JP6186303B2 - Traffic amount upper limit prediction apparatus, method and program - Google Patents

Description

  The present invention relates to a traffic volume upper limit prediction apparatus, method, and program, and more particularly, to a traffic volume upper limit prediction apparatus, method, and program for predicting an upper limit value of a future traffic volume from information on traffic flowing on a communication network.

  In recent years, the amount of traffic flowing on a communication network has fluctuated greatly. Therefore, it is very important for network facility operators to predict future traffic fluctuations in order to formulate a design / operation plan.

  In the conventional telephone network, in order to estimate future fluctuations in traffic volume, the average value of the most busy call volume on the top 30 days of the year is used as the basic call volume, and bandwidth design has been made based on the basic call volume. However, the amount of traffic flowing through the Internet and backbone network is currently increasing year by year. Such a packet network is more likely to generate bursty traffic than the telephone network, and its scale is large. For this reason, it is not appropriate to treat the basic call volume as a measure of the traffic volume fluctuation as in the case of the telephone network.

  Therefore, there are a time series prediction method for predicting an average traffic amount fluctuation (for example, see Patent Document 1), a traffic amount upper limit prediction method using a principal component analysis and a SARIMA model (for example, see Non-Patent Document 1), and the like. Proposed.

JP 2004-23114 A

Tatsuya Otoshi, Yuichi Ohshita, Masayuki Murata, Yosuke Takahashi, Keisuke Ishibashi, Hiroshi Shiomoto, "Traffic Prediction for Dynamic Traffic Engineering Considering Time Variation of Traffic," IEICE Technical Report (NS2012- 115), vol. 112, no. 287, pp. 65-70, Nov. 2012.

  The problems to be solved by the present invention with respect to the above conventional technique are the following two points.

  The first is the inefficiency of traffic design for each link. Here, inefficiency refers to overestimating the link bandwidth with respect to the actual traffic volume. For example, consider a situation where a large amount of traffic periodically arrives for a certain link. At this time, it is desirable to design the link bandwidth with a certain margin so that the utilization factor of the link bandwidth does not become 100%. However, if the margin is excessively estimated, there is a problem that the link utilization efficiency is lowered. On the other hand, if the margin is overestimated and overestimated, the link utilization rate reaches 100%, which may lead to frequent packet loss. At this time, if it is possible to know how the traffic volume generated in the past is probabilistically distributed, the traffic volume generated in the future can be estimated probabilistically.

The second point is inefficiency in bandwidth allocation for each session. Here, inefficiency refers to the fact that the bandwidth is allocated too much to the bandwidth actually requested by the session. Many sessions are set up within one physical link. At this time, the network allocates a usable bandwidth for each session. By allocating a bandwidth as small as possible, a single link can be shared by many sessions. However, if the bandwidth allocated to each session is too small, the bandwidth used by the session exceeds the allocated bandwidth, resulting in packet loss. However, if a large band is allocated too much, an excessively allocated band is not used, resulting in inefficient band allocation. At this time, if it is possible to know how the traffic used by the session in the past has been probabilistically distributed, it becomes possible to estimate the traffic used by the session in the future.
These two issues are summarized in the issue of calculating the probability distribution at each time point when given certain traffic time-series data and predicting the upper limit value of the traffic volume at each future time point using the distribution. .

  However, it is difficult to estimate such a stochastic distribution at each point in time when the conventional method of predicting the future traffic volume from only one series is used.

  The present invention has been made in view of the above points, and enables calculation of a predicted value of a traffic amount with high accuracy, and further allows calculation of a predicted value from traffic data of different scales of links. An object of the present invention is to provide a value prediction apparatus, method, and program.

According to one aspect, a traffic volume upper limit prediction device that predicts an upper limit value of a future traffic volume from information on traffic flowing on a communication network,
Traffic data collection means for acquiring the total traffic volume of each link in time series for each arbitrary period or the traffic volume in each session and storing it in the traffic DB;
A correlation sequence calculating means for correlating the acquired traffic data of a plurality of time series, extracting link information or session information of correlated traffic data, and storing it in a correlation sequence DB;
Based on a packet loss rate and a predicted link or session request, the link information or the session information is acquired from the correlation sequence DB, and traffic data of the link or session corresponding to the link information or the session information is obtained as the traffic. Perform normalization processing to obtain comparisons from the DB, compare them at the same scale, create a standard series from each normalized traffic data, convert the standard series to the original scale, and traffic the converted series There is provided a traffic volume upper limit predicting device having a traffic maximum value predicting means for calculating a predicted value and obtaining a maximum traffic volume.

  According to one aspect, it is possible to obtain a predicted value with high accuracy by correcting using past traffic data of a plurality of sequences, and further, even from traffic data of different scales of other links. By correcting to the same scale of the link, it is possible to calculate a predicted value, improving the future traffic prediction accuracy, and contributing to optimal network planning and equipment construction.

The structural example of the traffic amount upper limit prediction apparatus in one embodiment of this invention. The flowchart of the correlation series calculation part in the 1st Embodiment of this invention. The flowchart of the traffic maximum value estimation part in the 1st Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  Two patterns of link bandwidth design and bandwidth allocation are conceivable as embodiments of the present invention. A link bandwidth design will be described as a first embodiment, and bandwidth allocation will be described as a second embodiment.

  FIG. 1 shows a configuration example of a traffic volume upper limit prediction apparatus according to an embodiment of the present invention. The traffic volume upper limit prediction apparatus shown in the figure has the same configuration in both the first and second embodiments.

  The traffic volume upper limit prediction apparatus 10 is connected to the network 20.

  The traffic volume upper limit prediction apparatus 10 includes a traffic data collection unit 100, a correlation sequence calculation unit 200, a traffic maximum value prediction unit 300, and a user interface 400.

  The traffic data collection unit 100 shown in the figure is connected to the network 20, the correlation sequence calculation unit 200, and the traffic maximum value prediction unit 300, and includes a traffic data collection unit 101 and a traffic DB 102. The correlation sequence calculation unit 200 is connected to the traffic DB 102 and the traffic maximum value prediction unit 300 of the traffic data collection unit 100, and includes a correlation sequence calculation unit 201 and a correlation sequence DB 202. The traffic maximum value prediction unit 300 is connected to the traffic DB 102, the correlation sequence DB 202, and the user interface 400, and includes a normalization unit 301, a standard sequence creation unit 302, a denormalization unit 303, and a predicted value calculation unit 304.

  In the traffic DB 102, the amount of information flowing through the link at a certain time acquired from the network 20 in the traffic data collecting unit 101 is stored in time series. The data is used by the correlation sequence calculation unit 200 and the traffic maximum value prediction unit 300.

  The correlation sequence DB 202 stores a set of link information in which the Pearson product moment correlation coefficient calculated by the correlation sequence calculation unit 201 exceeds a predetermined threshold. The data is used in the traffic maximum value prediction unit 300.

[First Embodiment]
In the present embodiment, link band design will be described.

(1) Traffic data collection:
The traffic data collection unit 101 is connected to the network 20 and acquires the traffic volume (information volume) that flows through each link at predetermined time intervals such as every 10 minutes, for example, and calculates the total traffic volume as traffic. Store in DB102. For example, when data is acquired from a link y at time t, the amount of information that has flowed through the link y during time [t−10, t] is collected as y _t and stored in the traffic DB 102.

(2) Updating the correlation series DB:
The correlation sequence calculation unit 201 updates the correlation sequence DB according to the flow shown in FIG. 2 for the time series data acquired from the traffic DB 102.

Correlation sequence calculation unit 201 sets a set of sequences not calculated to x (step 101), and if x = φ is not (step 102, No), obtains traffic data in each link at each time point, The Pearson product moment correlation coefficient ρ is calculated for the link combination. For example, when there are 10 links and the traffic time-series data of each link up to time t is obtained as _{yt, 1} , yt _{, 2} , ..., yt _{, 10} , y _{t, 1} and y _{t, 2} , y _{t, 1} and y _{t, 3} ,..., y _{t, 9} and y _{t, 10} respectively, Pearson product moment correlation coefficients ρ _1,2 , ρ _1,3,. , Ρ _9,10 . Furthermore, a correlation sequence computing unit 201 to determine the stability of the correlation, to calculate the stability factor s _t (step 103). s _t is the following equation (1) using the correlation coefficients ρ _tk , ρ _{t-k + 1} , ..., ρ _{t-1 up} to k periods (k is the period in which time-series data is collected) Desired. However,

はk期前までの時点における相関係数ρ_t-k，ρ_t-k+1，…,ρ_t-1の算術平均である。

Correlation coefficient [rho _tk at the time of before k phase, ρ _{tk + 1,} ..., is the arithmetic mean of the [rho _t-1.

ステップ１０３で求められたρがしきい値th_corを超えたリンクの組であり（ステップ１０４、Yes）、かつ、安定係数s_tがしきい値th_staを下回る系列の組（ρ_t＞th_cor AND s_t＜th_sta）については（ステップ１０５、Yes）、安定して相関が見られると判断され、相関系列DB２０２に格納される(ステップ１０６)。

Step 103 [rho obtained is the set of links exceeds the threshold th _cor (step 104, Yes), and the sequence stability factor s _t is below the threshold th _sta pair (ρ _t> th the _{cor aND s t <th sta)} ( step 105, Yes), it is determined that stable correlation is observed, it is stored in the correlation sequence DB 202 (step 106).

Further, at this time, the correlation sequence calculation unit 201 has exceeded th _cor in the past, but the link set that has decreased below th _cor is deleted from the correlation sequence DB 202. It is assumed that th _cor is not limited to a constant and can be changed according to conditions. For example, consider a case where the number of links for collecting traffic data is given as a condition. When the number of links for collecting traffic is small, it is possible to increase the number of links having some correlation by setting th _cor small. Conversely, when the number of collected links is large, it is possible to acquire only a set of links having a high correlation by setting th _cor large.

  The parameter k is a value given by the operator via the user interface 400, and indicates to what point in the past the stability should be considered. If k is set to a small value, only a set of links that can be stably correlated in the short term can be stored in the correlation sequence DB 202. If k is set to a large value, a link that can be correlated stably in the long term. Can be stored in the correlation sequence DB 202 only.

(3) Traffic volume prediction:
The traffic maximum value prediction unit 300 performs processing according to the flow shown in FIG. 3 when the user wants to predict the future traffic volume on the link x.

  Via the user interface 400 of the maximum traffic value prediction unit 300, link information (for example, link name x) and a packet loss rate p to be predicted from the user are acquired and stored in a memory (not shown) (step 201). ).

  The normalization unit 301 in the traffic maximum value prediction unit 300 obtains a set of links correlated with the link from the correlation sequence DB 202 based on the link name to be predicted input from the user interface 400 (step). 202).

Next, the time series data of those links is acquired from the traffic DB 102 of the traffic data collection unit 100, and then the normalization unit 301 obtains the normalized series from which the influence of the trend tendency and the scale is excluded for each time series data. Obtain (step 203). Normalization is performed by assuming that the time series data to be predicted at time t is represented as x _t , the series of links correlated with x _t as a _{t 1} , a _{t 2 ,.}

は、(2)式もしくは(3)式を用いて計算することができる。

Can be calculated using equation (2) or equation (3).

ただし

However,

で表され、b₀,b₁は、最小二乗法を用いて計算される。

And b ₀ and b ₁ are calculated using the least square method.

  The standard sequence creation unit 302 estimates the traffic distribution at each time point based on the n + 1 normalized sequences obtained by the normalization unit 301. The traffic distribution is assumed to be a normal distribution (4), and parameters [mu] and [sigma] are estimated with the maximum likelihood using the expressions (5) and (6).

このようにして得られたトラヒック分布より、各時点におけるトラヒック量を見積もり、ある一つの標準系列

Based on the traffic distribution obtained in this way, the traffic volume at each time point is estimated, and one standard series

を得る。例えば、(7)式を用いて標準系列

Get. For example, the standard series using equation (7)

を計算することができる。

Can be calculated.

(6)式において、ユーザが入力したパケットロス率pが利用されている。この時、pを小さく設定すればするほど、将来のトラヒック量がリンク帯域を超過する可能性が低くなるため、よりロバストなリンク帯域設計が可能となり、pを大きく設定すればするほど、ロバスト性をある程度犠牲にすることで、より経済的なリンク帯域設計が可能となる（ステップ２０４）。

In equation (6), the packet loss rate p input by the user is used. At this time, the smaller p is set, the lower the possibility that the future traffic will exceed the link bandwidth. Therefore, a more robust link bandwidth design is possible, and the more p is set, the more robust Is sacrificed to some extent, a more economical link bandwidth design is possible (step 204).

  In the denormalization unit 303, the sequence in which the influence of the trend and the scale is returned by performing the inverse process of the process performed in the normalization unit 301 on the standard sequence created in the standard sequence creation unit 302.

を得ることができる（ステップ２０５）。

Can be obtained (step 205).

  In the predicted value calculation unit 304, the sequence obtained by the denormalization unit 303

をもとに、トラヒック予測値の計算を行う。例えば、(8)式に示すARIMAモデルを適用し、信頼区間99%の値を用いることで、トラヒック上限値の予測が可能となる（ステップ２０６）。

Based on the above, the traffic prediction value is calculated. For example, the traffic upper limit value can be predicted by applying the ARIMA model shown in Equation (8) and using the value of the confidence interval 99% (step 206).

ユーザインタフェース４００は、トラヒック最大値予測部３００に対し、予測を行いたいリンクの情報及び、パケットロス率pを入力し、上記の予測値計算ユニット３０４により求められた将来の各時点におけるトラヒック上限値をユーザに対して出力する（ステップ２０７）。このシステムを利用するユーザは、この情報を利用して、将来のリンク帯域設計をより効率的に行うことが可能となる。

The user interface 400 inputs information on the link to be predicted and the packet loss rate p to the traffic maximum value prediction unit 300, and the traffic upper limit value at each future time point obtained by the prediction value calculation unit 304 described above. Is output to the user (step 207). A user who uses this system can use this information to perform future link bandwidth design more efficiently.

  According to this embodiment, when the traffic information of each link is used to estimate the future traffic volume of a certain link, the link design of the network can be performed by adopting the upper limit value. . For example, it is assumed that the traffic generated at a certain time in the past is smaller than the average value of the traffic distribution at that time. At this time, a network with a lower packet loss rate can be realized by designing the future link design bandwidth as a value larger than the design bandwidth calculated by the conventional method. On the other hand, it is assumed that the traffic generated at a certain point in the past is larger than the average value of the traffic distribution at that point. At this time, by designing the future link design bandwidth as a value smaller than the design bandwidth calculated by the conventional method, a more economical network can be realized while suppressing the packet loss rate.

[Second Embodiment]
In this embodiment, bandwidth allocation will be described.

  The configuration of the traffic volume upper limit prediction apparatus in the present embodiment is the same as that in FIG.

  The flow of bandwidth allocation is basically the same as that of the first embodiment, but is the same as that of link bandwidth design, but only the traffic data acquisition method is different.

(1) Acquisition of traffic data:
In the link bandwidth design of the first embodiment, the traffic data collection unit uses the total traffic amount of each link as input data. However, in bandwidth allocation, the traffic amount in each session is used as input data. Estimate future required bandwidth. Therefore, the traffic data collection unit 101 collects traffic data for each session and stores it in the traffic DB 102.

(2) Updating the correlation series DB:
Similar to the link bandwidth design in the first embodiment, the correlation sequence DB 202 is updated. At this time, the data used in the correlation sequence calculation unit 201 is traffic data for each session of (1) above.

(3) Traffic volume prediction:
Similar to the link bandwidth design in the first embodiment, the traffic maximum value prediction unit 300 predicts the traffic volume at each future time point and outputs it to the user via the user interface 400. Thus, the user can estimate an appropriate bandwidth allocation amount for each session based on this result.

  According to the present embodiment, when the future traffic amount of a session is estimated using the traffic information of each session, efficient bandwidth allocation is performed by allocating the bandwidth to the upper limit value at each future time point. It becomes possible. For example, it is assumed that the traffic used by a session at a certain point in the past is smaller than the average value of the traffic distribution at that point. At this time, by assigning a value larger than the allocated bandwidth calculated by the conventional method to this session, traffic control with a further reduced packet loss rate becomes possible. On the other hand, it is assumed that the traffic generated at a certain point in the past is larger than the average value of the traffic distribution at that point. At this time, by assigning a value smaller than the assigned bandwidth calculated by the conventional method to this session, more sessions can use one link while suppressing the packet loss rate.

  The operation of each component of the traffic volume upper limit prediction apparatus 10 in FIG. 1 is constructed as a program and installed in a computer used as the traffic volume upper limit prediction apparatus, or is executed or distributed via a network. It is possible.

  The present invention is not limited to the above-described embodiments, and various modifications or applications are possible within the scope of the claims.

10 traffic volume upper limit prediction device 20 network 100 traffic data collection unit 101 traffic data collection unit 102 traffic DB
200 Correlation Sequence Calculation Unit 201 Correlation Sequence Calculation Unit 202 Correlation Sequence DB
300 traffic maximum value prediction unit 301 normalization unit 302 standard sequence creation unit 303 inverse normalization unit 304 prediction value calculation unit 400 user interface

Claims (7)

A traffic volume upper limit prediction device for predicting an upper limit value of future traffic volume from information on traffic flowing on a communication network,
Traffic data collection means for acquiring the total traffic volume of each link in time series for each arbitrary period or the traffic volume in each session and storing it in the traffic DB;
A correlation sequence calculating means for correlating the acquired traffic data of a plurality of time series, extracting link information or session information of correlated traffic data, and storing it in a correlation sequence DB;
Based on a packet loss rate and a predicted link or session request, the link information or the session information is acquired from the correlation sequence DB, and traffic data of the link or session corresponding to the link information or the session information is obtained as the traffic. Perform normalization processing to obtain comparisons from the DB, compare them at the same scale, create a standard series from each normalized traffic data, convert the standard series to the original scale, and traffic the converted series A traffic maximum value predicting means for calculating a predicted value and obtaining a maximum value of the traffic volume;
A traffic volume upper limit prediction apparatus characterized by comprising: The correlation sequence calculation means includes:
The traffic data for each link or session at each time point is acquired from the traffic DB, correlation coefficients are obtained for all links or combinations of all sessions, and the standard deviation of the correlation coefficient in the period up to any past time point Correlation coefficient calculating means for obtaining a stability coefficient by obtaining
Link information or session information of a link or session set in which the correlation coefficient is larger than a predetermined threshold th _cor and the stability coefficient is smaller than a predetermined threshold th _sta is extracted into the correlation sequence DB The traffic volume upper limit prediction apparatus according to claim 1, further comprising storage means for storing. The traffic maximum value predicting means includes:
When creating the standard series, it is assumed that the traffic amount at each time point follows a certain probability distribution, the traffic distribution is obtained from each normalized traffic data, and the traffic amount at each time point is estimated from the traffic distribution. The traffic amount upper limit prediction apparatus according to claim 1, further comprising a standard sequence creating means for making a standard sequence. The standard series creating means is
The traffic amount upper limit according to claim 3, further comprising means for estimating a traffic amount at each time point based on the packet loss rate for adjusting whether the acquired link robustness or economic importance is emphasized from the traffic distribution. Value prediction device. The traffic maximum value predicting means includes:
5. The denormalization unit according to claim 3 or 4 , further comprising a denormalization unit that performs a process reverse to the normalization process on the standard series created by the standard series creation unit to obtain a series in which an influence of a trend or a size is returned. Traffic volume upper limit prediction device. A traffic volume upper limit prediction method in an apparatus for predicting an upper limit value of a future traffic volume from information of traffic flowing on a communication network,
A traffic data collection step of acquiring the total traffic volume of each link in time series for each arbitrary period or acquiring the traffic volume in each session and storing it in the traffic DB;
A correlation sequence calculation step of correlating the acquired traffic data of a plurality of time series, extracting link information or session information of the correlated traffic data, and storing it in the correlation sequence DB;
Based on a packet loss rate and a predicted link or session request, the link information or the session information is acquired from the correlation sequence DB, and traffic data of the link or session corresponding to the link information or the session information is obtained as the traffic. Perform normalization processing to obtain comparisons from the DB, compare them at the same scale, create a standard series from each normalized traffic data, convert the standard series to the original scale, and traffic the converted series A traffic maximum value prediction step for calculating a predicted value and obtaining a maximum value of the traffic volume;
A traffic volume upper limit prediction method. Computer
A traffic volume upper limit prediction program that functions as each unit of the traffic volume upper limit prediction apparatus according to any one of claims 1 to 5.

Images