JP2016208212A - Traffic estimating apparatus, method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable estimation of the latent traffic volume.SOLUTION: The method includes: specifying a transition of the time series of the feature amount of first traffic on the basis of history data of first traffic in a first line in which congestion is observed; specifying a transition of the time series of the feature quantity of second traffic on the basis of history data of traffic in a second line in which congestion is not observed; on the basis of a comparison result between a transition of the time series of the feature quantity of the first traffic and a transition of the time series of the feature quantity of the second traffic in a first period in which congestion is not observed in the first line, and a transition of the time series of the feature amount of the second traffic in the second period in which the congestion is observed, correcting the feature amount of the first traffic in the second period; and on the basis of the corrected feature amount, estimating a transition of the time series of the traffic amount of the first traffic in the second period.SELECTED DRAWING: Figure 4

Description

  The present invention relates to a traffic estimation device, method, and program.

  Conventionally, when the line speed of a bottleneck link (congested line) is increased, the transition over time of the line usage rate (traffic volume in the line bandwidth) is ascertained and predicted from that transition. The required line bandwidth is designed based on this value. In addition, the design which applied queuing theory is also performed.

  When congestion occurs in the bottleneck link, the usage rate of the line where the congestion occurs is a value close to 100% (a state where the traffic amount has reached the upper limit of the line bandwidth). However, this seems to be only 100%, and it feels uncomfortable with a decrease in throughput (a decrease in traffic volume per unit time) and a decrease in communication quality due to the TCP (Transmission Control Protocol) congestion avoidance mechanism. In other words, a communication interruption by a user (reduction in the amount of requested traffic) has occurred, and thus a larger line bandwidth is originally required.

Such a decrease in the apparent traffic due to the upper limit of the line bandwidth is called “traffic latent”. The following factors can be cited as the cause of the latent traffic.
・ User withdrawal ・ Decrease in user's active communication volume (reduction of communication frequency, suspension of communication, change of service used, etc.)
・ Decrease of user dependent traffic (decrease of prefetch number, decrease of polling frequency, stop of software update, etc.)
・ Decrease in TCP throughput (feedback control by congestion avoidance mechanism)
Increase in packet loss Here, the traffic that would appear when there was no upper limit in the line bandwidth is called “latent traffic”, and the traffic volume of the potential traffic is called “latent traffic volume”.

  FIG. 1 is a diagram for explaining the amount of potential traffic. In the graphs (1) and (2) in FIG. 1, the horizontal axis corresponds to time, and the vertical axis corresponds to the traffic volume. Therefore, the curve shows a time-series transition of the traffic volume.

  (1) shows the transition of the traffic amount in the normal state (the state where congestion has not occurred). (2) shows the transition of the traffic amount in the congestion state. As shown in (2), the potential traffic and the volume refer to the difference between the original traffic volume that would have occurred when the bandwidth upper limit did not exist and the upper limit of the line bandwidth.

JP 2004-23114 A Japanese Patent No. 5258916

Jun Matsuda, Yoshishi Hayashi, Shuji Makino, "Switchboard Traffic Management Technology Based on Daily Traffic Prediction" NTT R & D, December 10, 1991, Vol.40, N0.12

  In a network with a bottleneck point, if the potential traffic at the bottleneck point is released (it becomes apparent) due to line speedup or traffic control, the traffic volume that has become apparent due to the improvement in throughput is New bottleneck points may arise. It should be noted that the latent traffic becoming obvious and becoming the original amount of traffic is called “latent traffic release”.

  FIG. 2 is a diagram illustrating the shift of the bottleneck point due to the release of latent traffic. In FIG. 2, (1) shows a state where the line L1 shared by the traffic T1 and the traffic T2 is a bottleneck point. The thickness of the arrows indicating the traffics T1 and T2 indicates the amount of traffic.

  (2) shows a state in which the traffic amount of traffic T1 is limited in response to the state of (1), and as a result, the potential traffic in traffic T2 is released and line L2 becomes a new bottleneck point. . In this case, the traffic T3 is affected by the line L2 becoming a bottleneck point. Therefore, in a congested line, considering the release of potential traffic, how much expansion should be performed. It is necessary to clarify how it works.

  Also, since the usage rate may change at a value of 100% in a bottleneck line where congestion is normally occurring, the bandwidth required by the conventional technology is designed based on the change in usage rate. It was difficult to do. As a result, by increasing the speed of the bottleneck link, potential traffic is released, and a new bottleneck link may be generated on the traffic route.

  Therefore, in order to grasp the influence of traffic accompanying the improvement of these bottleneck points, it is necessary to estimate the amount of potential traffic.

  The present invention has been made in view of the above points, and an object of the present invention is to enable estimation of the amount of potential traffic.

  Therefore, in order to solve the above-described problem, the traffic estimation device specifies the time-series transition of the feature amount of the first traffic based on the history data of the first traffic in the first line where congestion is observed, A specifying unit for identifying a time-series transition of a feature amount of the second traffic based on traffic history data in the second line where no congestion is observed; and a first part in which no congestion is observed in the first line. A comparison result of a time series transition of the feature quantity of the first traffic and a time series transition of the feature quantity of the second traffic in the period of 1, and the second period in which the congestion is observed The feature amount of the first traffic in the second period is corrected based on the time-series transition of the feature amount of the second traffic, and the second period is corrected based on the corrected feature amount. Having an estimation unit that estimates a change of the time series of the traffic volume of the first Traffic in.

  It is possible to estimate the amount of potential traffic.

It is a figure for demonstrating the amount of potential traffic. It is a figure which shows the shift of the bottleneck point by release of a latent traffic. It is a figure which shows the hardware structural example of the traffic estimation apparatus in embodiment of this invention. It is a figure which shows the function structural example of the traffic estimation apparatus in embodiment of this invention. It is a flowchart for demonstrating an example of the process sequence which a traffic estimation apparatus performs. It is a figure which shows the structural example of input traffic. It is a figure which shows notionally the process which a composition decomposition | disassembly part performs. It is a figure which shows notionally the process which a feature-value calculation part performs. It is a figure which shows notionally the process which a latent traffic amount estimation part performs. It is a figure which shows an example of the data specified by calculation of the feature-value in a 1st Example. It is a figure for demonstrating the regression analysis for every feature-value in a 1st Example. It is a figure which shows an example of the calculation result of the estimated value of the total traffic amount containing the potential traffic amount in a 1st Example. It is a figure which shows an example of the data specified by calculation of the total traffic amount for every subset in a 2nd Example. It is a figure for demonstrating the regression analysis of the total traffic amount in a 2nd Example. It is a figure which shows an example of the calculation result of the estimated value of the total traffic amount including the potential traffic amount in a 2nd Example. It is a figure which shows an example of the data specified by calculation of the number of users for every subset in a 3rd Example, and the average traffic amount for every user. It is a figure for demonstrating the regression analysis for every feature-value in a 3rd Example. It is a figure which shows an example of the calculation result of the estimated value of the total traffic amount including the potential traffic amount in a 3rd Example.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 3 is a diagram illustrating a hardware configuration example of the traffic estimation device according to the embodiment of the present invention. The traffic estimation device 10 in FIG. 3 includes a drive device 100, an auxiliary storage device 102, a memory device 103, a CPU 104, an interface device 105, and the like that are mutually connected by a bus B.

  A program that realizes processing in the traffic estimation apparatus 10 is provided by a recording medium 101 such as a CD-ROM. When the recording medium 101 storing the program is set in the drive device 100, the program is installed from the recording medium 101 to the auxiliary storage device 102 via the drive device 100. However, the program need not be installed from the recording medium 101, and may be downloaded via a network. The auxiliary storage device 102 stores the installed program and also stores necessary files and data.

  The memory device 103 reads the program from the auxiliary storage device 102 and stores it when there is an instruction to start the program. The CPU 104 executes a function related to the traffic estimation device 10 in accordance with a program stored in the memory device 103. The interface device 105 is used as an interface for connecting to a network.

  FIG. 4 is a diagram illustrating a functional configuration example of the traffic estimation device according to the embodiment of the present invention. In FIG. 4, the traffic estimation device 10 includes a composition decomposition unit 11, a feature amount calculation unit 12, a latent traffic amount estimation unit 13, and the like. Each of these units is realized by processing that one or more programs installed in the traffic estimation apparatus 10 cause the CPU 104 to execute. The traffic estimation apparatus 10 also uses an input traffic storage unit 121, a feature amount storage unit 122, and the like. Each of these storage units can be realized by using, for example, a storage device that can be connected to the auxiliary storage device 102 or the traffic estimation device 10 via a network.

  In the input traffic stored in the input traffic storage unit 121, the composition decomposition unit 11 generates a subset that does not overlap based on a pre-designated composition decomposition pattern, and sets each subset to the feature amount calculation unit 12. Output. Input traffic is log data (history data) of traffic observed for a certain line. The input traffic includes data that is input traffic (hereinafter referred to as “estimation target traffic”) used for estimation of the potential traffic volume, and input that is referred to (or referred to) for estimation of the potential traffic volume. There is data serving as traffic (hereinafter referred to as “reference traffic”). The estimation target traffic is input traffic related to traffic observed in a line in which occurrence of congestion is observed in the observation period (hereinafter referred to as “congested line”). The reference traffic is input traffic related to traffic observed in a line in which no occurrence of congestion is observed during the observation period (hereinafter referred to as “non-congested line”). That is, the congestion line and the non-congestion line are different lines.

  The feature quantity calculation unit 12 calculates the feature quantity of the input traffic for each subset generated by the composition decomposition unit 11. The feature amount calculation unit 12 passes the feature amount of the subset related to the estimation target traffic to the latent traffic amount estimation unit 13, and stores the feature amount of the subset related to the reference traffic in the feature amount storage unit 122. . Note that the feature amount is calculated for each subset generated by the composition decomposition unit 11, so that, for example, the application-specific traffic occupying the line or the composition ratio of the user's communication usage method type is the time when the traffic is observed ( The feature amount can be obtained in consideration of the fact that it varies depending on the date and time, the line (region) where the traffic is observed, and the like.

  In the present embodiment, the traffic feature amounts are the number of users, the average number of sessions per user, the average requested session size, and the average session transfer rate. This is because the traffic volume observed on the line is generally calculated by the following equation.

Traffic volume [Bytes] = number of users [persons] × average number of sessions per user [lines / person] × average requested session size [Bytes / line] × session average transfer rate [%]
That is, the traffic feature amount is a value that is a factor of the traffic amount. Details of each feature amount will be described later.

  The potential traffic amount estimation unit 13 calculates the potential traffic amount based on the feature amount of the estimation target traffic delivered from the feature amount calculation unit 12 and the feature amount of the reference traffic stored in the feature amount storage unit 122. Estimate and output the estimation result.

  Hereinafter, a processing procedure executed by the traffic estimation device 10 will be described. FIG. 5 is a flowchart for explaining an example of a processing procedure executed by the traffic estimation device.

  In step S <b> 101, the composition decomposition unit 11 acquires (reads out) the input traffic group stored in the input traffic storage unit 121.

  FIG. 6 is a diagram illustrating a configuration example of input traffic. As shown in FIG. 6, input traffic is log data of traffic in units of sessions, transmission start time, transmission end time, transmission source IP address, destination IP address, protocol, transmission source port number, destination port number, It includes parameters such as content type, content length, and transmission data length. Note that the traffic in the present embodiment is traffic in the downlink direction. The downlink direction is traffic related to a response from a server to a request from a client.

  The transmission start time is the session start time. The transmission end time is the session end time. The transmission source IP address is an IP address of a transmission source (server side) of traffic belonging to the session. The destination IP address is the IP address of the destination (client side) of the traffic belonging to the session. The protocol is a communication protocol used for traffic. FIG. 6 shows an example in which the protocol is classified as 6, 17, or other. 6 indicates TCP (Transmission Control Protocol). Reference numeral 17 denotes UDP (User Datagram Protocol). The transmission source port number is a port number of a transmission source of traffic belonging to the session. The destination port number is the port number of the destination of traffic belonging to the session. The content type is a type of content downloaded in the session. The content length is the data size of the content scheduled to be downloaded in the session. The transmission data length is the data size of the content actually downloaded in the session. For example, when the download is completely completed, the content length = the transmission data length, but when the download is interrupted, the content length> the transmission data length.

  A session refers to the period from the start to the end of communication in which parameters such as a source IP address, a destination IP address, a protocol, a source port number, and a destination port number are common. If the original data is data in units of packets, the input traffic may be generated by performing aggregation processing in advance in units of sessions based on the commonality of these parameters.

  Note that the input traffic storage unit 121 stores the estimation target traffic and the reference traffic separately. Each input traffic may further include a flag indicating whether or not it belongs to a congestion occurrence period (hereinafter referred to as “congestion period”) in the congestion line, and information indicating the congestion period is separately provided. It may be stored.

  Subsequently, the composition decomposition unit 11 divides the input traffic group into a plurality of subsets having no overlap according to a pre-designated composition decomposition pattern (step S102). The division into subsets is performed separately for each estimation target traffic group and each reference traffic group. Therefore, a subset in which the estimation target traffic and the reference traffic are mixed is not generated.

  FIG. 7 is a diagram conceptually showing processing executed by the composition decomposition unit. FIG. 7 shows an example in which the estimation target traffic group is divided into subsets 1 to 10 by the composition decomposition unit 11, and the reference traffic group is divided into subsets 1 ′ to 10 ′ by the composition decomposition unit 11. Has been. Note that the number of subsets generated for the estimation target traffic group and the number of subsets generated for the reference traffic group are not necessarily the same.

  The composition decomposition pattern includes information indicating (1) the target parameter and (2) the granularity of decomposition of the target parameter with respect to the parameters of the input traffic.

  For example, in the composition decomposition pattern, when (1) the target parameter is “transmission start time” and (2) the target parameter decomposition granularity is “minute granularity”, transmission starts An input traffic group having the same time unit is set as one set, and the input traffic group is divided into a plurality of subsets having no overlap.

  Further, for example, in the composition decomposition pattern, (1) the target parameter is “source IP address”, and (2) the target parameter decomposition granularity is “/ 24 (bit mask expression) granularity. ”, The input traffic group having the same upper 24 bits of the source IP address is taken as one set, and the input traffic group is divided into a plurality of subsets having no overlap.

  In the composition decomposition pattern, a plurality of conditions may be combined in parallel or in series. As an example of combination in parallel, (1) “transmission start time” is decomposed into (2) “granularity per minute”, and (1) “source IP address” is (2) “/ 32 granularity” In the case of the condition of disassembling the input traffic, the input traffic is divided into a plurality of non-overlapping subsets for each source IP address and every minute.

  As an example of serial combination, (1) “source IP address” is decomposed into (2) “/ 32 granularity”, and then (1) “transmission data length” is changed to (2) “granularity per 1 GB”. In the case of the condition that (1) “destination IP address” is decomposed into (2) “/ 32 granularity”, the destination IP address of the transmission source IP address whose transmission data length is greater than or equal to xGB and less than x + 1 GB For each address, the input traffic is divided into a plurality of non-overlapping sets.

  Subsequently, the feature amount calculation unit 12 calculates a feature amount for each subset generated by the composition decomposition unit 11 (step S103). Specifically, four types of statistics such as the number of users, the average number of sessions for each user, the average requested session size, and the average session transfer rate are calculated as feature quantities. However, a value obtained by multiplying some or all of these four types of feature amounts may be calculated as the feature amount.

  The number of users is obtained by counting the number of destination IP addresses of each input traffic included in the subset. That is, the number of users is acquired based on the assumption that the corresponding destination IP address is different for each user. However, if there is a mapping list between the destination IP address and the user, the user of each input traffic may be identified based on the mapping list.

  The average number of sessions per user is calculated by dividing the number of input traffic (that is, the number of sessions) included in the subset by the number of users.

  The average requested session size is calculated by dividing the total content length of each input traffic included in the subset by the number of input traffic included in the subset (that is, the number of sessions). For input traffic other than HTTP (HyperText Transfer Protocol), the transmission data length is used instead of the content length.

  The session average transfer rate is calculated by dividing the total transmission data length of each input traffic included in the subset by the total content length of each input traffic. For input traffic other than HTTP, the transmission data length is subject to addition to the total content length. In other words, for input traffic other than HTTP, the session average transfer rate is treated as 100%.

Note that a weighted session average transfer rate may be used as a feature amount instead of the session average transfer rate. When the requested session size of session i is Mi, the transfer rate is Ri, and the total number of sessions is N,
Session average transfer rate = Σi (Ri / N)
Whereas
Weighted session average transfer rate = Σi (Mi × Ri) / ΣiMi
It is. The weighted session average transfer rate is effective when the correlation between the requested session size and the transfer rate is strong.

  Note that the feature quantity calculated for the subset related to the reference traffic is stored in the feature quantity storage unit 122. The feature amount calculated for the subset related to the estimation target traffic is delivered to the latent traffic amount estimation unit 13.

FIG. 8 is a diagram conceptually illustrating processing executed by the feature amount calculation unit. In FIG. 8, the feature amounts B _{1 to} B ₁₀ calculated for the subsets 1 to 10 related to the estimation target traffic are delivered to the latent traffic amount estimation unit 13 and calculated for the subsets 1 to 10 related to the reference traffic. The feature amounts B ′ _{1 to} B ′ ₁₀ are stored in the feature amount storage unit 122. Note that the feature amount B _n and the feature amount B ′ _n (in FIG. 8, n is 1 to 10) each indicate four types of feature amounts.

  In the present embodiment, the composition decomposition pattern includes a condition in which the target parameter is “transmission start time”. Therefore, a subset in which at least “transmission start time” is divided for each unit time is generated. That is, a time series subset is generated. Further, among the subsets related to the estimation target traffic, a subset indicating that it belongs to the congestion period is given to the subset belonging to the congestion period. The subset belonging to the congestion period is a subset in which all or a part of the input traffic belonging to the subset belongs to the congestion period.

  Subsequently, the latent traffic amount estimation unit 13 stores the feature amount of each subset related to the reference traffic stored in the feature amount storage unit 122 and each of the estimation target traffic passed from the feature amount calculation unit 12. Based on the feature amount of the subset, the potential traffic volume of the congested line during the congestion period is estimated (S104).

  For example, the potential traffic amount estimation unit 13 compares the feature amount of each subset related to the reference traffic in the non-congestion period and the feature amount of each subset related to the estimation target traffic in the non-congestion period. The potential traffic amount estimation unit 13 assumes that the comparison result is established even in the congestion period, and applies the feature amount of each subset related to the reference traffic in the congestion period to the comparison result. Then, the feature amount of each subset related to the estimation target traffic is corrected. The potential traffic amount estimation unit 13 estimates the potential traffic amount by multiplying and adding the corrected feature amounts. That is, the potential traffic amount estimation unit 13 multiplies each estimated feature amount for each subset, and adds the multiplied results for each unit time. As a result, the true traffic volume of the congestion line (in time series) per unit time is estimated. Therefore, it is estimated that a value obtained by subtracting the upper limit of the bandwidth of the congestion line from the true traffic volume is the potential traffic volume of the congestion line. That is, using the regression analysis, the potential traffic volume of the congested line during the congestion period is estimated. The true traffic volume is a traffic volume including the potential traffic volume.

FIG. 9 is a diagram conceptually showing the processing executed by the latent traffic amount estimation unit 13. In FIG. 9, the feature amounts B ′ _{1 to} B ′ _{10 of} each subset related to the reference traffic stored in the feature amount storage unit 122, and each of the estimation target traffic transferred from the feature amount calculation unit 12. An example in which the amount of potential traffic is estimated based on the feature amounts B _{1 to} B _{10 of the} subset is shown. In the graph g1 shown in FIG. 9, the dashed curve is a time-series transition of the estimated value of the true traffic volume of the congestion line. Further, it is a portion where the broken curve exceeds the upper limit of the band, that is, the amount of potential traffic.

  For example, when the temporal transition of the feature amount cannot be used, the potential traffic amount may be estimated using a comparative analysis instead of the regression analysis. For example, the feature amount in the congestion period of the estimation target traffic may be corrected based on the ratio of the feature amount in the reference traffic, and the potential traffic amount may be estimated based on the corrected feature amount.

  Further, when a feature amount change model due to congestion can be used, the potential traffic amount may be estimated by model estimation. For example, the feature amount in the congestion period of the estimation target traffic may be corrected based on a model (function) given in advance.

  Subsequently, specific examples (first embodiment to fourth embodiment) of step S102 to step S104 will be described.

  In the first embodiment, in step S102, the input traffic group is divided into application-specific subsets, and the application-specific subsets are further divided into subsets for each unit time. Therefore, a subset related to the estimation target traffic and a subset related to the reference traffic are generated for each application and for each unit time. The unit time is applied to the transmission start time of input traffic. The application is distinguished based on a combination of values of the input traffic protocol, the source port number, the destination port number, and the content type. That is, the input traffic having the same value of these four parameters is classified into the same subset for each application. For input traffic that does not include a content type, the content type need not be considered.

In step S103, four types of feature quantities u _i (t), s _i (t), l _i (t), and r _i (t) are calculated for each subset. Here, i is an application index. That is, i indicates the application to which the subset belongs. t is an index of unit time. U, s, l, and r indicate the number of users, the average number of sessions per user, the average requested session size, and the average session transfer rate, respectively. Therefore, the meanings of u _i (t), s _i (t), l _i (t), and r _i (t) are as follows.
u _i (t): number of users in the subset related to application iεA (t) in unit time t s _i (t): average session per user of the subset related to application iεA (t) in unit time t Number l _i (t): Average request session size of subset related to application iεA (t) in unit time t r _i (t): Session of subset related to application iεA (t) in unit time t Average transfer rate In the first embodiment, A (t) is a set of applications in unit time t.

  By calculating the feature amount for each subset, data as shown in FIG. 10 is specified.

FIG. 10 is a diagram illustrating an example of data specified by the calculation of the feature amount in the first embodiment. FIG. 10 shows a time-series transition of the number of users (u _i (t)) of a certain application i and a time-series transition of the average number of sessions (s _i (t)) for each user of the application i. Has been. A solid curve is a transition related to the estimation target traffic related to the application i, and a broken curve is a transition related to the reference traffic related to the application i. That is, FIG. 10 shows a graph obtained by arranging the number of users in each subset per unit time and the average number of sessions per user in chronological order for a certain application i.

Although omitted in FIG. 10 for the sake of convenience, similar data can be obtained for l _i (t) and r _i (t). Similar data is specified for each application.

  In step S104, for each application, the ratio of each feature amount related to the reference traffic and each feature amount related to the estimation target traffic in the non-congested period (feature amount related to the estimation target traffic / feature amount related to the reference traffic). Is calculated by regression analysis. This ratio is hereinafter referred to as “feature amount ratio”.

  FIG. 11 is a diagram for explaining regression analysis for each feature amount in the first embodiment. In FIG. 11, the regression analysis regarding the number of users is shown. As shown in FIG. 11, the regression analysis is performed on the feature amount in the non-congested period. In the example of FIG. 11, an example is shown in which the characteristic amount ratio (number of users of the congested line ÷ number of users of the non-congested line) is obtained as a result of the regression analysis.

  The feature amount ratio is also calculated for feature amounts other than the number of users. That is, the feature amount ratio is calculated for each feature amount. For each application, the feature amount ratio is calculated for four types of feature amounts.

Further, in step S104, it is assumed that the feature amount ratio is maintained even in the congestion period, and each feature amount related to the reference traffic in the congestion period is multiplied by each feature amount ratio, thereby obtaining the estimation target traffic in the congestion period. Each feature amount is corrected. Hereinafter, the result obtained by multiplying the feature amount ratio is referred to as “correction feature amount”. When the feature quantity ratios for u _i (t), s _i (t), l _i (t), and r _i (t) are sequentially R _u , R _s , R _l , R _r , each feature quantity The corrected feature quantities u ′ _i (t), s ′ _i (t), l ′ _i (t), and r ′ _i (t) are as follows.
u ′ _i (t) = u _i (t) × R _u
s ′ _i (t) = s _i (t) × R _s
l ′ _i (t) = l _i (t) × R _l
r ′ _i (t) = r _i (t) × R _r
Where t is each unit time belonging to the congestion period, and u _i (t), s _i (t), l _i (t), and r _i (t) are feature quantities related to the reference traffic. .

  Subsequently, an estimated value of the total traffic volume including the potential traffic volume is calculated for each unit time t belonging to the congestion period. In the first embodiment, the calculation formula of the total traffic amount V (t) in the unit time t is as the following formula (1). Here, the “total traffic amount” refers to the total traffic amount by application.

したがって、潜在トラヒック量を含む総トラヒック量の推定値Ｖ'（ｔ）は、上記の式（１）におけるｕ_ｉ（ｔ）、ｓ_ｉ（ｔ）、ｌ_ｉ（ｔ）、ｒ_ｉ（ｔ）のそれぞれに対して、ｕ'_ｉ（ｔ）、ｓ'_ｉ（ｔ）、ｌ'_ｉ（ｔ）、ｒ'_ｉ（ｔ）が代入されることで算出される。

Therefore, the estimated value V ′ (t) of the total traffic amount including the potential traffic amount is represented by u _i (t), s _i (t), l _i (t), r _i (t) in the above equation (1). For each of u ′ _i (t), s ′ _i (t), l ′ _i (t), and r ′ _i (t).

  FIG. 12 is a diagram illustrating an example of a calculation result of the estimated value of the total traffic amount including the potential traffic amount in the first embodiment. In FIG. 12, a dotted line graph in the congestion period shows a time-series transition of the estimated value of the traffic volume including the potential traffic volume.

  As described above, according to the first embodiment, the total traffic amount including the potential traffic amount is estimated in consideration of the fact that the degree of change of the feature amount at the time of congestion may be different for each application, and as a result, The amount of potential traffic can be estimated. Therefore, the first embodiment is particularly effective when the degree of influence associated with deterioration of communication quality such as congestion differs for each application.

  It should be noted that the degree of change in the feature amount at the time of congestion for each application depends on whether or not a human is involved in application control. For example, in communication in which data download starts automatically and ends automatically like software update, even if the communication quality decreases, the average session transfer rate is not affected. On the other hand, for example, in a moving image or the like that a human is watching in real time, there is a high possibility that viewing will be stopped due to discomfort associated with a decrease in communication quality, and the session average transfer rate will decrease.

  Next, a second embodiment will be described. In the second embodiment, it is assumed that the ratio between the total traffic amount of the non-congested line during the non-congestion period and the total traffic amount of the congested line during the non-congestion period is maintained even during the congestion period, and this ratio is expressed as By multiplying the total traffic volume, the total traffic volume including the potential traffic volume is estimated, and thus the potential traffic volume is estimated.

  In the second embodiment, in step S102, the input traffic group is divided into subsets per unit time. Accordingly, for each unit time, a subset related to the estimation target traffic and a subset related to the reference traffic are generated. The unit time is applied to the transmission start time of input traffic.

  In step S103, the total traffic amount V (t) per unit time is calculated for each subset. The total traffic amount V (t) is obtained for each subset by obtaining the sum of the transmission data lengths of input traffic belonging to the subset. By calculating the total traffic amount V (t) for each subset (each unit time), data as shown in FIG. 10 is specified.

  FIG. 13 is a diagram illustrating an example of data specified by calculating the total traffic amount for each subset in the second embodiment. FIG. 13 shows a time-series transition of the total traffic amount. A solid curve is a transition related to the estimation target traffic, and a broken curve is a transition related to the reference traffic. That is, FIG. 13 shows a graph obtained by arranging the total traffic amounts of the respective subsets having different unit times in time series order.

  In step S104, the ratio of the total traffic amount related to the reference traffic and the total traffic amount related to the estimation target traffic in the non-congested period (the total traffic amount related to the reference traffic / the total traffic amount related to the reference traffic) is regressed. Is calculated by The ratio of the total traffic amount is also referred to as “feature amount ratio”. This is because the total traffic amount can also be considered as a traffic feature amount.

  FIG. 14 is a diagram for explaining regression analysis of the total traffic amount in the second embodiment. As shown in FIG. 14, the regression analysis is performed on the total traffic amount in the non-congested period. In the example of FIG. 14, an example is shown in which the characteristic amount ratio (total traffic amount of the congestion line / total traffic amount of the non-congestion line) is M times as a result of the regression analysis.

Further, in step S104, it is assumed that the feature amount ratio is maintained even in the congestion period, and by multiplying the total traffic amount related to the reference traffic in the congestion period by the feature amount ratio, the estimation target traffic in the congestion period is The total traffic volume including the potential traffic volume is estimated. That is, in the unit time (t), the total traffic amount of the reference traffic and V (t), when the feature amount ratio and R _v, total traffic volume V containing latent Traffic in congestion line '(t) is Is as follows.
V ′ (t) = V (t) × R _v
FIG. 15 is a diagram illustrating an example of the calculation result of the estimated value of the total traffic amount including the potential traffic amount in the second embodiment. In FIG. 15, a dotted line graph in the congestion period shows a time-series transition of the estimated value of the traffic volume including the potential traffic volume.

  As described above, in the second embodiment, the total traffic amount including the potential traffic amount is estimated without calculating the four types of feature amounts for each subset, and as a result, the potential traffic amount is estimated. . Therefore, in the second embodiment, it is difficult to derive four types of feature amounts from the input traffic, for example, when the input traffic is data including only “total traffic amount” every 5 minutes. It is especially effective in cases.

  The second embodiment is suitable when the user traffic usage pattern is the same for the non-congested line and the congested line, and the user traffic usage pattern does not change with time.

  Next, a third embodiment will be described. In the third embodiment, it is assumed that the ratio between the feature quantity for each user of the non-congested line in the non-congestion period and the feature quantity for each user of the congestion line in the non-congestion period is maintained even in the congestion period. By multiplying the ratio by the user-specific feature amount of the non-congested line, the total traffic amount including the potential traffic amount is estimated.

  In the third embodiment, in step S102, the input traffic group is divided into user-specific subsets, and the subsets are integrated based on the total transmission data length of each subset. For example, when there are three sections for the total transmission data length, a subset for each user is integrated into three subsets. The integrated subset is a subset for each set of users by traffic volume (hereinafter referred to as “traffic volume group”). Further, the subset for each traffic volume group is divided into subsets for each unit time. Therefore, a subset related to the estimation target traffic and a subset related to the reference traffic are generated for each traffic volume group and for each unit time. The unit time is applied to the transmission start time of input traffic. The user is distinguished based on the destination IP address of the input traffic.

  In step S103, the number of users and the average traffic amount per user are calculated for each subset. The number of users is the number of types of destination IP addresses of input traffic belonging to the subset. The average traffic volume for each user is an average value of the traffic volume for each user, and can be calculated by dividing the total transmission data length of input traffic belonging to the subset by the number of users in the subset. Note that the average traffic volume per user of a certain subset is equal to a value obtained by multiplying the three feature quantities other than the number of users, that is, the average session number per user, the average requested session size, and the average session transfer rate. Therefore, it can be said that the average traffic amount for each user is also a traffic feature amount.

  By calculating the number of users and the average traffic amount per user for each subset, data as shown in FIG. 16 is specified.

  FIG. 16 is a diagram illustrating an example of data specified by calculating the number of users for each subset and the average traffic amount for each user in the third embodiment. FIG. 16 shows a time-series transition of the number of users in a certain traffic volume group and a time-series transition of the average traffic volume for each user in the traffic volume group. A solid curve is a transition related to the estimation target traffic, and a broken curve is a transition related to the reference traffic. That is, FIG. 16 shows a graph obtained by arranging the number of users in each subset per unit time and the average traffic amount for each user in time series in relation to a certain traffic amount group.

  In step S104, the ratio of the number of users related to the reference traffic and the number of users related to the estimation target traffic in the non-congested period (the number of users related to the estimation target traffic / the number of users related to the reference traffic) for each traffic amount group. Is calculated by regression analysis. In addition, for each traffic volume group, the ratio of the average traffic volume per user related to the reference traffic during the non-congested period and the average traffic volume per user related to the estimation target traffic (average traffic volume per user related to the estimation target traffic / reference Average traffic volume for each user related to traffic for traffic) is calculated by regression analysis. These ratios are hereinafter referred to as “feature amount ratios”.

  FIG. 17 is a diagram for explaining regression analysis for each feature amount in the third embodiment. In FIG. 17, the regression analysis regarding the number of users is shown. As shown in FIG. 17, the regression analysis is performed on the feature amount in the non-congested period. In the example of FIG. 17, an example is shown in which the characteristic amount ratio (number of users of the congested line ÷ number of users of the non-congested line) is obtained as a result of the regression analysis. Note that the feature amount ratio is also calculated for the average traffic amount for each user.

Further, in step S104, it is assumed that the feature amount ratio is maintained even in the congestion period, and the number of users related to the reference traffic in the congestion period and the average traffic amount for each user are multiplied by the respective feature amount ratios. The number of users and the average traffic amount for each user related to the estimation target traffic during the congestion period are corrected. Hereinafter, the number of users after correction is referred to as “number of corrected users”, and the average traffic amount for each user after correction is referred to as “average traffic amount for each corrected user”. When the number of corrected users in the traffic volume group i in unit time t is u ′ _i (t) and the average traffic amount for each corrected user is w ′ _i (t), each is calculated as follows. .
u ′ _i (t) = u _i (t) × R _u
w ′ _i (t) = w _i (t) × R _w
However, u _i (t) is the number of users related to the reference traffic in the unit time t of the traffic volume group i, and w _i (t) is the user related to the reference traffic in the unit time t of the traffic volume group i. every average traffic volume, R _u, the feature quantity ratio according to the number of users, R _w is a characteristic quantity ratio according to the average amount of traffic for each user.

In FIG. 17, each time series transition of u ′ _i (t) and w ′ _i (t) is represented by dotted lines.

  Subsequently, an estimated value of the total traffic volume including the potential traffic volume is calculated for each unit time t belonging to the congestion period. In the third embodiment, the formula for calculating the total traffic amount V (t) in unit time t is as the following formula (2). Here, “total traffic volume” refers to the total traffic volume for each traffic volume group.

なお、式（２）において、Ａ（ｔ）は、単位時間ｔにおけるトラヒック量別グループの集合である。したがって、潜在トラヒック量を含む総トラヒック量の推定値Ｖ'（ｔ）は、上記の式（２）におけるｕ_ｉ（ｔ）、ｗ_ｉ（ｔ）のそれぞれに対して、ｕ'_ｉ（ｔ）、ｗ'_ｉ（ｔ）が代入されることにより算出される。

In equation (2), A (t) is a set of traffic volume groups in unit time t. Therefore, the estimated value V ′ (t) of the total traffic amount including the potential traffic amount is u ′ _i (t) for each of u _i (t) and w _i (t) in the above equation (2). , W ′ _i (t) is substituted.

  FIG. 18 is a diagram illustrating an example of a calculation result of the estimated value of the total traffic amount including the potential traffic amount in the third example. In FIG. 18, a dotted line graph in the congestion period shows a time-series transition of the estimated value of the total traffic amount including the potential traffic amount.

  As described above, according to the third embodiment, the total traffic amount including the potential traffic amount is calculated in consideration that the degree of change in the feature amount at the time of congestion may vary depending on the use method of communication by the user. As a result, the amount of potential traffic can be estimated. In addition, the difference in the usage method of the user communication is, for example, a difference between a heavy user and a light user, a difference between communication using a fixed terminal and communication using a mobile terminal, or the like.

  The third embodiment is suitable when the traffic usage pattern per user is the same and the traffic usage pattern per user does not change with time. In addition, the third embodiment is suitable when the traffic usage of a user decreases due to the presence of a user who stops (leaves) communication due to a decrease in communication quality, a decrease in usage time, or a decrease in throughput. is there.

  Next, a fourth embodiment will be described. In the fourth embodiment, it is assumed that the ratio between the traffic amount by protocol of the non-congested line in the non-congestion period and the traffic amount by protocol of the congestion line in the non-congestion period is maintained even in the congestion period. Is multiplied by the traffic amount for each non-congested channel protocol to estimate the total traffic amount including the potential traffic amount.

  In the fourth embodiment, in step S102, the input traffic group is divided into protocol-specific subsets, and the protocol-specific subsets are further divided into subsets for each unit time. Therefore, a subset related to the estimation target traffic and a subset related to the reference traffic are generated for each protocol and for each unit time. The unit time is applied to the transmission start time of input traffic. The protocol is distinguished based on the value of the input traffic protocol.

In step S103, for each subset, the traffic volume v _i (t) per unit time for each protocol is calculated. The traffic amount v _i (t) is obtained for each subset by obtaining the sum of the transmission data lengths of the input traffic belonging to the subset. In the fourth embodiment, i is an index for the protocol.

  In step S104, the ratio of the total traffic volume related to the reference traffic and the total traffic volume related to the estimation target traffic related to the TCP in the non-congestion period related to the TCP (the total traffic volume related to the reference traffic divided by the reference traffic). The total traffic amount) is calculated by regression analysis. This ratio is also referred to as “feature amount ratio”. On the other hand, the feature amount ratio does not have to be calculated for protocols other than TCP.

In step S104, regarding TCP, it is assumed that the feature amount ratio is maintained even in the congestion period, and the traffic amount related to the reference traffic in the congestion period is multiplied by the feature amount ratio, so that the estimation target in the congestion period is obtained. The traffic amount related to the traffic is corrected. _Assuming that the feature amount ratio is R _tcp , the TCP traffic amount v ′ _tcp (t) in which the latent traffic is considered is as follows.
v ′ _tcp (t) = v _tcp (t) × R _tcp
However, v _tcp (t) is the total traffic amount related to the reference traffic.

  On the other hand, except for TCP, the estimation target traffic during the congestion period is not corrected. This is based on the idea that, when congestion occurs, the throughput is reduced by the congestion avoidance mechanism for TCP and the traffic volume is reduced, but other protocols such as UDP are not easily affected by the congestion.

  Subsequently, an estimated value of the total traffic volume including the potential traffic volume is calculated for each unit time t belonging to the congestion period. Here, the calculation formula of the total traffic amount V (t) in the unit time t is as the following formula (3). Here, “total traffic volume” refers to the total traffic volume for each application.

但し、潜在トラヒック量を含む総トラヒック量の推定値Ｖ'（ｔ）は、上記の式（３）におけるｖ_ｉ（ｔ）に対して、ｖ'_ｉ（ｔ）が代入されることで算出される。なお、第４の実施例において、上記のＡ（ｔ）は、単位時間ｔにおけるプロトコルの集合である。

However, the estimated value V ′ (t) of the total traffic volume including the potential traffic volume is calculated by substituting v ′ _i (t) for v _i (t) in the above equation (3). The In the fourth embodiment, A (t) is a set of protocols in unit time t.

  As described above, in the fourth embodiment, it is possible to estimate the total traffic amount including the potential traffic amount in consideration of the influence of each protocol when the congestion occurs, and as a result, the potential traffic amount is reduced. Presumed.

  Note that the fourth embodiment is suitable when the usage pattern for each protocol is the same and the usage pattern for each protocol does not change with time even if there is a difference in the proportion of protocols.

  As described above, according to the present embodiment, the amount of potential traffic during the congestion period can be estimated. In the present embodiment, the traffic is compositionally decomposed (divided into subsets), and feature values are restored (corrected) for each composition (for each subset). Therefore, the amount of potential traffic can be estimated in consideration of the fact that the degree of influence at the time of congestion may differ for each composition.

  As a result, network operators estimate the amount of potential traffic at the bottleneck point and analyze the route, thereby improving traffic improvement measures (such as facility expansion and NW control) such as line speedup, cache, accommodation switching, and bandwidth control. It is possible to estimate the effect and evaluate the range of influence, and it is possible to support the implementation of the optimum line design necessary for eliminating the line congestion.

  In the present embodiment, the congestion line is an example of the first line. The non-congested line is an example of a second line. Input traffic is an example of history data. The congestion period is an example of a first period. The non-congestion period is an example of a second period. The feature amount calculation unit 12 is an example of a specifying unit. The potential traffic amount estimation unit 13 is an example of an estimation unit. The composition decomposition unit 11 is an example of a division unit.

  As mentioned above, although the Example of this invention was explained in full detail, this invention is not limited to such specific embodiment, In the range of the summary of this invention described in the claim, various deformation | transformation・ Change is possible.

DESCRIPTION OF SYMBOLS 10 Traffic estimation apparatus 11 Composition decomposition | disassembly part 12 Feature amount calculation part 13 Potential traffic amount estimation part 100 Drive apparatus 101 Recording medium 102 Auxiliary storage apparatus 103 Memory apparatus 104 CPU
105 Interface Device 121 Input Traffic Storage Unit 122 Feature Quantity Storage Unit B Bus

Claims (8)

Based on the history data of the first traffic in the first line where congestion is observed, the time-series transition of the feature quantity of the first traffic is specified, and the traffic in the second line where congestion is not observed A specifying unit for specifying a time-series transition of the feature amount of the second traffic based on the history data;
A comparison result between a time series transition of the feature quantity of the first traffic and a time series transition of the feature quantity of the second traffic in a first period in which no congestion is observed in the first line; Based on the time-series transition of the feature quantity of the second traffic in the second period in which congestion is observed in the first line, the feature quantity of the first traffic in the second period is calculated. An estimation unit that corrects and estimates a time-series transition of the traffic amount of the first traffic in the second period based on the corrected feature amount;
A traffic estimation apparatus comprising: Based on the parameter value of the history data, the history data of each of the first traffic and the second traffic is divided into a plurality of subsets for each of the first traffic and the second traffic. Having a dividing portion to
The specifying unit specifies, for each subset related to the first traffic, a time-series transition of the feature amount of the first traffic based on history data belonging to the subset, and the second traffic For each such subset, the time-series transition of the feature quantity of the second traffic is specified based on the history data belonging to the subset,
The estimation unit includes, for each subset related to the first traffic, a time-series transition of the feature amount of the first traffic in the first period, and a value of the parameter common to the subset and the parameter. Based on the comparison result with the time-series transition of the feature quantity of the second traffic relating to the subset relating to the second traffic, the first in the second period of the subset relating to the first traffic The traffic feature amount of the first traffic is corrected, the time-series transition of the traffic amount in the second period is estimated based on the feature amount corrected for each subset, and for each subset related to the first traffic Estimating a time-series transition of the traffic volume of the first traffic in the second period based on the estimated traffic volume;
The traffic estimation apparatus according to claim 1, wherein: The feature amount is calculated based on the history data, the number of users, the average number of sessions per user, the average requested session size, and the average session transfer rate,
The specifying unit specifies a time-series transition related to the first traffic and a time-series transition related to the second traffic for each feature amount,
The estimation unit compares, for each feature quantity, a time series transition of the feature quantity of the first traffic and a time series transition of the feature quantity of the second traffic in the first period. And correcting the feature amount of the first traffic in the second period based on the time-series transition of the feature amount of the second traffic in the previous second period, and for each feature amount Based on the correction result, the time-series transition of the traffic amount in the second period is estimated.
The traffic estimation apparatus according to claim 1 or 2, characterized in that The specifying unit specifies a time-series transition of the traffic amount of the first traffic based on history data of the first traffic on the first line, and based on the history data of traffic on the second line. Identify the time-series transition of the traffic volume of the second traffic,
The estimation unit includes a comparison result between a time series transition of the traffic amount of the first traffic and a time series transition of the traffic amount of the second traffic in the first period, and the second period. Correcting the traffic volume of the first traffic in the second period based on the time-series transition of the traffic volume of the second traffic in
The traffic estimation apparatus according to claim 1, wherein: The dividing unit divides each history data of the first traffic and the second traffic into a plurality of subsets based on a traffic amount for each user based on a parameter value of the history data.
For each subset related to the first traffic, the specifying unit specifies the time series transition of the number of users of the subset and the average traffic amount for each user based on history data belonging to the subset, For each subset related to the second traffic, based on the history data belonging to the subset, identify the number of users in the subset and the respective time-series transitions of the average traffic amount per user,
The estimation unit estimates a time-series transition of the traffic amount of the first traffic in the second period, using the number of users and the average traffic amount for each user as the feature amount.
The traffic estimation apparatus according to claim 2, wherein: The dividing unit divides each history data of the first traffic and the second traffic into a plurality of subsets for each protocol based on a parameter value of the history data.
The specifying unit specifies, for each subset related to the first traffic, a time-series transition of the traffic amount of the first traffic based on history data belonging to the subset, and sets the second traffic as the second traffic. For each such subset, the time-series transition of the traffic volume of the second traffic is identified based on historical data belonging to the subset,
The estimation unit, for a subset related to a part of the protocol, the time-series transition of the traffic amount of the first traffic in the first period, and the value of the parameter common to the subset and the parameter. Based on a comparison result with a time-series transition of the traffic volume of the second traffic related to the subset related to the second traffic, the first of the subset related to the first traffic in the second period The traffic amount of the traffic is corrected, the transition of the corrected traffic amount in time series, and the subset related to the protocol excluding the part of the protocol, specified by the specifying unit with respect to the second period, The traffic volume of the first traffic in the second period is estimated based on the time-series transition of the traffic volume of the first traffic. ,
The traffic estimation apparatus according to claim 2, wherein: Computer
Based on the history data of the first traffic in the first line where congestion is observed, the time-series transition of the feature quantity of the first traffic is specified, and the traffic in the second line where congestion is not observed A specific procedure for identifying the time-series transition of the feature amount of the second traffic based on the history data;
A comparison result between a time series transition of the feature quantity of the first traffic and a time series transition of the feature quantity of the second traffic in a first period in which no congestion is observed in the first line; Based on the time-series transition of the feature quantity of the second traffic in the second period in which congestion is observed in the first line, the feature quantity of the first traffic in the second period is calculated. An estimation procedure for correcting and estimating a time-series transition of the traffic amount of the first traffic in the second period based on the corrected feature amount;
The traffic estimation method characterized by performing this.   A program that causes a computer to function as each unit according to claim 1.

Images