JP6618971B2 - Estimation apparatus, estimation method, and program - Google Patents

Description

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

  With the widespread use of the Internet, whether or not a web browser can be used comfortably occupies a large position as to whether the user can use the Internet comfortably.

  Therefore, it is important to perform quality control for the display waiting time when the user browses the web page.

  In general, in a mobile network represented by LTE (Long Term Evolution), the congestion status varies depending on the area and time zone, so the use quality differs for each area, and the quality also varies depending on the time zone.

  As a method of grasping the quality for each area / time zone, there is a method (active measurement) in which content is actually accessed and measured from an individual terminal (for example, Patent Document 1).

JP 2017-92661 A

  However, it is not economical to perform active measurement in each area and time zone. In addition, the web page server is very different, and there is a difference in the influence on the display waiting time even under the influence of the same network quality degradation for each individual web page. Judgment of quality (deterioration degree) is difficult.

  The present invention has been made in view of the above points, and an object of the present invention is to make it possible to estimate the quality of a display waiting time from which the influence of individual web pages is removed.

  Therefore, in order to solve the above-described problem, the estimation device determines the quality of the display waiting time based on the display waiting time measurement data on the web page for each of a plurality of trials in which at least one of the region and the time zone is different. A calculation unit that calculates a representative value of the index to be indicated, a comparison unit that outputs a logical value indicating a comparison result with a threshold value for the representative value related to the trial for each trial, and a trial for each trial. An extraction unit that extracts parameters indicating the status of communication traffic in each region and time zone, and learning that causes the estimator to learn using the logical value of each trial as an objective variable and the parameters of each trial as explanatory variables And the parameters in a certain region and time zone are input to the estimator, and the display latency quality of the web page in the region and time zone Having an estimation unit for estimating for.

  It is possible to estimate the quality of the display latency without the influence of individual web pages.

It is a figure which shows the hardware structural example of the estimation apparatus 10 in embodiment of this invention. It is a figure which shows the function structural example of the estimation apparatus 10 in embodiment of this invention. It is a flowchart for demonstrating an example of the process sequence which the web deterioration analysis part 11 performs. It is a flowchart for demonstrating an example of the process sequence which the traffic information extraction part 12 performs. It is a flowchart for demonstrating an example of the process sequence which the quality estimation part 13 performs in a learning stage. It is a flowchart for demonstrating an example of the process sequence which the estimation apparatus 10 performs in an estimation stage.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a diagram illustrating a hardware configuration example of an estimation apparatus 10 according to an embodiment of the present invention. The estimation device 10 in FIG. 1 is a computer having a drive device 100, an auxiliary storage device 102, a memory device 103, a CPU 104, an inter-stage device 105, and the like that are mutually connected by a bus B.

  A program that realizes processing in the 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 from another computer 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 estimation device 10 according to a program stored in the memory device 103. The inter-stage device 105 is used as an inter-stage for connecting to a network.

  FIG. 2 is a diagram illustrating a functional configuration example of the estimation device 10 according to the embodiment of the present invention. In FIG. 2, the estimation device 10 includes a web deterioration analysis unit 11, a traffic information extraction unit 12, a quality estimation unit 13, and the like. Each of these units is realized by processing that one or more programs installed in the estimation apparatus 10 cause the CPU 104 to execute.

  The web degradation analysis unit 11 displays the display waiting time at the terminal for one or more types (K types) of web pages (web contents) for each of Z trials distinguished by the combination of area (region) and time zone. Terminal measurement data D1, which is a set of measurement results, is analyzed. As a result of the analysis, the web deterioration analysis unit 11 outputs a deterioration degree, which is an example of an index indicating the quality of the display waiting time.

  The traffic information extraction unit 12 extracts (acquires) a parameter or the like indicating the state of communication traffic from the network traffic measurement data D2 in each trial area and time zone.

  In the learning stage, the quality estimation unit 13 causes the estimator to learn one or more information output from the web deterioration analysis unit 11 and information extracted by the traffic information extraction unit 12. In addition, when the traffic measurement data D2 in an arbitrary area and time zone is input in the estimation stage, the quality estimation unit 13 sets the degree of deterioration of the display waiting time of the web page in the area and time zone to 1 or more. Estimate using an estimator.

  The estimator is, for example, an SVM (Support Vector Machine). However, other estimators may be used as long as one logical value can be estimated from a numerical vector.

  An example of a method for collecting terminal measurement data D1 will be described. Here, URL_1, URL_2... URL_K: (K measurement target URLs; K is an arbitrary natural number of 1 or more (that is, one URL may be used)) is set as the URL of the measurement target web page. Suppose that The URL is not limited to a specific URL, but it is desirable to include various types of web pages that are frequently used.

  For each trial: The trial is a concept that is distinguished by a combination of an area to be measured and a time zone (period) as described above. Therefore, if at least one of the area and the time zone is different, the trial is different.

  (1) For the area / time zone corresponding to one trial, the display waiting time is measured P times for K measurement target URLs using a terminal capable of displaying a web page. That is, P × K measurements are performed for one trial. However, the value of P may not be common for all URLs. Moreover, P = 1 may be sufficient. That is, it is not always necessary to perform multiple measurements for each URL. In order to obtain the representative values of the area and time zone, it is desirable that measurements are performed at somewhat different places and times within the same area and time zone, but the measurement pattern (for example, the same area and the same time zone) And how to select the place and time) may be arbitrary, and the measurement method is not limited to a specific one. The method for acquiring the display wait time is not limited to a specific method, but a method using a time difference between navigationStart and loadEventEnd using the navigation timing API may be used. Note that the P measurements in each trial are completed within the total time width W.

  (2) Perform (1) for Z trials.

  The terminal measurement data D1 acquired as a result of the execution of (1) and (2) is a set of measurement values (display waiting time) denoted as “L (z, y, x)” below. Here, L (z, y, x) is a display waiting time in the x-th (x: 1 to P) measurement of the y-th URL (y: 1 to K) of the z-th (z: 1 to Z) trial. Indicates.

  Next, a method for collecting the traffic measurement data D2 will be described. Here, in order to evaluate the communication traffic quality information for each area and each time zone corresponding to each trial without the individual active measurement, for example, all of the area or (if sampling) one Traffic measurement data D2 is collected by connecting a DPI (Deep Packet Inspection) device to a monitoring port of a device such as a router that processes the communication traffic of each unit and monitoring the traffic using the DPI device. Therefore, the traffic measurement data D2 may be history information (log) of traffic monitored by the DPI device. The monitoring target is the traffic corresponding to each of the Z trials and the traffic in the time zone (time width W) (however, it is not limited to the traffic related to the measurement of the display waiting time). If the display waiting time measurement is mobile / LTE, traffic in a specific base station (eNB) or a specific sector (ECI) is to be monitored, and if it is a fixed network, traffic in a specific accommodating router or the like is to be monitored. However, the identification method of these base stations, sectors, accommodating routers, etc. is not limited to a predetermined one. It is assumed that those who have the authority to monitor at the upstream side of the network are naturally understood.

  Hereinafter, the process procedure which the estimation apparatus 10 performs is demonstrated. FIG. 3 is a flowchart for explaining an example of a processing procedure executed by the web deterioration analysis unit 11.

In step S101, the web deterioration analysis unit 11 calculates a reference value related to the display waiting time for each of the K URLs. For example, if it is a reference value for a URL, the median value, 10% value, minimum value, etc. of Z × P measurement values related to the URL may be used as the reference value. Alternatively, the reference value may be set in advance by the user of the estimation device 10. Hereinafter, reference values for URL_1, URL_2 ... URL_K are set as Wb_1, Wb_2,. . Let Wb_K. In this case, if the median is used as the reference value, the reference value of Wb_1 is as follows.
Wb_1 = (median value of the set of display waiting times of L (z, 1, x))
Subsequently, the web deterioration analysis unit 11 calculates “deterioration degree (Dw (z, y, x))” for each of the K URLs for each of Z trials as follows (S102).
Dw (z, y, x) = L (z, y, x) / Wb_y
That is, the deterioration degree Dw (z, y, x) is a ratio of the measured value to the reference value, and is an example of an index indicating the quality (deterioration degree) of the display wait time of the web page in the present embodiment. . Note that the degradation degree Dw (z, y, x) may be calculated by another method as long as the quality of the display waiting time of the web page can be grasped.

  Subsequently, the web deterioration analysis unit 11 calculates a “deterioration degree representative value (D_z)” for every Z trials (S103).

  More specifically, the web deterioration analysis unit 11 determines that a set of D_z = (Dw (i, j, l), i is fixed to z, and j and l are all possible values (ie, , A set including P × K elements) is extracted for all z, and the respective d_th quantile values of each set are calculated. d_th is a deterioration threshold given by the quantile, and is set in advance. For example, if d_th is 0.8 (that is, the 80th percentile), the 80% value (80th percentile) of the set of Dw (z, j, l) is calculated. When K = 1 and P = 1, the deterioration level representative value (D_z) = deterioration level (Dw (z, 1, 1) may be set.

  Subsequently, the web deterioration analysis unit 11 calculates “deterioration degree logical value (Db_ (z, s))” for each Z trials (S104). Here, s is a variable for the deterioration determination threshold for D_z. In the present embodiment, (−m, − (m−1),. ,...,..., N−1, n) are set to m + n + 1 multi-stage Tsh_s. That is, m and n are parameters that define the number (the number of steps) of the deterioration determination threshold value Tsh_s, and are set in advance. Note that n = m or n ≠ m. Further, when only one estimator is used, m = n = 0 may be used (that is, one Tsh_s may be used).

Therefore, the number of Db_ (z, s) calculated as follows for one trial is m + n + 1.
Db_ (z, s) = boole (D_z> Tsh_s)
That is, for each trial, for each deterioration determination threshold value Tsh_s, D_z is compared with the deterioration determination threshold value Tsh_s, and a logical value indicating the comparison result is substituted into Db_ (z, s).

  Db_ (z, s) is a logical value group that represents “deterioration” if true, and “non-deterioration” if false, and represents different intensity degradations with different thresholds (for each s). The deterioration level representative value D_z is independent of s (threshold type), and the deterioration determination threshold value Tsh_s is independent of trial (z).

  Next, a processing procedure executed by the traffic information extraction unit 12 will be described. FIG. 4 is a flowchart for explaining an example of a processing procedure executed by the traffic information extraction unit 12.

  In step S201, the traffic information extraction unit 12 substitutes 1 for a variable i. The variable i is a variable for identifying each trial in the description of FIG. Hereinafter, the i-th trial is referred to as “trial i”.

  Subsequently, the traffic information extraction unit 12 acquires log data corresponding to the trial i from the traffic measurement data D2 (S202). Specifically, log data corresponding to the area and time zone (total time width W) corresponding to trial i is acquired. Hereinafter, the communication traffic related to the acquired log data is referred to as “target traffic”. The area corresponding to the trial i is based on the eNB IP address of the area corresponding to the trial i when the area is a base station unit or a sector unit and the log data is assigned the IP address or CID of the eNB. Or the log data of the target traffic can be narrowed down based on the CID of the sector in the area. On the other hand, when the IP address or CID of the eNB is not given to the log data, the range that can be observed at the monitoring port of the area corresponding to the trial i is the log data regarding the area corresponding to the trial i.

  In subsequent steps S203 to S207, the traffic information extraction unit 12 extracts a parameter indicating the status of the communication traffic related to the trial i from the log data related to the target traffic.

  In step S203, the traffic information extraction unit 12 performs the following steps S203-1 to S203-3 on the basis of the in-time throughput and the total transfer amount related to the specific port number of the target traffic, thereby regarding the trial i. Calculate a representative value for low throughput.

In step S203-1, the traffic information extraction unit 12 extracts a TCP flow that satisfies the following conditions (1) and (2) from the target traffic.
(1) The TCP port is thpt_port. Here, thpt_port is one of preset parameters, and indicates the port number of the TCP throughput to be counted. In the present embodiment, the value of thpt_port is “80”.
(2) Further, if th_thpt_L and th_thpt_U are set as the preset parameters, the TCP flow corresponding to (1) is a TCP flow having a total transfer amount of th_thpt_L or more and th_thpt_U or less. Here, th_thpt_L is the lower limit flow size to be extracted, and th_thpt_U represents the upper limit flow size to be extracted. Note that if both th_thpt_L and th_thpt_U are set to 0, the condition (2) is invalidated.

  In step S203-2, the traffic information extraction unit 12 determines the throughput value (within the time) for each extracted TCP flow for each aggregation unit time τ that divides (subdivides) the time zone (total time width W) of trial i. Transfer amount / τ) is totaled, and the value of thpt_p1 quantile of the set of throughput values is calculated.

  Here, thpt_p1 and the following thpt_p2 are one of the preset parameters, and are throughput value extraction threshold values designated by quantiles (percentiles). Therefore, for example, when thpt_p1 is 0.2, the 20th percentile of the aggregated throughput value set is calculated. Also, τ is one of the preset parameters, and W = τ * κ (κ is an integer). Accordingly, step S203-2 is repeated κ times to obtain a set of κ thpt_p1 quantile values.

  In step S203-3, the traffic information extraction unit 12 calculates a thpt_p2 (for example, 0.2 etc.) quantile value in a set of κ values. The value of thpt_p2 quantile calculated in step S203-3 is a representative value of throughput (hereinafter referred to as “Th1_z” (z is 1. Z)).

  Subsequent to step S203, the traffic information extracting unit 12 executes the following steps S204-1 to S204-3 based on the RTT sample calculated from the ack response of the TCP flow in the target traffic, thereby performing trial i. The representative value of the high RTT is calculated (S204). Note that a method for calculating RTT (Round Trip Time) from a TCP packet sequence is not limited to a predetermined method. As an example of an existing method, in the packet analysis tool wireshark, tcp. analysis. There is a method of reading with ack_rtt.

  In step S204-1, the traffic information extraction unit 12 extracts an RTT data group for the target traffic for each total unit time τ of trial i.

  In step S204-2, the traffic information extraction unit 12 calculates the value of the rtt_p1 quantile of the extracted RTT data group for each aggregation unit time τ of trial i. In step S204-1, κ RTT data groups are extracted. Accordingly, in step S204-2, a set of values of κ rtt_p1 quantiles is calculated. Note that rtt_p1 and the following thpt_p2 are one of the preset parameters, and are values that specify the throughput extraction threshold value with quantile points (for example, 0.1, 0.2, etc.).

  In step S204-3, the traffic information extraction unit 12 calculates the value of the rtt_p2 quantile of κ rtt_p1 quantile values. The value of the rtt_p2 quantile calculated in step S204-3 is a representative value of high RTT (hereinafter referred to as “Rtt1_z” (z is 1. Z)).

  Subsequent to step S204, the traffic information extraction unit 12 performs the following step S205− based on the TCP retransmission occurrence ratio (ratio of retransmission packets in TCP packets) for each τ calculated from the analysis of the TCP flow in the target traffic. By executing 1 to S205-2, the representative value of the retransmission rate for trial i is calculated (S205). The method for detecting retransmission from a TCP packet sequence is not limited to a predetermined method. As an example of an existing method, in the packet analysis tool wireshark, tcp. analysis. There is a method of evaluating from parameters such as transmission.

  In step S205-1, the traffic information extraction unit 12 extracts a retransmission rate for the target traffic for each counting unit time τ. Therefore, a set of κ retransmission rate values is obtained.

  In step S205-2, the traffic information extraction unit 12 calculates the value of the ret_p quantile of κ retransmission rates. Note that ret_p is one of the preset parameters, and is a value that designates the retransmission rate extraction threshold with a quantile (eg, 0.9, 0.8, etc.). The value of the ret_p quantile is a representative value of the retransmission rate (hereinafter referred to as “Ret1_z” (z is 1... Z)).

  Subsequent to step S205, the traffic information extraction unit 12 calculates the representative value of the DNS response time by executing the following steps S206-1 to S206-3 (S206). The DNS response time is the time from the DNS search to the DNS response calculated from the DNS search analysis in the target traffic.

  In step S206-1, the traffic information extraction unit 12 extracts a DNS-RTT data set for the target traffic for each aggregation unit time τ of trial i.

  In step S206-2, the traffic information extraction unit 12 calculates the value of the dnst_p1 quantile for the extracted DNS-RTT data set for each aggregation unit time τ of trial i. In step S206-1, κ DNS-RTT data sets are extracted. Therefore, in step S206-2, a set of values of κ dnst_p1 quantiles is calculated. Note that dnst_p1 and the following dnst_p2 are one of the preset parameters, and are values that specify the DNS response time extraction threshold value with a quantile (for example, 0.9, 0.8, etc.).

  In step S206-3, the traffic information extraction unit 12 calculates the value of the dnst_p2 quantile of the κ dnst_p1 quantile values. The value of the dnst_p2 quantile is a representative value of DNS response time (DNS-RTT) (hereinafter referred to as “Drt1_z” (z is 1. Z)).

  Following step S206, the traffic information extraction unit 12 calculates the representative value of the DNS response rate by executing the following steps S207-1 to S207-3 (S207). The DNS response rate is a ratio of responses to the DNS search calculated from the DNS search analysis in the monitored traffic.

  In step S207-1, the traffic information extraction unit 12 extracts the number of DNS searches for the target traffic and the number of DNS responses to the search for each aggregation unit time τ of trial i. In addition, since the response may be a time later than τ, the time until dns_timeout elapses is counted. dns_timeout is one of the preset parameters, and is a timeout time when calculating the DNS response rate.

  In step S207-2, the traffic information extraction unit 12 calculates a response rate (number of DNS responses / number of DNS searches) for each aggregation unit time τ of trial i. There is no value for τ where the number of searches is 0. In step S207-2, κ response rates are calculated. Therefore, a set of κ response rates is obtained.

  In step S207-3, the traffic information extracting unit 12 excludes any “no value” from the set of κ response rates, and dnsr_p of κ (may be less than κ). Calculate the quantile value. dnsr_p is one of the preset parameters, and is a value that designates the DNS response rate extraction threshold using a quantile (for example, 0.2). The value of this dnsr_p quantile is a representative value of the DNS response rate (hereinafter referred to as “Drp1_z” (z is 1.... Z)).

Steps S202 to S207 are executed for all Z trials (S208, S209). As a result, the following vector is obtained as the throughput representative value Th1.
Th1 = (Th_1, Th1_2, Th1_3... Th1_z)
A plurality of sets of thpt_p1 and thpt_p2 may be set. In that case, vectors such as Th2_z and Th3_z are obtained for each group.

Further, as the high RTT representative value Rtt1, the following vector is obtained.
Rtt1 = (Rtt1_1, Rtt1_2,... Rtt1_z)
A plurality of sets of rtt_p1 and rtt_p2 may be set. In that case, vectors such as Rtt2_z and Rtt3_z are obtained for each group.

Further, the following vector is obtained as the retransmission rate representative value Ret1.
Ret1 = (Ret1_1, Ret1_2,... Ret1_z)
A plurality of values may be set for ret_p. In that case, vectors such as Ret2_z and Ret3_z are obtained for each value.

Further, as the DNS response time representative value Drt1, the following vector is obtained.
Drt1 = (Drt1_1, Drt1_2, ... Drt1_z)
A plurality of sets of dnst_p1 and dnst_p2 may be set. In that case, vectors such as Drt2_z and Drt3_z are obtained for each group.

Furthermore, the following vectors are obtained as the DNS response rate Drp1.
Drp1 = (Drp1_1, Drp1_2,... Drp1_z)
A plurality of values may be set for dnsr_p. In that case, vectors such as Drp2_z and Drp3_z are obtained for each value.

Subsequently, the traffic information extraction unit 12 outputs the following DPIvec with each vector of the representative values having a length Z as a set (matrix) (S210).
DPIvec = (Th1 (, Th2, Th3 ..), Rtt1 (, Rtt2, Rtt3 ..), Ret1 (, Ret2, Ret3 ..), Drt1 (, Drt2, Drt3), Drp1 (, Drp2, Drp3 ..) )
(, Th2, Th3 ...), (, Rtt2, Rtt3 ...), (, Ret2, Ret3 ...), (, Drt2, Drt3) and (, Drp2, Drp3 ...) are preset parameters. This indicates that the number of constituent vectors changes depending on the number of threshold values (quantile points).

  Then, the process procedure which the quality estimation part 13 performs in a learning stage is demonstrated. FIG. 5 is a flowchart for explaining an example of a processing procedure executed by the quality estimation unit 13 in the learning stage.

In step S301, the quality estimation unit 13 inputs the deterioration degree logical value Db_ (z, s), which is the following logical value vector of m + n + 1 lengths Z calculated by the web deterioration analysis unit 11.
Db_ (z, s) = (Db_ (z, −m), Db_ (z, −m + 1),... Db_ (z, n−1), Db_ (z, n))
Note that Db_ (z, s) is a vector having Z elements when s is fixed. Since s takes m + n + 1 values, Db_ (z, s) is a vector whose elements are m + n + 1 Z vectors.

Subsequently, the quality estimation unit 13 inputs the following set of representative value vectors (matrix) output by the traffic information extraction unit 12 (S302).
DPIvec = (Th1 (, Th2, Th3 ..), Rtt1 (, Rtt2, Rtt3 ..), Ret1 (, Ret2, Ret3 ..), Drt1 (, Drt2, Drt3), Drp1 (, Drp2, Drp3 ..) )
Subsequently, the quality estimation unit 13 generates m + n + 1 (that is, one or more) instances of the estimator (S303). That is, as many estimators as the number of deterioration determination threshold values Tsh_s (for each deterioration determination threshold value Tsh_s) are prepared. Each estimator is denoted as “S_s”. Here, s corresponds to m + n + 1 types of Tsh_s.

  Subsequently, the quality estimator 13 learns each S_s using Db_ (z, s) as an objective variable and DPIvec as an explanatory variable (S304). Each S_s learns about Db_ (z, s) with common s. S_s is now learned, and can be used to estimate the quality of the display latency of the web page.

  Then, the process procedure which the estimation apparatus 10 performs in an estimation stage is demonstrated. FIG. 6 is a flowchart for explaining an example of a processing procedure executed by the estimation apparatus 10 in the estimation stage.

  As a premise of the processing procedure of FIG. 6, it is assumed that traffic measurement data D2 (for example, DPI log data) corresponding to the estimation target area and time zone is collected. The traffic measurement data D2 is hereinafter referred to as “estimation target measurement data”. It is not necessary to measure the display wait time of the web page. This is because the degree of deterioration of the display waiting time is an objective variable (estimation target).

  In step S401, the traffic information extraction unit 12 inputs estimation target measurement data. In subsequent steps S402 to S406, the traffic information extraction unit 12 calculates (extracts) each representative value from the estimation target measurement data by executing the same processing as in steps S203 to S207 of FIG. However, since the estimation target area and the time zone are one, Z = 1.

  As a result, the traffic information extraction unit 12 obtains a matrix of the same format as DPIvec, with a length Z = 1, as in step S210 of FIG. Hereinafter, the matrix is referred to as “DPIvec_new”.

Subsequently, the quality estimation unit 13 inputs DPIvec_new to each S_s (each estimator) as follows, and thereby sets of logical values indicating whether or not the degradation is in accordance with each degradation determination threshold value Tsh_s ( Est_s which is a vector) is obtained (S407).
Est_s = S_s (DPIvec_new) (for multiple s)
Est_s is a set (vector) of logical values having a length of m + n + 1.

  Typically, the value of Tsh_s is true (deteriorated) for the largest s, and when s is changed so that Tsh_s decreases in order, it becomes false (non-degraded) when a certain s is θ, In the following Tsh_s, the pattern is all false (non-degraded) (however, all false, all true, or the order may be broken).

  The quality estimation unit 13 searches for such θ from Est_s, and outputs Tsh_θ corresponding to the searched θ as an “estimated degradation level” (estimated degradation level for the estimation target area and time zone). (S409). Specifically, for example, if all are false (non-degraded), the quality estimation unit 13 outputs “non-degraded”. In this case, the degradation is less than (minimum) Tsh_s. If all are true (deterioration), the quality estimation unit 13 outputs “(maximum) Tsh_s or more”. In addition, when all are not false and all are not true, even if false and true are not classified with a certain Tsh_s as a boundary, the quality estimation unit 13 checks the values in order from the side with the largest Tsh_s, Using Tsh_θ that becomes false first, “Tsh_θ or more” is output. This Tsh_θ is an estimation result of the degree of occurrence of deterioration of the display latency of the web page (which does not depend only on the individual web page).

  In this embodiment, as an example of a parameter for extracting a parameter indicating the state of communication traffic, a low throughput representative value, a high RTT representative value, a retransmission rate representative value, a DNS response time representative value, and a DNS response rate representative value However, all of these may not necessarily be used, and some (for example, any one) of them may be used. In addition, parameters other than these may be used as long as the parameters can grasp the status of communication traffic.

  As described above, according to the present embodiment, the estimator learns the relationship between the measurement value of the display latency of the web page and the information related to the communication traffic corresponding to the display latency measurement area and the time zone. Then, by inputting information related to communication traffic of unknown data to be estimated to the estimator, it is possible to estimate the degree of deterioration of the waiting time of the web page (which does not depend only on the individual page). Therefore, according to the present embodiment, it is possible to estimate the quality of the display waiting time from which the influence of individual web pages is removed.

  Further, in this embodiment, a plurality of web page servers are used for learning, and a general characteristic of “deterioration of display latency of web pages” is partially (because of the reason of the web page side instead of network origin). In order to extract by excluding the influence of what deteriorates only, by using the concept of “degradation degree” normalized by the standard for each web page and the method of extracting data by the quantile of multiple measurement data, By absorbing the difference in waiting time for each individual web page, the deterioration of the waiting time of the web page is captured. Therefore, it is possible to solve the problem that the web page server is various and the waiting time varies depending on each page even if the web page server is used at the same time and place.

  In the present embodiment, in response to instability of the contents of passive measurement data (communication data analysis log (DPI log) with different users, communication partners, and purposes), there is also a log corresponding to the deterioration of the display waiting time. This is instability compared with the case of extracting the average value etc. by extracting the traffic data quantile, which has good compatibility with deterioration information and is not easily influenced by extreme abnormal values. Can respond appropriately.

In addition, in the present embodiment, a plurality of target web pages are handled in a lump and evaluated by their statistics, although the influence on the display latency degradation differs for each individual web page even under the same NW degradation effect. By doing so, it can be quantified as “typical appearance of degradation effect”. In general, deterioration determination is performed with binary values (logical values). In this embodiment, a logical value estimator is also used. However, in this embodiment, since a quantitative value is used as the “degradation degree”, when converting to a logical value vector during learning (corresponding to a plurality of deterioration levels) By preparing a plurality of estimators having different levels, it is possible to realize the determination of the degree of deterioration. In this embodiment, the web deterioration analysis unit 11 includes a calculation unit and a comparison unit. It is an example. The traffic information extraction unit 12 is an example of an acquisition unit. The quality estimation unit 13 is an example of a learning unit and an estimation unit.

  Although the embodiments of the present invention have been described in detail above, the present invention is not limited to such specific embodiments, and various modifications can be made within the scope of the gist of the present invention described in the claims. Deformation / change is possible.

DESCRIPTION OF SYMBOLS 10 Estimation apparatus 11 Web deterioration analysis part 12 Traffic information extraction part 13 Quality estimation part 100 Drive apparatus 101 Recording medium 102 Auxiliary storage apparatus 103 Memory apparatus 104 CPU
105 Inter stage device B bus

Claims (7)

A calculation unit that calculates a representative value of the index indicating the quality of the display waiting time based on the measurement data of the display waiting time on the web page for each of a plurality of trials in which at least one of the region and the time zone is different.
For each trial, for the representative value related to the trial, a comparison unit that outputs a logical value indicating a comparison result with a threshold value;
For each trial, an acquisition unit that obtains parameters indicating the status of communication traffic in each of the regions and time zones involved in the trial;
A learning unit that learns the estimator using the logical value of each trial as an objective variable and the parameter of each trial as an explanatory variable;
An estimation unit that inputs the parameters in a certain region and time zone to the estimator and estimates the quality of display latency of the web page in the region and time zone;
The estimation apparatus characterized by having. The calculation unit, for each trial, calculates a representative value of an index indicating the quality of display latency of the web page based on measurement data of display latency of each of a plurality of web pages.
The estimation apparatus according to claim 1. For each trial, the comparison unit outputs a logical value group indicating a comparison result with a plurality of threshold values for the representative value related to the trial,
The learning unit uses the logical value group of each trial as an objective variable, the parameter of each trial as an explanatory variable, and learns a plurality of estimators corresponding to each of the plurality of threshold values,
The estimation unit inputs the parameter in a certain region and time zone to the estimator corresponding to each of the plurality of thresholds, and estimates the quality of the display waiting time of the web page in the region and time zone.
The estimation apparatus according to claim 1 or 2, wherein The acquisition unit acquires the parameter for each of a plurality of unit times that divide the time zone, calculates a representative value of the parameter acquired for each unit time,
The learning unit uses the logical value of each trial as an objective variable, and trains an estimator using a representative value of the parameter of each trial as an explanatory variable.
The estimation apparatus according to claim 1, wherein: The calculation unit calculates a representative value of an index indicating the quality of the display latency of the web page based on a plurality of measurement data of the display latency of the web page for each trial.
The estimation apparatus according to claim 1, wherein: A calculation procedure for calculating a representative value indicating the quality of the display waiting time based on the measurement data of the display waiting time on the web page for each of a plurality of trials in which at least one of the region and the time zone is different.
For each trial, for the representative value related to the trial, a comparison procedure for outputting a logical value indicating a comparison result with a threshold value;
For each trial, an acquisition procedure for obtaining parameters indicating the status of communication traffic in each of the regions and time zones involved in the trial;
A learning procedure for learning an estimator using the logical value of each trial as an objective variable and the parameter of each trial as an explanatory variable;
An estimation procedure for inputting the parameters in a certain region and time zone to the estimator to estimate the quality of display latency of the web page in the region and time zone;
Is performed by a computer. 6. A program that causes a computer to function as each unit according to claim 1.

Images