CN101610199B - Heterogeneous network packet loss differentiating method based on fuzzy comprehensive judgment - Google Patents

Abstract

The invention provides a heterogeneous network packet loss differentiating method based on fuzzy comprehensive judgment. The method comprises the following steps: firstly, using unidirectional transmission time delay and average packet loss rate of a packet pair detection frame as factor set parameters of the fuzzy comprehensive judgment; during packet loss, working out independent judgment by using the unidirectional transmission time delay and the average packet loss rate respectively to acquire judgment values of the two factor set parameters on the current network condition; calculating dynamic weight distribution functions related with the unidirectional transmission time delay and the average packet loss rate respectively, and carrying out normalized processing for the two dynamic weight distribution functions to determine a relative weight value of the unidirectional transmission time delay and the average packet loss rate; and carrying out fuzzy transformation for the relative weight value to acquire comprehensive judgment so as to differentiate packet loss reasons. The related experiment shows that the method has better implementation effect compared with the prior TCP versions and the prior differentiating method.

Description

Heterogeneous network packet loss differentiating method based on fuzzy comprehensive evoluation

Technical field

The present invention relates to a kind of under heterogeneous network environment; The method that the reason of data packet loss is distinguished; Relate in particular to a kind of under the heterogeneous network environment of Wireless/wired fusion; Based on fuzzy (Fuzzy) Model for Comprehensive, the method to the two kinds of reasons (error code packet loss, congestion packet loss) that cause data-bag lost are effectively distinguished belongs to network service quality (QoS) technical field.

Background technology

Continuous development along with the communication technology; The trend that the coexistence of polytype communication network occurred; As shown in Figure 1, multiple Radio Network Systems such as wireless wide area network (like GPRS, UMTS etc.), WLAN (like IEEE802.11), satellite communication network, blueteeth network just progressively replace traditional cable network to become the final jump that the Internet inserts.For the user who uses wireless network can connect Internet, need make Radio Network System with based on realization fusion between the Int ernet of cable network.The coexistence of this dissimilar communication networks just is called heterogeneous network (heterogeneous network).At present, there's a widespread conviction that realizes the seamless link of miscellaneous heterogeneous networks for people, i.e. the fusion of heterogeneous network is the inexorable trend of future network development.

In the heterogeneous network environment of Wireless/wired fusion, prior art all packet loss simply owing to the blindness severe exacerbation of network congestion the performance of TCP in the wireless network.If it must be the principle that is caused by congested that TCP still follows packet loss, blindly reduce transmission rate, just certainly will reduce bandwidth availability ratio, cause the meaningless deterioration of TCP performance.Therefore, under the heterogeneous network environment that wire/wireless mixes, the control strategy that realize ideal, the packet loss that the packet loss that just must correctly distinguish wireless error code and caused and network congestion are caused, thus take the Different control strategy respectively.

To the problems referred to above, some researchs have been carried out both at home and abroad.Generally speaking, the solution of this respect mainly is divided into two kinds:

1. segment connection method promptly separates the method for monitoring respectively in transport layer with wireless, cable environment.This method is divided into two sections from the base station with transport layer, and the TCP transmit leg has been shielded the error of transmission of Radio Link and the packet loss that causes fully, can use other congested control technology such as SACK etc. in addition to wireless portion.

In segment connection method, owing to carry out transfer of data and affirmation separately, therefore need not to change the TCP realization mechanism of transmitting terminal on the border of cable network and wireless network, just can realize the needed additional functionality of wireless network segment.But this method is in implementation process, and before data did not have the actual recipient of being sent to, the affirmation of forgery just was fed to transmit leg, connects semantic thereby destroyed end-to-end TCP.Therefore, implement segment connection method and often need make material alteration, realize that difficulty is bigger traditional agreement.

From end to end based on the differentiating method of data statistics, promptly data transmission parameters is carried out statistics and analysis, to infer the congested of current network and the state of makeing mistakes at receiving terminal or transmitting terminal.

This differentiating method can be to network model more widely, and is less to requiring of changing of the current agreement of network, thereby more and more receives people's attention.But this differentiating method is through the parameter in the direct Measurement Network transmission course, perhaps estimates network traffic parameter indirectly, carries out statistics and analysis then, and infers the reason of current network packet loss in view of the above.In the practice; The reason that numerical value and variation tendency thereof through certain single parameter is difficult to accurately judge the current network packet loss; Tracing it to its cause is that these parameters often receive the influence of two kinds of lost reason simultaneously, is embodied in the ambiguity that often demonstrates certain intercrossing and probability judge on numerical value and the variation tendency thereof.Therefore, if Radio Link is also made mistakes and caused packet loss in network congestion, then be difficult to judge the reason of packet loss according to the variation that related parameter is arranged.

In two pieces of papers that the inventor delivers on international academic conference ICSNC 2006 (" A ROD based Fuzzy packet loss differentiating algorithm for TCP in the hybrid wired/wireless network " and " End-to-End Differentiation of Congestion and Wireless Losses using a Fuzzy Arithmetic based on Relative Entropy "); Analyzed to the differentiating method that adopts single parameter; When only utilizing ROD (unidirectional transmission time delay) to carry out packet loss differentiating; Under underloaded situation, obtain packet loss differentiating effect preferably; But under the heavier situation of offered load, distinguish accuracy and obviously descend.When only utilizing average packet loss ratio to carry out packet loss differentiating, when the wireless error rate is big, distinguishes accuracy and obviously descend.

According to above-mentioned achievement in research,, use simple, common differentiating method often to be difficult to make clear and definite judgement for the network state of complicacy.

Summary of the invention

It is a kind of under heterogeneous network environment that technical problem to be solved by this invention is to provide, based on fuzzy synthetic evaluation model, and the method that the two kinds of reasons (error code packet loss, congestion packet loss) that cause data-bag lost are effectively distinguished.

For realizing above-mentioned goal of the invention, the present invention adopts following technical scheme:

A kind of heterogeneous network packet loss differentiating method based on fuzzy comprehensive evoluation is characterized in that:

(1) with average packet loss ratio and bag to the unidirectional transmission time delay of explore frame set of factors parameter as fuzzy comprehensive evoluation;

(2) when packet loss takes place, utilize said average packet loss ratio and said unidirectional transmission time delay to make independent judge respectively, obtain the judge value of two said set of factors parameters current network conditions;

(3) calculate the changeable weight distribution function relevant with said unidirectional transmission time delay respectively, and two changeable weight distribution functions are done normalization handle, to confirm the relative weighting value of said average packet loss ratio and said unidirectional transmission time delay with said average packet loss ratio;

(4) said relative weighting value is carried out blurring mapping, obtain multifactorial evaluation, thereby lost reason is distinguished.

Wherein, be provided with set of factors U={x in the said fuzzy comprehensive evoluation ₁, x ₂..., x _n, pass judgment on collection V={y ₁, y ₂..., y _m, each parameter has weight allocation separately in the said set of factors, is designated as: A={a ₁, a ₂..., a _n∈ F (U), wherein a _iBe the weight of i factor among the U, and satisfy

The result of multifactorial evaluation is designated as: B={b for passing judgment on the fuzzy set on the collection V ₁, b ₂..., b _m∈ F (V), wherein b _jRepresent the shared status of j kind comment in passing judgment on target population V.

Said blurring mapping is:

T _R：F(U)→F(V)

Wherein, R is the fuzzy relationship matrix r=(r of U to V _Ij) _{N * m}

In said step (3), the changeable weight distribution function relevant with said average packet loss ratio is:

$W_p=\cos (N({\underset{~}{A}}_c^p,{\underset{~}{A}}_w^p)\·\frac{\mathrm{\π}}{2})$

Where

as congestion losses fuzzy sets

and wireless error loss fuzzy sets

closeness between cells.

In said step (3), the changeable weight distribution function relevant with said unidirectional transmission time delay is:

Wherein $N({\underset{~}{A}}_c^r,{\underset{~}{A}}_w^r)=\mathrm{Exp}(-{(({{\mathrm{\μ}}^r}_c-{{\mathrm{\μ}}^r}_w)/({{\mathrm{\σ}}^r}_c+{{\mathrm{\σ}}^r}_w))}^2).$

In the said step (3), obtain the relative weighting value of said unidirectional transmission time delay according to following formula:

W′ _ROD＝W _ROD/(W _p+W _ROD)，

W′ _p＝W _p/(W _p+W _ROD)

Wherein, W _pBe the changeable weight distribution function relevant, W with said average packet loss ratio _RODBe the changeable weight distribution function relevant, W with said unidirectional transmission time delay _RODBe the relative weighting value of the said unidirectional transmission time delay in normalization processing back, W ' _pHandle the relative weighting value of the said average packet loss ratio in back for normalization.

Heterogeneous network packet loss differentiating method provided by the present invention causes the reason of packet loss creatively to combine for these two kinds to the ROD of explore frame average packet loss ratio and bag; Distribute through constructing the changeable weight that is fit under the different network environments, adopt fuzzy synthetic evaluation model that the heterogeneous network lost reason is distinguished.Relevant experiment shows that this method can obtain better effect under the various network environment.

Description of drawings

Below in conjunction with accompanying drawing and embodiment the present invention is further described.

Fig. 1 is that final jump is the heterogeneous network topological structure sketch map of wireless network;

Fig. 2 is the multifactorial evaluation schematic flow sheet of this heterogeneous network packet loss differentiating method;

Fig. 3 is that the employing fuzzy synthetic evaluation model carries out packet loss differentiating and utilizes ROD to carry out the comparison sketch map of packet loss differentiating, Reno, NewReno throughput simulation result under different topology structure, the different error rate;

Fig. 4 is that the employing fuzzy synthetic evaluation model carries out packet loss differentiating and utilizes average packet loss ratio to carry out the comparison sketch map of packet loss differentiating, Reno, NewReno throughput simulation result under different topology structure, the different error rate.

Embodiment

Before address, for the heterogeneous network environment of complicacy, adopt differentiating method often to be difficult to make right judgement based on single parameter.For this reason; The inventor further thinks on the basis of existing research: for heterogeneous network environment; The material elements that need average packet loss ratio and bag be influenced network service quality to the ROD (unidirectional transmission time delay) of (Packet Pair) explore frame etc. is according to the dynamic different weight allocation of structure of network conditions, thereby the lost reason under the heterogeneous network environment is distinguished accurately.In corresponding analysis and judgement process, need the influence of comprehensive considering various effects, for certain purpose is utilized blurring mapping, thereby final result is made comprehensive judgement.Therefore, above-mentioned problem under heterogeneous network environment, how to distinguish lost reason has significantly fuzzy (Fuzzy) characteristic, can consider that introducing fuzzy synthetic evaluation model handles.

Below, at first the fuzzy synthetic evaluation model that uses among the present invention is described.

Fuzzy synthetic evaluation model is fuzzy multiobjective decision-making (fuzzy multiple goals decision making); Be under fuzzy enviroment; Consider influence of various factors, for certain purpose is utilized blurring mapping, the method for the multifactorial evaluation that affairs are made.

Be provided with set of factors U={x ₁, x ₂..., x _n, pass judgment on collection V={y ₁, y ₂..., y _m, because each parameter is inconsistent to the influence of being passed judgment on affairs in the set of factors, so each parameter should have weight allocation separately, it is a Fuzzy vector on the domain, is designated as: A={a ₁, a ₂..., a _n∈ F (U), wherein a _iBe the weight of i factor among the U, and satisfy

The result of multifactorial evaluation is designated as: B={b for passing judgment on the fuzzy set on the collection V ₁, b ₂..., b _m∈ F (V), wherein b _jRepresent the shared status of j kind comment in passing judgment on target population V.

In order to make up the Fuzzy Multi-Objective Decision model of multifactorial evaluation, the structure blurring mapping

T _R：F(U)→F(V)

Wherein, R is the fuzzy relationship matrix r=(r of U to V _Ij) _{N * m}So (U, V R) constitute fuzzy synthetic evaluation model, and a given weight allocation A is through blurring mapping T by triplet _R, can obtain a multifactorial evaluation B.About further specifying of above-mentioned fuzzy synthetic evaluation model, can consult " fuzzy mathematics method and application thereof " that Xie Jijian and Liu Chengping collaborate (ISBN:7560937950), just do not given unnecessary details in detail at this.

Based on above-mentioned fuzzy synthetic evaluation model, referring to shown in Figure 2, at first select bag to the unidirectional transmission time delay (ROD) of explore frame and average packet loss ratio as the set of factors parameter.When detecting generation packet loss (in the present embodiment, will receive 3 repeat ACKs or overtime being defined as taken place packet loss takes place), confirm the membership function of these two set of factors parameters respectively, thereby obtain the judge value of two set of factors parameters current network conditions.Wherein, judge value decides according to the determined degree of membership size of membership function.Carry out bright specifically in the face of this down.

(1) bag is to the unidirectional transmission time delay of explore frame

For the unidirectional transmission time delay (ROD) of bag to explore frame, adopt the membership function under the conditional probability structure different packet loss pattern, distinguish thereby carry out lost reason according to maximum subjection principle.Its membership function is following:

In formula (2),

With

Be average under the congestion packet loss pattern and variance,

With

Be average and variance under the congestion packet loss pattern mutually, ξ for from congested with intersect congested delay factor when existing simultaneously, n is the node number, And σ _AwBe average and the variance under the error code packet loss pattern.

(2) average packet loss ratio

Average packet loss ratio is defined as the probability of packet loss incident.The packet loss incident is meant the situation that has a packet loss at a RTT in the time at least.Time period between the double packet loss is defined as one takes turns, the number that defines this packet between taking turns is for losing at interval.Because wireless error code and network congestion error code are when stable state; Its average packet loss ratio is obeyed different normal distributions; Membership function when therefore adopting average packet loss ratio conditional probability structure stable state under the different packet loss pattern carries out lost reason according to maximum subjection principle and distinguishes.

The membership function that average packet loss ratio is confirmed is following:

In the formula (3) in,

and

as congestion losses modes mean and variance; and

error packet loss mode for the mean and variance.

Then, calculate the changeable weight distribution function of average packet loss ratio and unidirectional transmission time delay (ROD) respectively, do normalization and handle, confirm the relative weighting value.According to this result of determination, distinguish lost reason through blurring mapping, thereby correspondingly adjust congestion window.

Introduce the concrete steps of calculating the changeable weight distribution function below.

Particularly; In the paper " End-to-End Differentiation of Congestion and Wireless Losses using a Fuzzy Arithmetic based on Relative Entropy " that background technology is partly mentioned, propose to utilize average packet loss ratio to distinguish lost reason.Based on the research of this paper, its distance of distinguishing average packet loss ratio under efficient and wireless error code pattern and the network congestion pattern is relevant.This distance is big more, and it is good more then to distinguish effect.Therefore, on the basis of this paper studies, the changeable weight distribution function that definition of the present invention is relevant with average packet loss ratio is following:

$W_p=\cos (N({\underset{~}{A}}_c^p,{\underset{~}{A}}_w^p)\·\frac{\mathrm{\π}}{2})---\left(4\right)$

Wherein

is the lattice approach degree between congestion packet loss fuzzy set

and the wireless error code packet loss fuzzy set

, i.e.

Because Be the normal fuzzy collection, so $N({\underset{~}{A}}_c^p,{\underset{~}{A}}_w^p)=\mathrm{Exp}(-{(({{\mathrm{\μ}}^p}_c-{{\mathrm{\μ}}^p}_w)/({{\mathrm{\σ}}^p}_c+{{\mathrm{\σ}}^p}_w))}^2).$

Similarly; In the paper " A ROD based Fuzzy packet loss differentiating algorithm for TCP in the hybrid wired/wireless network " that background technology is partly mentioned, select bag that the unidirectional transmission time delay (ROD) of explore frame is carried out packet loss differentiating.Because the unidirectional transmission time delay of bag centering probe data packet has been embodied a concentrated reflection of its queuing situation in formation; Under the congestion packet loss pattern; Packet before this ROD reflection packet loss passes through the required maximum queuing delay of this link, and under wireless error code packet loss pattern, explore frame ROD statistic and maximum queuing delay are irrelevant; But the previous detection data of reflection lost package wraps in the average queuing delay in the transmission link formation; Under the heavier situation of offered load, the queuing delay under two kinds of patterns is close, causes distinguishing accuracy and descends.

Therefore, on the basis of above-mentioned paper studies, it is following to the changeable weight distribution function of explore frame ROD that the present invention defines under the different packet loss pattern bag:

Wherein $N({\underset{~}{A}}_c^r,{\underset{~}{A}}_w^r)=\mathrm{Exp}(-{(({{\mathrm{\μ}}^r}_c-{{\mathrm{\μ}}^r}_w)/({{\mathrm{\σ}}^r}_c+{{\mathrm{\σ}}^r}_w))}^2)$

After the changeable weight distribution function that obtains unidirectional transmission time delay (ROD) and these two parameters of average packet loss ratio, next step work is the changeable weight distribution function to be carried out normalization handle, so that confirm the actual relative weighting that is adopted.

Formula (4) and formula (5) have shown the computational process of relative weighting value:

W′ _p＝W _p/(W _p+W _ROD) (6)

W′ _ROD＝W _ROD/(W _p+W _ROD) (7)

Wherein, W _pBe the changeable weight distribution function of average packet loss ratio, W _RODBe the changeable weight distribution function of bag to explore frame ROD.W ' _pBe the relative weighting value of normalization processing back average packet loss ratio, W ' _RODFor the relative weighting value of back bag to explore frame ROD handled in normalization.

Obtained average packet loss ratio and bag relative weighting value, also just obtained above-mentioned weight allocation A simultaneously explore frame ROD.According to this weight allocation A, the blurring mapping TR that mentions in the through type (1) just can obtain a multifactorial evaluation result.The result can distinguish lost reason according to this multifactorial evaluation.If be judged as congestion packet loss, then retransmit this packet, and congestion window is reduced by half; If be judged as the error code packet loss, then only retransmit this packet, and do not adjust the size of congestion window.

In order to verify the actual effect of this method, be wireless emulation topological model (referring to Fig. 1) through setting up final jump, on the NS-2 platform, carried out emulation.Result such as Fig. 3 and shown in Figure 4 through the emulation experiment acquisition.Wherein Fig. 3 has shown that under different topology structure, the different error rate, adopting fuzzy synthetic evaluation model to carry out packet loss differentiating carries out packet loss differentiating, Reno, NewReno throughput simulation result relatively with utilizing ROD.Fig. 4 has shown that under different topology structure, the different error rate, adopting fuzzy synthetic evaluation model to carry out packet loss differentiating carries out packet loss differentiating, Reno, NewReno throughput simulation result relatively with utilizing average packet loss ratio.FCELDA among the figure is the abbreviation of Fuzzy Comprehensive Evaluation based loss differentiation algorithm, means packet loss differentiating fuzzy comprehensive evaluation method agreement; FPRD is the abbreviation of Fuzzy Pattern Recognition based Differentiating algorithm, means based on ROD packet loss differentiating method agreement; Reno representes TCP Reno agreement (not comprising packet loss differentiating judges); NewReno representes TCP NewReno agreement (not comprising packet loss differentiating judges).

Can find out from Fig. 3 and simulation result shown in Figure 4, only utilize ROD to carry out packet loss differentiating, under underloaded situation, obtain packet loss differentiating effect preferably, but under the heavier situation of offered load, distinguish accuracy and obviously descend.Only utilize average packet loss ratio to carry out packet loss differentiating, when the wireless error rate is big, distinguishes accuracy and obviously descend.The fuzzy synthetic evaluation model that utilizes this method to provide carries out packet loss differentiating, and the obviously more above-mentioned two kinds of methods of performance are stable.

Above the heterogeneous network packet loss differentiating method based on fuzzy comprehensive evoluation of the present invention has been carried out detailed explanation, but obviously concrete way of realization of the present invention is not limited thereto.For the those skilled in the art in present technique field, the various conspicuous change of under the situation that does not deviate from claim scope of the present invention, it being carried out is all within protection scope of the present invention.

Claims (5)

1. heterogeneous network packet loss differentiating method based on fuzzy comprehensive evoluation is characterized in that:

(1) will wrap the unidirectional transmission time delay of explore frame and average packet loss ratio set of factors parameter as fuzzy comprehensive evoluation;

(2) when packet loss takes place, utilize said average packet loss ratio and said unidirectional transmission time delay to pass judgment on respectively, obtain the judge value of two said set of factors parameters current network conditions;

(3) calculate the changeable weight distribution function relevant with said unidirectional transmission time delay respectively, and two changeable weight distribution functions are done normalization handle, confirm the relative weighting value of said average packet loss ratio and said unidirectional transmission time delay with said average packet loss ratio;

(4) said relative weighting value is carried out blurring mapping, obtain multifactorial evaluation, thereby lost reason is distinguished.

2. the heterogeneous network packet loss differentiating method based on fuzzy comprehensive evoluation as claimed in claim 1 is characterized in that:

In the said step (3), the changeable weight distribution function relevant with said average packet loss ratio is:

$W_p=\cos (N({\underset{~}{A}}_c^p,{\underset{~}{A}}_w^p)\·\frac{\mathrm{\π}}{2})$

Where

is the congestion losses fuzzy sets

and wireless error loss fuzzy sets

closeness between cells.

3. the heterogeneous network packet loss differentiating method based on fuzzy comprehensive evoluation as claimed in claim 2 is characterized in that:

In the said step (3), the changeable weight distribution function relevant with said unidirectional transmission time delay is:

Wherein $N({\underset{~}{A}}_c^r,{\underset{~}{A}}_w^r)=\mathrm{Exp}(-{(({{\mathrm{\μ}}^r}_c-{{\mathrm{\μ}}^r}_w)/({{\mathrm{\σ}}^r}_c+{{\mathrm{\σ}}^r}_w))}^2).$

4. like claim 2 or 3 described heterogeneous network packet loss differentiating methods, it is characterized in that based on fuzzy comprehensive evoluation:

In the said step (3), obtain the relative weighting value of said unidirectional transmission time delay according to following formula:

W′ _ROD＝W _ROD/(W _p+W _ROD)

W wherein _pBe the changeable weight distribution function relevant, W with said average packet loss ratio _RODBe the changeable weight distribution function relevant, W ' with said unidirectional transmission time delay _RODHandle the relative weighting value of the said unidirectional transmission time delay in back for normalization.

5. like claim 2 or 3 described heterogeneous network packet loss differentiating methods, it is characterized in that based on fuzzy comprehensive evoluation:

In the said step (3), obtain the relative weighting value of said average packet loss ratio according to following formula:

W′ _p＝W _p/(W _p+W _ROD)

W wherein _pBe the changeable weight distribution function relevant, W with said average packet loss ratio _RODBe the changeable weight distribution function relevant, W ' with said unidirectional transmission time delay _pHandle the relative weighting value of the said average packet loss ratio in back for normalization.

Images