JP2005064971A - Load generation distribution estimation method and apparatus, and high load determination method and apparatus - Google Patents

Abstract

An object of the present invention is to estimate a load generation distribution in units of a short period based on easily collectable information measured at relatively long intervals so that a high load determination can be performed.
An average load amount ρi and an average resource usage qmi of a node 3 in a measurement period Ti obtained from MIB information of an arbitrary measurement period Ti by a distribution estimation unit 15B of a control unit 15, and a load in the node 3 From the relational expression showing the relationship between the amount and the average resource usage, the load generation distribution Qi in the short period Ts in the measurement period Ti is estimated, and the cumulative distribution calculation means 15C is used for each measurement period T1 to Tn. The accumulated load generation distribution Qa corresponding to the determination period Ta is calculated by averaging the load generation distributions Q1 to Qn estimated in step (b).
[Selection] Figure 2

Description

  The present invention relates to a technique for managing a network, a server, and the like in a service such as content distribution using a packet switching network such as the Internet, and more particularly to a technique for determining a load status of a packet switching network.

In general, a service provider who provides a content providing service to a user via a packet switching network such as the Internet checks whether the provided service is provided with a quality that can satisfy the user. The load status for the resource is monitored. Conventionally, such load status monitoring is generally performed by installing a line load measuring device in a packet switching network to be managed.
However, since this method requires equipment costs, as an alternative, MIB (Management Information Base) information is acquired as node operation status information from routers and switches that constitute the node of the packet switching network, and the packet switching network A method for monitoring the load status on resources has been proposed (see, for example, Non-Patent Document 1).

The applicant has not yet found prior art documents related to the present invention by the time of filing other than the prior art documents specified by the prior art document information described in this specification.
Chiba, Tokuhisa, Nogami, Abe, "A method for determining bottlenecks in managed networks using MIB information", 2002 IEICE Communication Society Conference, B-7-71, p231

However, in such a conventional technique, the information obtained by the MIB information is an average value during the measurement period, and thus there is a possibility that an instantaneous quality deterioration in a period shorter than the measurement period may be missed. For example, when distributing contents such as audio and video, it is necessary to be able to detect short-term traffic fluctuations of several milliseconds to several seconds so that the voice and video are not interrupted. Since the information collection interval is every several minutes to several hours, quality deterioration due to such short-term traffic fluctuations cannot be detected.
The present invention is for solving such a problem, and a load generation distribution estimation method capable of estimating a load generation distribution in a short period unit based on easily collectable information measured at relatively long intervals, and It is an object of the present invention to provide an apparatus and a high load determination method and apparatus capable of determining whether or not a high load is generated in a short period unit.

  In order to achieve such an object, a load distribution estimation method according to the present invention includes a plurality of measurement periods based on operation state information obtained for each predetermined measurement period from a node that performs packet transfer processing in a packet switching network. In a load distribution estimation method for estimating a load generation distribution at a node corresponding to a determination period consisting of: an average load amount and an average resource usage of a node in the measurement period obtained from operation state information of an arbitrary measurement period; A distribution estimation step for executing a first step of estimating a load generation distribution in units of a period shorter than the measurement period in the measurement period from a relational expression indicating a relation between the load amount and the average resource usage in the node; Cumulative negative to calculate the cumulative load generation distribution corresponding to the judgment period by accumulating the load generation distribution estimated for each measurement period In which and a generation distribution calculating step.

In the distribution estimation step, load generation in the measurement period is further performed by using an average quality in the measurement period obtained from operation state information in an arbitrary measurement period and a relational expression indicating a relationship between the load amount and the quality in the node. You may make it perform the 2nd step which estimates distribution.
At this time, the distribution estimation step may execute the second step when the average quality is larger than a predetermined threshold value. When the average load amount is larger than the predetermined load amount, Steps may be executed.

  As a specific example of the first step, the average load amount ρi, the average resource usage amount qmi, the load amount ρmin that is smaller than the average load amount ρi and does not cause quality degradation, the larger than the average load amount ρi, and the load amount and the quality In the relationship, the load amount ρr, the load occurrence frequency X1i of the load amount ρmin, the load occurrence frequency X2i of the load amount ρr, and the load generation of the average load amount ρi are smaller than the critical point where the quality starts to deteriorate rapidly as the load amount increases. {X1i, X2i, Yi} may be estimated as the load generation distribution Qi in the measurement period Ti based on the later-described Equation 3 regarding the frequency Yi.

  As a specific example of the second step, the average load amount ρi, the average quality Li, the average resource usage qmi, the load amount ρmin that is smaller than the average load amount ρi and does not cause quality degradation, the larger than the average load amount ρi, and the quality degradation The load ρmax indicating the allowable limit value of the load, the load that is located between the load ρmin and the load ρmin, and the load is smaller than the critical point where the quality starts to deteriorate rapidly as the load increases in relation to the load This measurement is based on the following equation 4 regarding the load occurrence frequency X1i of the load ρr, the load amount ρmin, the load occurrence frequency X2i of the load amount ρr, the load occurrence frequency X3i of the load amount ρmax, and the load occurrence frequency Yi of the average load ρi {X1i, X2i, X3i, Yi} may be estimated as the load generation distribution Qi in the period Ti.

  The load distribution estimation apparatus according to the present invention is connected to a node that performs packet transfer processing in a packet switching network, and includes a plurality of measurement periods based on operation state information obtained for each predetermined measurement period from the node. A load distribution estimation apparatus that estimates a load generation distribution of a node corresponding to a determination period includes a control unit that estimates a cumulative load generation distribution corresponding to the determination period using the load distribution estimation method described above.

  Further, the high load determination method according to the present invention is based on the operation state information obtained for each predetermined measurement period from the node that performs packet transfer in the packet switching network, and the high load in units of a period shorter than the measurement period for the node. In the high load determination method for determining the presence or absence of occurrence, an estimation step for estimating a cumulative load generation distribution corresponding to a determination period composed of a plurality of measurement periods using the load distribution estimation method, and a load estimated by the estimation step A determination step for determining whether or not a high load has occurred by comparing a time ratio in which the quality deteriorates from a preset quality target value in the occurrence distribution and a quality deterioration allowable time ratio with respect to the quality target value; Is.

  In addition, the high load determination device according to the present invention is connected to a node that performs packet transfer in a packet switching network, and has a period shorter than the measurement period for the node based on the operation state information obtained for each predetermined measurement period from the node. In the high load determination apparatus for determining the presence or absence of high load occurrence in units of the above, a control unit for determining the presence or absence of high load using the above high load determination method is provided.

According to the present invention, it is possible to determine a high load of a resource of a packet switching network without overlooking instantaneous quality degradation only by measuring a long-term interval of MIB information. As a judgment period, for example, by collecting MIB information at intervals that can be collected relatively easily, such as every measurement period of several minutes to several hours over a unit of one day to several days, quality degradation of several milliseconds to several seconds can be achieved. The load determination can be performed without overlooking. As a result, even for services that require short-term quality degradation detection, it is possible to perform high-load detection without installing a measurement device in the packet-switched network, so services provided via a large-scale network The effect of drastically reducing the quality control cost can be obtained.
In addition, when estimating the load generation distribution for each measurement period, the load generation distribution is estimated using a relational expression indicating the relationship between the load amount ρ and the average resource usage. Even in a small region, it is possible to estimate the load generation distribution in a short period unit with high accuracy.

Next, embodiments of the present invention will be described with reference to the drawings.
First, a high load determination device according to an embodiment of the present invention will be described with reference to FIG. FIG. 1 is a conceptual diagram showing a configuration example of a high load determination system of a packet switching network using a high load determination device according to an embodiment of the present invention.
In the figure, the high load determination device 1 is configured to operate from the plurality of nodes 3 such as routers that configure the packet switching network 2 to be managed and perform packet transfer processing, respectively, via a predetermined communication network. It is configured to collect MIB (Management Information Base) information as information. In this case, an IP network including a node 3 such as a plurality of IP (Internet Protocol) routers and a link connecting them is shown as the packet switching network 2.

  In such a configuration, the high load determination device 1 according to the present exemplary embodiment has a predetermined period of about 1 day to several weeks, for example, a measurement period of about 5 minutes to 1 hour from the node 3 by SNMP (Simple Network Management Protocol). By collecting the MIB information 4 including the average load amount and quality data, it is determined whether or not a high load that causes deterioration in quality within a short period of several milliseconds to several seconds has occurred. is there. Here, “degradation of quality” refers to a state in which a service interruption or delay that is unacceptable to the user occurs in a service provided via the managed packet switching network 2, and the amount of load caused by this state is high. It is defined as

[Configuration of high load judgment device]
FIG. 2 is a block diagram illustrating a functional configuration of the high load determination device according to the present embodiment. As shown in FIG. 2, the high load determination device 1 includes a computer as a whole, and includes a communication interface unit (hereinafter referred to as an I / F unit) 11, a screen display unit 12, an operation input unit 13, a storage unit 14, and A control unit 15 is provided.
The communication I / F unit 11 is a circuit unit that collects MIB information from an arbitrary node 3 using SNMP. The screen display unit 12 includes a screen display device such as a CRT or LCD, and displays the estimated load generation distribution and the result of the high load determination on the screen in response to an instruction from the control unit 15. The operation input unit 13 includes a keyboard and a mouse pointer, detects an operation of the administrator, and outputs it to the control unit 15.

The storage unit 14 includes a storage device such as a hard disk and a memory, and stores various processing information and programs used for processing in the control unit 15.
The control unit 15 includes a microprocessor such as a CPU and peripheral circuits thereof, and reads and executes a program in the storage unit 14 to realize various functional units by causing the hardware and the program to cooperate with each other. Part.

Functional means realized by the control unit 15 include data acquisition means 15A, distribution estimation means 15B, cumulative distribution calculation means 15C, and high load determination means 15D.
The data acquisition unit 15 </ b> A collects MIB information from the node 3 using SNMP by performing data communication with an arbitrary node 3 via the communication I / F unit 11. The distribution estimation unit 15B estimates the load generation distribution in units of short periods shorter than the measurement period for each measurement period based on the MIB information collected for each measurement period by the data acquisition unit 15A. The cumulative distribution calculation unit 15C calculates the cumulative load generation distribution from the high load generation distribution in each measurement period estimated by the distribution estimation unit 15B. The high load determination means 15D determines the presence or absence of high load occurrence in a short period based on the cumulative load occurrence distribution calculated by the cumulative distribution calculation means 15C.

[High load judgment processing]
Next, with reference to FIG. 3 and FIG. 4, a high load determination process will be described as an operation of the high load determination device according to the present embodiment. FIG. 3 is a flowchart showing the high load determination process. FIG. 4 is an explanatory diagram showing a high load determination process.
The control unit 15 starts the high load determination process of FIG. 3 periodically or in response to an instruction from the administrator input from the operation input unit 13.
First, the data acquisition unit 15A acquires MIB information by performing data communication with the monitoring target node 3 via the communication I / F unit 11, and calculates measurement data used by the distribution estimation unit 15B (step 100). ). At this time, as shown in FIG. 4, a plurality of measurement periods T1 to Tn (n is an integer of 2 or more) are provided within a predetermined determination period Ta, and each measurement period Ti (i is an integer of 1 to n) is provided. Obtain MIB information.

The data acquisition unit 15A calculates the average load amount ρi, the average quality Li, and the average resource usage qmi for the measurement period Ti as measurement data.
As the average load amount ρi, for example, an average line usage rate (%) at the target node 3 may be used. This average line usage rate can be calculated from the transfer information amount (ifOutOctetc) and interface speed (ifSpeed) for each line interface in the node 3 obtained by the standard MIB. The transfer information amount is counter information at the time of SNMP polling, and indicates an average characteristic between two polling times.

As the average quality Li, for example, an average packet loss rate (%) at the target node 3 may be used. This average packet loss rate can be calculated from the number of lost packets (ifOutDiscards, ifOutErrors) and the number of transferred packets (ifOutUcastPkts, ifOutNUcastPkts) for each line interface in the node 3 obtained by the standard MIB.
As the average resource usage qm, for example, the average line transmission waiting buffer usage rate (%) at the target node 3 may be used. The average line transmission waiting buffer usage rate can be calculated from the ratio of the current queue length of the line transmission waiting buffer in the node 3 and the buffer size obtained by the node's QoS command.

If the queue length of the line transmission waiting buffer cannot be obtained, the average line transmission waiting buffer usage rate may be calculated from the packet transfer processing time T at the target node 3 and the average packet length H of the transfer packet. The packet transfer processing time T may be obtained by actually transferring the packet at the target node and actually measuring the packet transmission / reception time difference between the input side and the output side at that time. The average packet length H is calculated from the MIB information used for calculating the line usage rate and the packet loss rate.
H (bit) = ifOutOctets × 8
/ [(ifOutUcastPkts + ifOutNUcastPkts)-(ifOutDiscards + ifOutError)]
It can be calculated by And C is the line speed (bps) and K is the buffer size.
qm = 100 × C × T / [H × (K + 1)]
Thus, the average line transmission waiting buffer usage rate qm may be calculated.

In this way, by executing the load distribution estimation process in the distribution estimation means 15B based on the average load amount ρi, average quality Li, and average resource usage qmi obtained for each measurement period Ti, the node 3 For each measurement period Ti, a load generation distribution (probability distribution) Qi in a short period Ts shorter than the measurement period is estimated (step 101).
In this load distribution estimation process, a function qm = G (ρ) indicating the relationship between the load amount ρ and the average resource usage qm, and further a function L = F (ρ) indicating the relationship between the load amount ρ and the quality L are obtained. Use.

Usually, the function L = F (ρ) has a critical point R. When the load amount ρ exceeds the critical point R, the quality L rapidly deteriorates, and the quality L until the load amount ρ exceeds the critical point R. Has a characteristic of almost no loss (packet loss rate≈0).
Therefore, in the region where the load amount ρ is smaller than the critical point R, the quality L hardly changes and it is difficult to capture the load generation distribution in units of short periods. In the present embodiment, even in a region where the load amount ρ is smaller than the critical point R, attention is paid to the average resource usage amount qm that increases or decreases with respect to the change in the load amount ρ, and the load amount ρ and the average resource usage amount qm By estimating the load generation distribution Qi using the function qm = G (ρ) indicating the relationship, it is possible to estimate the load generation distribution in a short period unit with high accuracy. Details of the load distribution estimation process will be described later.

In this way, after estimating the load generation distribution Qi for each measurement period Ti, the cumulative distribution calculation means 15C averages (accumulates) these load generation distributions and loads the load generation distribution corresponding to the determination period Ta, that is, cumulative. A load generation distribution Qa is calculated (step 102).
Then, the cumulative load generation distribution Qa is converted into a load generation distribution QL for the quality L, and the load generation distribution QL is calculated based on the quality target value Lt set in advance as a high load determination condition and the high load generation probability P. Inspection is performed to determine whether or not a high load has occurred in a short period (step 103), and a series of high load determination processing is terminated.

Thus, in the present embodiment, the average load amount ρi, the average quality Li, and the average resource usage amount qmi are collected from the target node for each measurement period, and the function qm = G (ρ) and the function L = Since the load generation distribution Qi is estimated for each measurement period using F (ρ), the cumulative load generation distribution Qa corresponding to the determination period Ta is calculated from these Qi, and the high load determination is performed. It is possible to determine the high load of the resources of the packet switching network without overlooking instantaneous quality degradation only by measuring the long-term interval of information.
As the determination period Ta, for example, over several days to several days, the MIB information is collected at a relatively easily collectable interval of several minutes to several hours of measurement periods T1 to Tn. It is possible to make a load determination without overlooking the quality degradation.

As a result, even for services that require short-term quality degradation detection, it is possible to perform high-load detection and determination without installing a measurement device in the packet switching network. The effect of greatly reducing the quality control cost of the service that is provided.
In addition, when estimating the load generation distribution for each measurement period, the load generation distribution Qi is estimated using the function qm = G (ρ) indicating the relationship between the load amount ρ and the average resource usage qm. Even in a region where the load amount ρ is smaller than the critical point R, the load generation distribution in a short period can be estimated with high accuracy.

[Load distribution estimation processing]
Next, the load distribution estimation process executed by the distribution estimation unit 15B will be described with reference to FIG. FIG. 5 is a flowchart showing the load distribution estimation process.
The distribution estimation means 15B first reads the average load amount ρi, average quality Li, and average resource usage qmi of the measurement period Ti of the node 3 to be estimated from the storage unit 14 (step 110). It is assumed that the measurement data for each measurement period Ti calculated based on the MIB information acquired from the node 3 by the data acquisition unit 15A is already stored in the storage unit 14.

Next, the distribution estimation means 15B determines whether Li is ε or less and ρi is ρr or less (step 111). This ε is a threshold value for defining a region where the load amount ρ is smaller than the critical point R by the value of the quality L, and is set in the storage unit 14 in advance. Similarly, ρr is a threshold value for defining a region where the load amount ρ is smaller than the critical point R by the load amount ρ, and is set in the storage unit 14 in advance.
Here, when Li is ε or less and ρi is ρr or less (step 111: YES), the load generation distribution is estimated using only the function qm = G (ρ) (step 112). The load distribution estimation process for the measurement period Ti is terminated.

Here, the load generation distribution estimation process (first step) in step 112 will be described with reference to FIG. FIG. 6 is an explanatory diagram showing a load generation distribution estimation process when only the function qm = G (ρ) is used. 6A is a characteristic diagram showing characteristics of the average quality L and the average resource usage qm with respect to the load quantity ρ, and FIG. 6B is a frequency of load occurrence in a short period with respect to the load quantity ρ. FIG.
Region E represents a region where Li is ε or less and ρi is ρr or less. R is a critical point.

When estimating the load generation distribution using only the function qm = G (ρ), the frequency of load generation is estimated for the preset load amounts ρmin, ρr and the average load amount ρi for each measurement period Ti. Note that ρmin is a load amount ρ that gives the minimum value (best value) of the quality L necessary for high load determination as ρ that does not cause quality degradation. For example, ρmin is the most actually measured in each measurement period T1 to Tn. What is necessary is just to use the value below a small average load amount. Ρr is larger than ρi and smaller than the critical point R.
When the average load amount ρi, the average quality Li, and the average resource usage amount qmi are given for the measurement period Ti, the load occurrence frequencies X1i, X2i, Yi corresponding to the load amounts ρmin, ρr and the average load amount ρi are: It has the following equation (3).

  In Equation 3, G (ρmin), G (ρr), and G (ρi) are known by using qm = G (ρ), and by solving these as simultaneous equations, unknown load occurrence frequencies X1i, X2i, The value of Yi is obtained. When these load occurrence frequencies are captured on a time axis, it can be regarded as a distribution of loads generated for each short period Ts within the measurement period Ti.

On the other hand, if Li is not less than ε or ρi is not less than ρr in step 111 of FIG. 5 (step 111: NO), is Li larger than ε and there are solutions X1i, X2i, X3i, Yi? Judgment is made (step 113).
If Li is larger than ε and solutions X1i, X2i, X3i, and Yi exist (step 113: YES), a load is generated using the function L = F (ρ) and the function qm = G (ρ). The distribution is estimated (step 117), and the load distribution estimation process for the measurement period Ti is terminated.

Here, the load generation distribution estimation process (second step) in step 117 will be described with reference to FIG. FIG. 7 is an explanatory diagram showing a load generation distribution estimation process when the function L = F (ρ) and the function qm = G (ρ) are used.
7A is a characteristic diagram showing characteristics of the average quality L and the average resource usage qm with respect to the load quantity ρ. FIG. 7B shows the load occurrence frequency in a short period with respect to the load quantity ρ. It is a probability distribution diagram shown.

When estimating the load generation distribution using the function L = F (ρ) and the function qm = G (ρ), the preset load amounts ρmin, ρr, ρmax and the average load amount ρi for each measurement period Ti The frequency of load occurrence is estimated for. Note that ρmax is a load amount ρ that gives the minimum value (worst value) of the quality L necessary for high load determination as ρ, which is an allowable limit value for quality degradation, and is actually measured, for example, in each measurement period T1 to Tn. A value equal to or greater than the largest average load amount may be used.
When the average load amount ρi, the average quality Li, and the average resource usage amount qmi are given for the measurement period Ti, the load occurrence frequencies X1i, X2i, and X3i corresponding to the load amounts ρmin, ρr, ρmax, and the average load amount ρi. , Yi have the following relationship:

In Equation 4, by using L = F (ρ) and qm = G (ρ), F (ρmin), F (ρr), F (ρmax), F (ρi), G (ρmin), G (ρr ), G (ρmax), G (ρi) are known, and unknown load generation frequencies X1i, X2i, X3i, Yi are obtained by solving these as simultaneous equations. When these load occurrence frequencies are captured on a time axis, it can be regarded as a distribution of loads generated for each short period Ts within the measurement period Ti.
As described above, when the point (ρi, Li) is outside the region E and the solutions X1i, X2i, X3i, Yi are obtained, the load is calculated using L = F (ρ) and qm = G (ρ). Since the generation distribution is estimated, the load generation distribution can be estimated with high accuracy from two functions in a wide range including a region where the load amount ρ exceeds the critical point R.

  Note that whether or not the solutions X1i, X2i, X3i, Yi are obtained in step 113 is, for example, that the point (ρi, Li) is the point (ρmin, L (ρmin)), the point (ρi, L (ρi) ), Points (ρr, L (ρr)), points (ρmax, L (ρmax)), and the points (ρi, qmi) are points (ρmin, G (ρmin)). ), A point (ρi, G (ρi)), a point (ρr, G (ρr)), and a point (ρmax, G (ρmax)).

On the other hand, if there is no solution X1i, X2i, X3i, Yi in step 113 of FIG. 5 (step 113: NO), it is determined whether or not the average quality Li is greater than F (ρi) (step 114).
Here, if Li> F (ρi) (step 114: YES), ρmax is corrected to ρmax ′ on the safe side in the quality determination (step 115), and the function L = F (ρ) and the function qm = The load generation distribution is estimated using G (ρ) (step 117), and the load distribution estimation process for the measurement period Ti is terminated.

Here, with reference to FIG. 8, the correction of ρmax in step 115 will be described. FIG. 8 is an explanatory diagram showing the correction process of ρmax.
8A is a characteristic diagram showing characteristics of the average quality L and the average resource usage qm with respect to the load quantity ρ, and FIG. 8B shows the load occurrence frequency in a short period with respect to the load quantity ρ. It is a probability distribution diagram shown.

As described in FIG. 7, when the point (ρi, Li) exists outside the convex region V, the solutions X1i, X2i, X3i, Yi cannot be obtained. At this time, if Li> F (ρi), the point (ρi, Li) can be placed inside the convex region V by correcting ρmax in the direction of increasing, and the solutions X1i, X2i, X3i, Yi Will be obtained. It should be noted that correcting ρmax in the increasing direction selects ρmax ′ on the safer side in quality determination and does not adversely affect high load determination.
In this way, by adjusting ρmax, it is possible to increase the estimation period that can be estimated, and it is possible to realize more stable high load determination.

On the other hand, if Li> F (ρi) is not satisfied in step 114 of FIG. 5 (step 114: NO), after replacing the average quality Li with F (ρi) (step 116), the function L = F (ρ) is obtained. The load generation distribution is estimated using the function qm = G (ρ) (step 117), and the load distribution estimation process for the measurement period Ti is terminated.
Thereby, the estimation period which can be estimated can be increased and the more stable high load determination can be realized.

[Cumulative load generation distribution calculation processing]
Next, a cumulative load occurrence distribution calculation process in the cumulative distribution calculation unit 15C will be described with reference to FIG. FIG. 9 is an explanatory diagram showing a cumulative load generation distribution calculation process.
In the cumulative distribution calculation means 15C, the load generation distribution Qi (X1i, X2i, X3i, Yi) estimated for each measurement period Ti by the distribution estimation means 15B is averaged to calculate the cumulative load generation distribution Qa.

In this case, as shown in FIG. 9, the cumulative distribution calculating unit 15 </ b> C has a load distribution ρ distribution based on the data sections D <b> 1 to Dm (m is a natural number) preset in the storage unit 14. And a cumulative load occurrence distribution Qa composed of relative frequency distributions in the respective data sections D1 to Dm is calculated.
Here, the data section D1 is specified to include ρmin, and the data section Dm is specified to include ρmax. The section widths of the data sections D1 to Dm may be equal or different.

  About the relative frequency of the load amount of each data section D1-Dm, it calculates using Formula 5-7. In particular, the data interval D1 including ρmin is obtained from Equation 5, and the data interval Dm including ρmax is obtained from Equation 6. Further, other data sections D2 to D (m−1) are obtained by Equation 7. In Equations 5 to 7, j represents the jth measurement period Tj in the determination period Ta, and takes an integer of j = 1 to n. Here, n is the number of effective measurements within the determination period Ta.

  In this way, the cumulative distribution calculation means 15C averages the load generation distribution Qi (X1i, X2i, X3i, Yi) estimated for each measurement period Ti by the distribution estimation means 15B, as shown in FIG. The cumulative load generation distribution Qa in the plurality of measurement periods T1 to Tn, that is, the determination period Ta is calculated.

[High load judgment processing]
Next, with reference to FIG. 11, the high load determination process in the high load determination means 15D will be described. FIG. 11 is an example of the load generation distribution QL for the quality L obtained by conversion by the high load determination means 15D.
The high load determination unit 15D converts the cumulative load generation distribution Qa calculated by the cumulative distribution calculation unit 15C into a load generation distribution QL for the quality L as shown in FIG. 11 based on the function L = F (ρ). .

  Then, the time ratio P in which the quality has deteriorated from the quality target value Lt set in advance as a high load determination condition in the storage unit 14 is represented as a data section on the quality deterioration side from the quality target value Lt, in the example of FIG. m-2) to Dm are determined from the total area ratio of the relative frequencies. Then, the quality deterioration allowable time ratio Pt preset in the storage unit 14 is compared with the actual time ratio P. When Pt <P, a high load is generated in a short period Ts shorter than the measurement period Ti. Judge.

In the above description, the high load determination device (method) is described as an example. However, the load generation distributions Q1 to Qn are estimated for a plurality of measurement periods T1 to Tn, and the cumulative load corresponding to the determination period Ta is determined from these Q1 to Qn. The load distribution estimation process for calculating the occurrence distribution Qa may be used independently for apparatuses and processes other than the high load determination, and the same operational effects as the operational effects relating to the load distribution estimation described above can be obtained.
Further, when estimating the load generation distribution Qi in the measurement period Ti using qm = G (ρi), Qi may be estimated by using Equation 4 on the assumption that X3i = 0 instead of Equation 3. .

In the present embodiment, for the function L = F (ρ) used in the load distribution estimation process, the quality characteristics of the router and the packet transmission line are evaluated on the desk or by verification. For example, using a queuing model, if the packet input process to the router is a Poisson process, the packet length is an exponential distribution, and the buffer size is K (M / M / 1 / K model), F (ρ ) = Ρ ^ K · (1-P) / (1-ρ ^ K). However, ρ ^ K indicates ρ to the Kth power.
When using real machine verification, L = F (ρ) is obtained by multivariate regression analysis based on data of measurement results {ρi, Li, qmi} i = 1 to K in a real machine in a steady state.
For the function qm = G (ρ) used in the load distribution estimation process, for example, when the resource usage is qm, the relationship between the packet transmission line usage rate and the queue length in the transmission waiting buffer is the same as above. = G (ρ) is obtained.

It is a conceptual diagram which shows the high load determination system of the packet switching network using the high load determination apparatus concerning one embodiment of this invention. It is a block diagram which shows the structure of the high load determination apparatus concerning one embodiment of this invention. It is a flowchart which shows the high load determination process in the high load determination apparatus concerning one embodiment of this invention. It is explanatory drawing which shows a high load determination process. It is a flowchart which shows a load distribution estimation process. It is explanatory drawing which shows the estimation process of load generation distribution in the case of using only function qm = G (ρ). It is explanatory drawing which shows the estimation process of load generation distribution in the case of using a function L = F (ρ) and a function qm = G (ρ). It is explanatory drawing which shows the correction process of (rho) max. It is explanatory drawing which shows a cumulative load generation distribution calculation process. It is a load accumulation load generation distribution corresponding to a judgment period. It is an example of load generation distribution with respect to quality.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... High load determination apparatus, 2 ... Packet switching network, 3 ... Node, 4 ... MIB information, 11 ... Communication I / F part, 12 ... Screen display part, 13 ... Operation input part, 14 ... Memory | storage part, 15 ... Control 15A ... data acquisition means, 15B ... distribution estimation means, 15C ... cumulative distribution calculation means, 15D ... high load determination means, Ta ... determination period, Ti, T1 to Tn ... measurement period, Ts ... short period, ρi, ρ1 ~ Ρn ... average load, Li, L1 to Ln ... average quality, qmi, qm1 to qmn ... average resource usage, Qi, Q1 to Qn ... load generation distribution, Qa ... cumulative load generation distribution, QL ... load generation distribution ( Quality), X1i, X2i, X3i, Yi ... Frequency, D1 to Dm ... Data section, Lt ... Quality target value, Pt ... Quality degradation allowable time ratio.

Claims (9)

A load distribution for estimating a load generation distribution in the node corresponding to a determination period consisting of a plurality of measurement periods based on operation state information obtained for each predetermined measurement period from a node that performs packet transfer processing in a packet switching network In the estimation method,
From the average load amount and average resource usage amount of the node in the measurement period obtained from the operation state information of an arbitrary measurement period, and a relational expression indicating the relationship between the load amount and average resource usage amount in the node, A distribution estimation step for executing a first step of estimating a load generation distribution in units of a period shorter than the measurement period in the measurement period;
A load distribution estimation method comprising: a cumulative load generation distribution calculating step of calculating a cumulative load generation distribution corresponding to the determination period by accumulating the load generation distribution estimated for each measurement period. The load distribution estimation method according to claim 1,
The distribution estimation step further uses an average quality in the measurement period obtained from the operation state information in an arbitrary measurement period, and a relational expression indicating a relationship between a load amount and quality in the node, A load distribution estimation method comprising: executing a second step of estimating the load generation distribution in The load distribution estimation method according to claim 2,
The load estimating method is characterized in that the distribution estimating step executes the second step when the average quality is larger than a predetermined threshold value. The load distribution estimation method according to claim 2,
In the distribution estimation step, the second step is executed when the average load amount is larger than a predetermined load amount. The load distribution estimation method according to claim 1,
The first step includes the average load amount ρi, the average resource usage amount qmi, the load amount ρmin that is smaller than the average load amount ρi and does not cause quality degradation, the larger than the average load amount ρi, and the load amount and quality. The load amount ρr that is smaller than the critical point where the quality starts to deteriorate rapidly as the load amount increases, the load occurrence frequency X1i of the load amount ρmin, the load occurrence frequency X2i of the load amount ρr, and the average load The following relational expression regarding the load occurrence frequency Yi of the quantity ρi

Then, {X1i, X2i, Yi} is estimated as the load generation distribution Qi in the measurement period Ti. In the load distribution estimation method according to any one of claims 2 to 4,
In the second step, the average load amount ρi, the average quality Li, the average resource usage amount qmi, the load amount ρmin that is smaller than the average load amount ρi and does not cause quality degradation, the larger than the average load amount ρi, and A load amount ρmax indicating an acceptable limit value of quality deterioration, a critical value that is located between the load amount ρmin and the load amount ρmin, and that the quality starts to deteriorate rapidly as the load amount increases in the relationship between the load amount and the quality The load amount ρr smaller than the point, the load occurrence frequency X1i of the load amount ρmin, the load occurrence frequency X2i of the load amount ρr, the load occurrence frequency X3i of the load amount ρmax, and the load occurrence frequency Yi of the average load amount ρi Relational expression

Then, {X1i, X2i, X3i, Yi} is estimated as the load generation distribution Qi in the measurement period Ti. A load generation of the node corresponding to a determination period composed of a plurality of measurement periods based on operation state information obtained from the node for each predetermined measurement period connected to a node that performs packet transfer processing in a packet switching network In the load distribution estimation device for estimating the distribution,
A load distribution estimation apparatus comprising: a control unit that estimates a cumulative load generation distribution corresponding to the determination period using the load distribution estimation method according to claim 1. High load determination method for determining presence / absence of occurrence of high load for each node in a unit shorter than the measurement period based on operation state information obtained for each predetermined measurement period from a node that performs packet transfer in a packet switching network In
An estimation step of estimating a cumulative load generation distribution corresponding to the determination period including the plurality of measurement periods using the load distribution estimation method according to claim 1;
By comparing the time ratio at which the quality deteriorates from a preset quality target value in the load generation distribution estimated at the estimation step, and the quality deterioration allowable time ratio with respect to the quality target value, the presence or absence of the high load occurrence And a determination step for determining the high load. Presence or absence of occurrence of high load in units of a period shorter than the measurement period for the node based on operation state information obtained from the node every predetermined measurement period connected to a node that performs packet transfer in a packet switching network In the high load determination device for determining
A high load determination apparatus comprising: a control unit that determines whether or not the high load is generated using the high load determination method according to claim 8.

Images