JP6430360B2 - Simulation device, simulation method, and simulation program - Google Patents

Description

  The present invention relates to an apparatus, a method, and a program for calculating a loss rate when a statistically effective number of samples are obtained in a traffic simulation.

  For example, video distribution is actively performed on the Internet. In addition to private networks such as an in-house IP (Internet Protocol) network, video and audio for training, instruction, lectures, etc. Multi-point distribution (multicast) to PCs and the like is becoming widespread.

  Multicast packets, such as video and audio, sent from a multicast server that performs multi-point distribution via a network are UDP (User Datagram Protocol) and are not retransmitted by TCP. Therefore, FEC (Forward Error Correction) by video applications, etc. Even if the protection works, the effect of packet loss may appear in the quality of video and audio. For this reason, it is important to predict in advance packet loss that occurs in the network in accordance with various traffic forms and traffic volumes.

  Various approximate calculation formulas have been proposed so far, but in most cases it is unknown how much error there is depending on the traffic type and traffic volume. It is necessary to verify the correctness of the approximate calculation formula. Thus, traffic simulation is still an important means even today.

  The datagram referred to here includes packets of various communication protocols including IP packets, Ethernet frames, ATM (Asynchronous Transfer Mode) cells, X. It refers to a batch of bit strings for transferring some information such as 25 packets.

  When the packet loss is obtained by performing simulation, the simulation time until one loss is obtained increases as the buffer length increases, that is, as the loss rate becomes extremely small. For this reason, in order to shorten the simulation time, we want to obtain the datagram loss rate with as few losses as possible. However, we have not found a specific method for how many datagrams are lost statistically. As described in Non-Patent Document 1 and Non-Patent Document 2, simulations were conducted empirically.

  Therefore, in order to shorten the simulation time, a method of obtaining a datagram loss rate with the smallest possible number of losses using a statistical method has been devised (see, for example, Patent Document 1).

JP2013-192005A

Murata Masayuki, Osaki Hiroyuki, Harai Hiroaki, Miyahara Hideo, “Communication Traffic Theory and Its Applications [VI] -Simulation Techniques and Their Applications,” IEICE Journal, vol. 78, no. 3, pp. 264-270, Mar. 1995. Koji Ogawa, Kenji Nakagawa, “Optimum Simulation Distribution in Importance Sampling Simulation Method”, Shingaku Sodai, Telecommunications (2), B-646, p. 78, Mar. 1996. R. S. Burlington C. Mei, translated by Tomio Hayashi and Kazumasa Wakimoto, “Probability and Statistical Analysis Handbook”, Morikita Publishing, 1975.

In the method of Patent Document 1, if the simulation end condition is satisfied when the number of datagram loss L _j in FIG. 1 is satisfied, the number of datagram inputs N _i generated in the simulation from the start until then is used.
(Equation 1)
R = L _j / N _i (1)
As a result, the loss rate R of the datagram is obtained.

  However, it is considered that a more accurate loss rate can be obtained by calculating the loss rate R by using the loss number L generated so far for an arbitrary number N of inputs.

Here, it is assumed that the next (L _j +1) th loss occurs when the number of inputs is (N _i + k + 1). If the number of inputs for calculating the loss rate is determined randomly, regardless of the loss number L _j ,
Probability calculating the loss rate when the number of inputs N _i P _i,
Probability P _{i + 1} for calculating the loss rate when the number of inputs is N _i +1,
...
Probability calculating the loss rate when the number of inputs _{N i + k P i + k} ,
Are all considered equal.

The number of losses in the above case is all L _j , and the average number of these inputs is (N _i + k / 2). Therefore, the more accurate loss rate R is (Equation 2)
R = L _j / (N _i + k / 2) (2)
It becomes.

In the case of a general long-time simulation, since N _i >> k / 2,
(Equation 3)
L _j / (N _i + k / 2) ≈L _j / N _i (3)
Therefore, even if the loss rate R is calculated by either the formula (1) or the formula (2), the value is almost the same, and there is substantially no problem.

However, in order to shorten the simulation time, when obtaining the loss rate R with the smallest possible number of losses, _Ni is a small value, so in order to obtain a more accurate loss rate R, k / 2 is ignored. I can't do it. For example, in Patent Document 1, the simulation is terminated with L _j = 20 at the earliest.

Therefore, it is necessary to obtain k / 2. In order to obtain k / 2, it is necessary to perform simulation until the number of losses becomes (L _j +1) even though it is desired to shorten the simulation time. is there.

Therefore, when the number of losses reaches L _j , a method for calculating the loss rate in consideration of k / 2 described above is necessary.

The simulation apparatus according to the present invention is
In traffic simulation,
When the number of lost datagrams reaches the set L _j
The number of datagram inputs N _m when the number of lost datagrams is (L _j −1), and the difference between the number of datagram inputs N _i when the number of lost data is L _j and the N _m Apply the value k _d , that is, the value obtained from Equation (4) and the L _j to Equation (5), or the N _m , the N _i , and the L _j By applying to (5 ′), the loss rate R of the datagram is calculated.

The simulation apparatus according to the present invention is
In traffic simulation,
When the number of lost datagrams reaches the set L _j
The number N _{i of} datagrams when the number of losses is L _j , and k _a that is the “average number of inputs per loss” from the start of simulation until the number of losses reaches L _j , that is, the equation By applying the value obtained from (6) and the L _j to the equation (7), or applying the N _i and the L _j to the equation (7 ′), The loss rate R is calculated.

The simulation method according to the present invention includes:
A traffic simulation method executed by a computer,
When the number of lost datagrams reaches the set L _j
The number of datagram inputs N _m when the number of lost datagrams is (L _j −1), and the difference between the number of datagram inputs N _i when the number of lost data is L _j and the above N _m k _d , that is, the value obtained from the equation (4) and the L _j are applied to the equation (5), or the N _m , the N _i , and the L _j are expressed by the equation (5). 5 ′) applying the datagram loss rate R by applying
Is provided.

The simulation method according to the present invention includes:
A traffic simulation method executed by a computer,
When the number of lost datagrams reaches the set L _j
The number N _{i of} datagrams when the number of losses is L _j , and k _a that is the “average number of inputs per loss” from the start of simulation until the number of losses reaches L _j , that is, the equation By applying the value obtained from (6) and the L _j to the equation (7), or applying the N _i and the L _j to the equation (7 ′), Calculating a loss rate R;
Is provided.

  The program according to the present invention is a simulation program for causing a computer to execute each step provided in the simulation method according to the present invention.

  As described above, in the method of Patent Document 1, the loss rate R is calculated without considering k / 2. However, according to the present invention, the loss rate R considering k / 2 can be obtained. A more accurate loss rate can be obtained.

An example of simulation datagram input and loss generation in the present invention. It is a block diagram of the simulation apparatus in the 3rd Embodiment of this invention. It is a flowchart of the simulation in the 3rd Embodiment of this invention. FIG. 6 is a transition diagram of a two-state MMPP (Markov Modulated Poison Process) that generates simulation traffic.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In addition, this invention is not limited to embodiment shown below. These embodiments are merely examples, and the present invention can be implemented in various modifications and improvements based on the knowledge of those skilled in the art. In the present specification and drawings, the same reference numerals denote the same components.

[First Embodiment]
As a first embodiment of the present invention of the method for obtaining the loss rate R in consideration of k / 2, the number of datagram losses during simulation is (L _j −1). record the number of inputs N _m-number, number of lost datagrams using the input speed N _i number of datagrams when it becomes L _j-number,
(Equation 4)
k _d = N _i −N _m (4)
As
(Equation 5)
R = (L _j −1) / (N _m + k _d / 2) (5)
Or directly
R = 2 (L _j −1) / (N _m + N _i ) (5 ′)
Calculate, do things.

  As an example, the content of the present invention can be applied to the invention of Patent Document 1 to obtain a more accurate loss rate. The above method can be realized by software on a computer.

  Further, an electronic circuit that performs the above-described operation can be mounted on hardware to perform a simulation. The hardware is an electronic circuit in which logic elements are combined, and may be realized by mounting a logic IC (Integrated Circuit) or the like on a substrate, or an FPGA (Field-Programmable Gate Array) or PAL (Programmable Array). (Logic) or the like, or may be realized during LSI (Large Scale Integration).

Since the number of outputs = the number of inputs N _m −the number of losses (L _j −1), the number of losses (L _j −1) is very small compared to the number of inputs N _m (for example, loss rate = 10 ^{− In} the case of ⁶ , the number of losses (L _j −1): the number of inputs N _m = 1: 1,000,000), the number of outputs is nearly equal to the number of inputs. Therefore, in this embodiment, instead of the number of inputs N _m using the output number when the number of lost becomes (L j _-1) number, the number of lost instead of the input speed N _i is also a method for calculating using the number of output when it is L _j number.

[Second Embodiment]
As a second embodiment of the present invention for calculating the loss rate R in consideration of k / 2, k is the “average number of inputs per loss” from the start of the simulation to k _a the use, i.e., with the input speed N _i number of datagrams when number of lost datagrams becomes L _j-number,
(Equation 6)
k _a = N _i / L _j (6)
As
(Equation 7)
_{R = L j / (N i} + k a / 2) (7)
Or directly
R = 2L _j ² / (N _i × (2L _j +1)) (7 ′)
Calculate, do things.

  As an example, the content of the present invention can be applied to the invention of Patent Document 1 to obtain a more accurate loss rate. The above method can be realized by software on a computer.

  Further, an electronic circuit that performs the above-described operation can be mounted on hardware to perform a simulation. The hardware is an electronic circuit in which logic elements are combined, and may be realized by mounting a logic IC or the like on a substrate, may be realized by FPGA or PAL, or may be realized in an LSI. May be.

Since the number of outputs = the number of inputs N _i −the number of losses L _j , when the number of losses L _j is extremely smaller than the number of inputs N _i (for example, when the loss rate = 10 ⁻⁶ , the number of losses L _j : Number of inputs N _i = 1: 1,000,000), the number of outputs is nearly equal to the number of inputs. In this embodiment, the number of losses is L _j instead of the number of inputs N _i . There is also a method of calculating using the number of outputs.

[Third Embodiment]
In this embodiment, an application example of the first embodiment to the invention of Patent Document 1 will be described.

FIG. 2 shows an example of a simulation apparatus according to this embodiment. The simulation apparatus according to the present embodiment includes a storage unit 11, a datagram generation unit 12, a buffer unit 13, a loss detection unit 14, a time interval measurement unit 15, a time interval determination unit 16, and a χ ² test unit. 17, a loss rate calculation unit 18, and an occurrence count determination unit 19. The χ ² test unit 17 includes an actual number counting unit 21, an actual number determining unit 22, a statistic calculating unit 23, and a statistic determining unit 24.

  As an example, the simulation method according to the present embodiment will be described with reference to a flowchart shown in FIG. 3, with reference to a flowchart of FIG. 3. The simulation method according to the present embodiment includes a loss detection procedure, a time interval measurement procedure, a statistic determination procedure, and a simulation end procedure in this order. In the loss detection procedure, steps S101 to S103 are executed. In the time interval measurement procedure, steps S104 to S105 are executed. In the statistic determination procedure, steps S106 to S110 are executed. In the simulation end procedure, steps S111 to S112 are executed.

When using the present invention, the following operations are performed during simulation.
In step S100, prior to the simulation, the storage unit 11 stores the set initial value (details will be described later).
In step S101, the datagram generation unit 12 generates a datagram and generates traffic. In the case of the present embodiment, the datagram has a fixed length, which is hereinafter referred to as a unit length. The method for generating the datagram is arbitrary, but for example, it is generated using a pseudo random number by a computer. In the present embodiment, the datagram generation unit 12 acquires the MMPP parameter set in the storage unit 11, and generates a datagram according to the parameter.

  In step S <b> 102, the datagram generation unit 12 outputs the generated datagram to the buffer unit 13. As a result, the datagram is written in the buffer unit 13 and the situation is notified to the loss detection unit 14. In the case of the present embodiment, the datagram written in the buffer unit 13 is read one unit length (= 1 datagram) per unit time. The buffer unit 13 may actually write the generated datagram into the buffer, but it is only necessary to be able to manage at least the amount of free or in-use buffer and to know that a loss has occurred because there is no free space in the buffer. .

  In step S <b> 103, the loss detection unit 14 determines whether or not a loss has occurred based on the notification from the buffer unit 13, and detects the loss.

When a loss occurs, the loss detection unit 14 stores the datagram loss number L _j and the datagram input number N _i in the storage unit 11. The number of datagram inputs N _i is the number of datagrams written in the buffer unit 13. If a loss occurs, the loss detection unit 14 outputs the time at which the loss occurred to the time interval measurement unit 15 and proceeds to step S104. On the other hand, if no loss occurs, the loss detection unit 14 notifies the datagram generation unit 12 that the next datagram is to be generated, and proceeds to step S101.
In step S <b> 104, the time interval measurement unit 15 measures the time interval t at which the loss has occurred using the time acquired from the loss detection unit 14.

In step S105, the time interval determination unit 16 determines whether or not the time interval t measured by the time interval measurement unit 15 is greater than or equal to the threshold value TH. Thus, the loss is classified into a random loss (a loss that occurs randomly) and a subsequent loss (a loss that follows a loss that occurs randomly) according to the loss occurrence time interval. If the time interval t is equal to or greater than the threshold TH, that is, when a random loss, the time interval determination unit 16, and proceeds to notify the time interval t to the real power counting section 21 to the step S106, the following chi ² test Do. On the other hand, when the time interval t is less than the threshold value TH in step S105, that is, when the loss is not random, the time interval determination unit 16 notifies the datagram generation unit 12 that the next datagram is to be generated, and step S101. Return to datagram generation.

In step S106, the actual number counting unit 21 stores the notified random loss occurrence time interval and increments the random loss occurrence count m by one. Here, “m” in the present embodiment corresponds to the loss number L _j .
In step S107, the real number counting unit 21 calculates an average occurrence time interval using the random loss occurrence time intervals stored so far.
Next, the real number counting unit 21 calculates (L−1) boundary values that are the boundaries of the L intervals for the χ ² test, using the calculated average occurrence time interval.
Then, the real power counting section 21, stored to have random losses each section with a time interval of generation of i (i = 1,2, ···, L) counts the actual frequency f _i of.

In step S108, the actual number determination unit 22 determines whether all the actual numbers f _{i in} each section _i are _{equal to} or greater than f _min .
If the actual number f _i is equal to or greater than f _min in all the sections, the actual number determination unit 22 outputs the actual number f _i of each section _i to the statistic calculation unit 23 and proceeds to step S109.
On the other hand, if the actual frequency f _i is is less than five sections, the real power determination unit 22 notifies to generate the next datagram to datagram generator 12 returns to the datagram generated in step S101.

In step S109, the statistic calculator 23 calculates statistics chi ² for chi ² test from the actual frequency _{f i} of each interval i.

In step S110, the statistic determination unit 24 determines whether or not the statistic χ ² is larger than the critical value χ ₀ ² . When the statistic χ ² ≦ the critical value χ ₀ ² , the statistic determination unit 24 notifies the loss rate calculation unit 18 to that effect and proceeds to step S111. On the other hand, when the statistic χ ² > the critical value χ ₀ ² , the statistic determination unit 24 outputs the random loss occurrence count m to the occurrence count determination unit 19 and proceeds to step S113.

In step S111, the loss rate calculation unit 18 calculates the loss rate R from the total number of losses detected by the loss detection unit 14 and the number of all datagrams generated. At this time, the loss rate calculation unit 18 _obtains “k _d ” using the equation (4) and calculates the loss rate R using the equation (5), or the loss rate using the equation (5 ′). The rate R is calculated, or “k _a ” is calculated using the equation (6) and the loss rate R is calculated using the equation (7), or the loss rate is calculated using the equation (7 ′). Calculate R.
In step S112, the loss rate calculation unit 18 outputs a loss rate as a calculation result.

In step S113, the occurrence count determination unit 19 determines whether the random loss occurrence count m that has occurred so far is greater than the maximum occurrence count M or not.
In the case of m ≦ M, the occurrence count determination unit 19 notifies the datagram generation unit 12 that the next datagram is to be generated, proceeds to step S101, and returns to datagram generation.
When m> M, the occurrence count determination unit 19 executes Step S114 and outputs a message to that effect.

  In the case of the MMPP / D / 1 / K simulation example, the initial value set in step S100, that is, what value the initial value is set for each item in the initial setting portion of the flowchart of FIG. Is described below.

(A) MMPP Parameter Value MMPP is a Markov Modulated Poisson Process that is one of traffic generation forms, and is a model in which traffic is generated at a plurality of different densities that are switched randomly as shown in FIG. Although the MMPP can have a plurality of different density traffic generation states, the simplest two states in FIG. 4 will be described as an example.
Λ ₀ and λ ₁ are the average arrival rates of datagrams generated in each of the states 0 and 1, and are usually 0 or more and 1 or less.
Γ ₀₁ and γ ₁₀ are the transition probability from MMPP state 0 to state 1 and the transition probability from state 1 to state 0, respectively, and are values from 0 to 1. Similarly, 1-γ ₀₁ and 1-γ ₁₀ are probabilities that state 0 and state 1 will continue, respectively.

(B) Buffer size (K-1)
The size of the buffer unit 13 into which the generated MMPP traffic flows.

(C) Threshold value TH
Every time a loss occurs, the generation time interval t with the previous loss is calculated. The threshold value TH is a reference value of the occurrence time interval t set in advance, and is classified into a random loss and a subsequent loss by comparing with this.

(D) Critical value χ ₀ ²
In order to set the critical value χ ₀ ² , first, it is necessary to set the significance level α. The significance level α is that the probability that the statistic χ ² obtained from the sample exceeds the critical value χ ₀ ² (first type error) is α when the sample obtained so far follows an exponential distribution. Meaning. In this embodiment, a random loss occurrence time interval is used for the sample.
The value of α is usually 5% or 1%. In this embodiment, as will be described later, the degree of freedom d = 1, and the critical value χ ₀ ² is obtained from the statistical value table of the distribution of the statistic χ ² .
When α = 5%, χ ₀ ² = 3.841
When α = 1%, χ ₀ ² = 6.635
It is.

(E) Maximum value M of random loss
Since the statistic χ ² tends to gradually increase as the number of losses increases, the statistic χ ² ≦ critical value χ ₀ ² when the random loss number m does not end up to about several hundred. There is a possibility that it will not be. In order to cope with this case, the maximum value M of the random loss number m for ending the simulation is set (the process ends when the random loss number m exceeds the maximum value M (steps S113 to S114)).
L and f _min will be described later.

The random loss and the subsequent loss described in the operation (step S105) performed during the simulation will be described.
If a loss occurs rarely due to the input of MMPP traffic to the buffer unit 13 in a state where the space is large, it is considered that the loss occurs randomly (random loss). However, when a loss occurs due to the input of MMPP traffic to the buffer unit 13 once the loss has occurred, there is almost no vacancy in the buffer unit 13, and thus the loss occurs at zero or very close intervals. Increase (subsequent loss). Therefore, it is considered that the loss generated in the above-described traffic model includes two types of loss, a random loss and a subsequent loss.

  The occurrence time interval of random loss is considered to follow an exponential distribution, and usually occurs when the output time of D of MMPP / D / 1 / K, that is, the read time of the datagram from the buffer unit 13 is 1 [unit time]. The time interval is several hundred thousand [unit time], and the existence probability of the occurrence time interval is thin and widely distributed. On the other hand, since the subsequent loss occurrence time interval is zero or a value close to zero, for example, 100 [unit time] or less, it can be separated by the threshold value TH, although not strictly.

An outline of the χ ² test of the operations (steps S106 to S110) performed during the simulation will be described.
As a method of numerically evaluating and judging whether or not the coincidence degree (fitness) of whether or not the occurrence time interval of random loss occurring in the simulation (hereinafter referred to as random loss occurrence time interval) follows the exponential distribution is χ ² test is is there. The χ ² test is a method often used to test how well the sample distribution matches the population (theoretical) distribution. It is known that the occurrence interval of randomly occurring events has an exponential distribution. In such a case, an exponential distribution is assumed as a population distribution. In addition to this, various distributions such as binomial distribution and normal distribution are assumed depending on the target event.

In the χ ² test, first, the difference in distribution between the sample and the population is obtained as a statistic χ ² . Here, the sample is a random loss occurrence time interval w _j (j = 1, 2,..., N), and the set is W _n . The population distribution is an exponential distribution. The statistic χ ² generally tends to decrease as the sample distribution approaches the population distribution.

Next, the statistic χ ² is compared with the critical value χ ₀ ² obtained in advance, and if the statistic χ ² is less than or equal to the critical value χ ₀ ² , the sample has a population distribution with a significance level α%. Is done. Specifically, every time the random loss occurrence time interval w _j is obtained by simulation, a χ ² test is performed on the set W _n , and when it is determined that the distribution is exponential, The loss rate is calculated from the number of inputs to the buffer unit 13, and the simulation is terminated. The critical value χ ₀ ² can be obtained by complex calculation using the degree of freedom d and the significance level α%, but frequently used values are shown as a statistical numerical table in Non-Patent Document 3, etc. .

A specific method of implementing the χ ² test of the operation (step S107) performed during the simulation (four sections, three boundary values serving as boundaries), and the like will be described.

-Population distribution parameters In this embodiment, the population distribution is an exponential distribution, but the parameters (mean and variance) are unknown.
Therefore, the average occurrence time interval 1 / τ of random loss is calculated from the sample, and the population distribution is estimated using this. In the exponential distribution, when the average value is 1 / τ, the variance is 1 / τ ² . Therefore, an average value is calculated every time data of a new generation time interval due to the occurrence of a new “random loss” is obtained, and a variance is calculated from the average value to be a parameter of the population distribution.

・ Number of intervals (category) for χ ² test
In the χ ² test, each occurrence time interval w _j of random loss obtained by simulation is classified into one of a plurality of intervals according to the value, and the actual number is counted. For this reason, it is necessary to determine the number of sections and the range of each section.
The number of intervals seems to be several to several tens in actual tests, but it is required to be as small as possible in order to shorten the calculation time and the calculation load.
In the χ ² test, when d is the degree of freedom, L is the number of sections, and s is the number of parameters estimated from the maximum likelihood from the sample,
(Equation 8)
d = L-1-s ≧ 1 (8)
It is. As described above, since the parameters of the population distribution are the mean and variance estimated from the sample, s = 2. Therefore, from Equation (8) (Equation 9)
L ≧ 4 (9)
It becomes. Therefore, the section number L = 4.

・ Minimum actual number of intervals f _min
In the χ ² test, the number of samples in each section can be set freely. However, since the number of samples for which the χ ² test is established is generally 5 or more, in this embodiment, the minimum actual number f _min of each section is set to 5. As a result, the minimum number of losses required for this embodiment is 20.

・ Range of each section When the number of samples is small, the sample values are often biased. However, since it is shown that when the number of samples increases, it matches the exponential distribution, so the range of each section is determined assuming the exponential distribution.
Since the number of sections for performing the χ ² test is 4, and the minimum actual number f _min of each section is equal to 5, the range of the section is equal to each cumulative probability distribution of the four sections in the exponential distribution, that is, 0 A range of .25 is considered reasonable.

The four sections are called the section i (i = 1,..., 4) from the left side of the exponential distribution (w ≧ 0), and the boundary value between the section i and the section i + 1 is a boundary value y _i (where i = 1). , 2, 3). Each boundary value y _i is obtained from equation (8):
(Equation 10)
0.25 × i = 1−exp (−τy _i ) (10)
Solve this,
(Equation 11)
y _i = −log _e (1−0.25 × i) / τ (11)
It becomes.

· Χ ² test statistic χ ²
In order to perform the χ ² test, the random loss occurrence time intervals w _j of the samples obtained so far are classified into four sections using the three boundary values obtained by the equation (11). If the actual number f _i (= number of loss) in any section is less than 5, which is the minimum actual number f _min , the simulation proceeds again until the next random loss occurs (step S108).

ただし、Ｆ_ｉは理論度数（実数）で、本実施形態では、

で表される。 When the actual number f _{i of} all the sections is 5 or more, which is the minimum actual number f _min , the statistic χ ² of the following equation (12) is calculated (step S109).

However, F _i is a theoretical frequency (real number), and in this embodiment,

It is represented by

The statistical value χ ² obtained in this way is used for comparison with the critical value χ ₀ ² of the χ ² test (step S110).

If the statistic χ ² > the critical value χ ₀ ² , it is determined that the sample value obtained so far does not statistically follow the exponential distribution at the significance level α. Continue the simulation until loss occurs.

If the statistic χ ² ≦ the critical value χ ₀ ² , it is determined that the random loss follows the exponential distribution, and the process proceeds to the loss rate calculation (step S111). In step S111, the loss rate is calculated by dividing the total number of losses so far by the total number of inputs, and the simulation is terminated.

  It is also possible to perform simulation by mounting an electronic circuit that performs the above-described operation on hardware. As hardware, it is an electronic circuit that combines logic elements, which can be realized by mounting a logic IC etc. on a substrate, FPGA, PAL, etc., or in a custom LSI May be. The apparatus of the present invention can also be realized by a computer and a program, and can be recorded on a recording medium or provided through a network.

  The simulation method, the simulation apparatus, and the simulation program according to the present invention have been described above for the case of performing the MMPP / D / 1 / K simulation, but the present invention is not limited to the MMPP / D / 1 / K simulation. For example, the datagram can use any transmission / reception data used in traffic simulation other than MMPP / D / 1 / K, such as M / M / 1 / K and M / G / 1 / K. When K is a finite value, it is a finite buffer, and when K is an infinite value, it means an infinite buffer (in the case of an infinite buffer, K may be omitted). The present invention includes IP packets, Ethernet (registered trademark) frames, ATM cells, X. It can be applied to the simulation of any communication system using any datagram such as 25 packets.

  Multipoint distribution (multicast using UDP) of video and audio for live internet distribution such as concerts on the Internet, training, instruction, lectures, etc. on private networks such as in-house IP network to each employee's receiving device (PC etc.) In the situation, as a network design, it is necessary to predict loss rates in various traffic forms and traffic volumes before distribution.

  In such a case, the simulation is still one of the effective methods, and in the method of ending the simulation using the statistical method described in the embodiment of the present invention, when the simulation is judged to be statistically effective. Since the loss rate can be calculated, the simulation can be completed in a short time. In addition, the obtained loss rate can be given statistical validity.

  The present invention provides a calculation method capable of obtaining a more accurate loss rate when the simulation is completed in a short time.

11: storage unit 12: datagram generation unit 13: buffer unit 14: loss detection unit 15: time interval measurement unit 16: time interval determination unit 17: χ ² test unit 18: loss rate calculation unit 19: number of occurrences determination unit 21 : Actual number counting unit 22: actual number determining unit 23: statistic calculating unit 24: statistic determining unit

Claims (5)

In traffic simulation,
When the number of lost datagrams reaches the set L _j
The number of datagram inputs N _m when the number of lost datagrams is (L _j −1), and the difference between the number of datagram inputs N _i when the number of lost data is L _j and the N _m By applying the value k _d and the L _j to the equation (C11), or applying the N _m , the N _i , and the L _j to the equation (C12), Calculating the loss rate R;
Simulation device.
(Number C11)
R = (L _j −1) / (N _m + k _d / 2) (C11)
(Number C12)
R = 2 (L _j −1) / (N _m + N _i ) (C12) In traffic simulation,
When the number of lost datagrams reaches the set L _j
Input speed N _i datagram when the number of losses of L _j-number, and the number of losses from the beginning of the simulation is L of _j until number "average input per 1 Loss" k _a, and the The loss rate R of the datagram is calculated by applying L _j to the equation (C21) or applying the N _i and the L _j to the equation (C22).
Simulation device.
(Number C21)
_{R = L j / (N i} + k a / 2) (C21)
(Number C22)
R = 2L _j ² / (N _i × (2L _j +1)) (C22) A traffic simulation method executed by a computer,
When the number of lost datagrams reaches the set L _j
The number of datagram inputs N _m when the number of lost datagrams is (L _j −1), and the difference between the number of datagram inputs N _i when the number of lost data is L _j and the above N _m Applying k _d and L _j to equation (C31), or applying N _m , N _i , and L _j to equation (C32), loss of datagram Calculating the rate R;
Simulation method.
(Number C31)
R = (L _j −1) / (N _m + k _d / 2) (C31)
(Number C32)
R = 2 (L _j −1) / (N _m + N _i ) (C32) A traffic simulation method executed by a computer,
When the number of lost datagrams reaches the set L _j
Input speed N _i datagram when the number of losses of L _j-number, and the number of losses from the beginning of the simulation is L of _j until number "average input per 1 Loss" k _a, and the The loss rate R of the datagram is calculated by applying L _j to the equation (C41) or applying the N _i and the L _j to the equation (C42).
Simulation method.
(Number C41)
_{R = L j / (N i} + k a / 2) (C41)
(Number C42)
R = 2L _j ² / (N _i × (2L _j +1)) (C42)   The simulation program for making a computer perform each operation | movement of Claim 3 or 4.

Images