2].According to certainty of measurement, be the minizone of a plurality of discretizations with solution vector space equal portions; Repeatedly carry out genetic algorithm then, can obtain the suboptimal solution of many group link performances, each suboptimal solution being dropped on the situation in discretization interval adds up, can obtain the probability distribution graph of link performance, its maximum probability the interval central value of corresponding discretization can be expressed as the value of this link performance parameters, also be the value of the link performance that will measure of the present invention.

Claims (1)

1. a network link performance measurement method is characterized in that comprising the steps:

(a) on phase-split network loss of link rate and time of delay and path Loss Rate and time of delay basis, set up the unified linear model that network link performance is analyzed, Y=AX;

In the formula, Y is the end to end performance parameter, is a vectorial transposition of row; A is a route matrix; X is the link performance parameters column vector that will find the solution;

This linear model represents that it is R (A)=R (A|Y) that nonhomogeneous equation has the precondition of separating, and promptly the order of route matrix A must equal the order of its augmented matrix (A|Y);

(b) if R (A) ≠ R (A|Y), promptly the order of route matrix A is not equal to the order of its augmented matrix (A|Y), then

minf ₁＝|a _1，1x ₁+a _1，2x ₂+...+a _1，mx _m-Y ₁|

minf ₂＝|a _2，1x ₁+a _2，2x ₂+...+a _2，mx _m-Y ₂|

......

minf _n＝|a _n，1x ₁+a _n，2x ₂+...+a _n，mx _m-Y _n|

x ₁∈[Z _1，1，Z _1，2]，x ₂∈[Z _2，1，Z _2，2]，...，x _m∈[Z _m，1，Z _m，2]

Utilize genetic algorithm to try to achieve the suboptimal solution of multiple-objection optimization;

(c) in order to overcome " hamming steep cliff (Hamming Cliff) " problem that traditional binary coding is brought, adopt Gray code, fitness function is defined as:

$F\left(x\right)=\frac{1}{f\left(x\right)}=\frac{1}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_i{f}_i\left(x\right)};$

In the formula, ω _i(i=1...n) expression weight, ω ₁=ω ₂=...=ω _n=1/n, n are measured path numbers;

(d) by to the solution space discretization, add up the situation that suboptimal solution drops on the discretization solution space, obtain the link performance probability distribution graph; Select the center numerical value in the corresponding interval of maximum institute in the probability distribution graph, i.e. loss of link rate or link delay time.

Images