JPH10336244A - Method for deciding parameter for ip quality assurance service by application and recording medium recording parameter decision program for ip quality assurance service - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a parameter decision method for a IP quality assurance service by an application where waste of a resource is omitted and quality deterioration due to a frequency band estimate error is avoided. SOLUTION: An upper limit Rbound calculation section receives a request delay Dreq and a traffic parameter TSpec from an application main body 3, a network state through which a data flow passes from a network 4, calculates a band upper limit Rbound to warrant the delay Dreq as a packet transfer delay time W and gives the limit to an R calculation section 7. A probability Rprob calculation section 6 calculates a probability approximation value Rprob based on the Dreq received from the application main body 3, a probability α to permit its excess, an average rate (r) of the TSpec, and a statistic property of the traffic stored in the section 6 according to the probability Pr(W>Dreq )=α of the W in excess of the Dreq and gives the result to the R calculation section 7. The R calculation section 7 estimates a warrant band R recording to equation R=(1-β)Rbound +βRpprob (0<=β<=1) and returns the result to the application main body 3. The application main body 3 requests flow setting to the network 4 by using the calculated R as a quality parameter and when the data flow is set, a transmission terminal 5 sends data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an IP (Int) network in which a router having a scheduling mechanism capable of guaranteeing a bandwidth for each application flow is connected in a network.
When a real-time application that operates between terminals connected to each other via an Internet Protocol (IP) network uses the IP quality assurance service by requesting the guarantee of end-to-end packet transfer delay time, the service is provided. The present invention relates to a method for determining a parameter of an IP quality assurance service by an application which determines a value of a guaranteed bandwidth which is a quality parameter to be defined.

[0002] This parameter determination method is a form as one function of a GUI (Graphical User Interface) type browser operated by a user himself, and an operating system (OS).
This is provided in the form of one function of communication processing of em), in the form of being embedded in an application program, and in the form of one function of a library function of a programming language.

[0003]

2. Description of the Related Art Data that is continuously sampled from an input device (camera, microphone, etc.) of a transmitting terminal, such as a telephone, broadcasting, or a video conference, is transferred to a receiving terminal via a network to reproduce the original video and audio. Real-time applications that need to guarantee end-to-end packet transfer delay time, such as reading data periodically from a storage medium on the sending terminal, transferring the data to the receiving terminal via a network, and playing back video and audio The use of Internet on the Internet is being considered.

[0004] However, the current Internet provides only the best-effort service that does not guarantee the quality. Therefore, if there is a congestion router on a packet transfer path, loss or excessive delay is caused, and reproduction information has a predetermined quality. May not be satisfied, and a quality comparable to current broadcast media with a large amount of information cannot be obtained. Internet Enginee
The Ring Task Force (IETF) adopts Weighted Fair Queuing (WFQ) as packet scheduling to control the packet transfer order in the router, etc., and guarantees the bandwidth for each application flow, so that the end-to-end packet transfer delay time We are considering providing an IP quality assurance service that guarantees the following.

[Prior Art 1] <Band (R
Method for calculating the _bound )> “Internet Draft” that defines the IP quality assurance service (S. Shenker, CP
artridge and R. Guerin, “Specification of Guarante
ed Quality of Service, "Internet Draft, Feb., 19
97. ) Gives the upper limit value _Dmax of the end-to-end packet transfer delay time based on the traffic parameters (TSpec: average rate r, peak rate p, burst size b, maximum packet size M) that give the upper limit of the outgoing traffic. From the formula, there has been proposed a method of estimating the guaranteed bandwidth R in the transit router so as to satisfy the end-to-end required delay D _req . The application specifies the guaranteed bandwidth R as a quality parameter (RSpec).

According to this, the upper limit value of the delay time due to packet scheduling in the passing router (i) occurs depending on the term estimated by fluid approximation and the guaranteed bandwidth (R) in consideration of the packet size. The term (C _i / R) can be divided into a term (D _i ) that does not depend on the guaranteed bandwidth. For example, if the scheduling of the router is WFQ, assuming that μ and MTU are the bandwidth of the transmission link from the router i and the maximum transfer unit, respectively, C _i =
M, D _i = MTU / μ.

In the case of passing through a plurality of routers, it is possible to estimate the upper limit value of the packet transfer delay time between the end-end terminals by adding the terms C _i and D _i of all the routers passing the flow. is there. Here, C _i and D _i added for all routers passing through the flow are represented as C _tot and D _tot , respectively, and the pair of them is called Adspec.

[0008] If this Adspec is notified in advance to the application on the terminal that sets the flow, the upper limit of the packet transfer delay time between the end-end terminals is calculated in consideration of the TSpec that gives the upper limit of the outgoing traffic.

(Equation 1) Can be estimated from the relational expression. Conversely, _assuming that the packet transfer delay time between the end and end terminals required by the application is a required delay D _req , a band R required to absolutely guarantee this is obtained by setting D _max = D _req in the above equation. Can be In order to distinguish the value from the value estimated by the method described later, the guaranteed band R estimated by this equation is described as R _bound . This R _bound is a band that guarantees the required delay time even when the worst traffic is assumed under the constraints of TSpec, and is the upper limit of the guaranteed band.

FIG. 1 shows an embodiment for realizing the parameter estimation method proposed in the Internet draft. That is, the upper limit R _bound calculation unit 2 provided in the receiving terminal 1 receives the request delay D _req and TSpec from the application main body 3 and receives C _tot and D _tot from the network 4. Then, based on these, R = R _bound is calculated by the above-described calculation formula, and the result is returned to the application main body 3. The application requests the network 4 to set the flow, using the guaranteed bandwidth R as RSpec. When the flow is set, data is transmitted from the transmitting terminal 5.

[Prior art 2] <Method of calculating an approximate value (R _prob ) of a band that guarantees a packet transfer delay with a probability α> In a network, each router through which an application flow passes has a constant If the bandwidth is guaranteed and completely isolated from other flows, the network portion can be approximated by a single server with a guaranteed bandwidth. Therefore, assuming some input traffic model, the probability Pr (W> D _req ) of the end-to-end packet transfer delay time W exceeding the required delay D _req is strictly or approximately solved using the guaranteed bandwidth R. Can be. Utilizing this, the probability Pr that there is no problem if it is smaller than that for the user or application
If the value of (W> D _req ) is α, the relational expression Pr (W> D _req )
_req ) = α, an approximate value R _prob = R of the required band can be estimated. However, the assumed traffic model cannot represent actual traffic, or depending on the model, only an approximate solution can be obtained. Therefore, the approximate value R _prob estimated here includes some error.

Here, a two-state Markov modulated Poisson proc commonly used as a traffic model
When ess (MMPP) is adopted, the model parameters are the duration of each high-load / low-load state and the load of each state (for example, H.Heffes and DMLucantoni, “A Markov
Modulated Characterization of Packetized Voiceand
Data Traffic and Related Statistical Multiplexer
Performance, "IEEEJournal on Selected Area in Com
munications, Vol. 4, No. 6, Sept., 1986.). These parameters cannot be easily converted from statistics that can be measured directly by the application. Further, when solving this model, there is a problem in that Laplace transform and iterative calculation are required, and it takes time to calculate a numerical solution.

FIG. 2 shows an embodiment for realizing the parameter estimation method according to the present method. Note that, in the figure, the same components as those in FIG. 1 are denoted by the same reference numerals. Probabilistic value R _prob calculator 6 provided in receiving terminal 1
Receives the request delay D _req , the probability α of exceeding the required delay D, and the average rate r of TSpec from the application main body 3. Then, based on these and the statistical properties of the traffic held by itself, R = R
_{The prob} is calculated, and the result is returned to the application main body 3.
The application requests the network 4 to set the flow, using the guaranteed bandwidth R as RSpec. When the flow is set, data is transmitted from the transmitting terminal 5.

[0013]

[Problems to be solved by the invention]

[Problem 1] R = R _bound estimated from TSpec is excessive.
Real application, it will not produce a worst traffic defined by the declared traffic parameters (TSpec), that is, the actual traffic is less than TSpec, assuming a worst traffic TSpec from request delay D _req When the calculated upper limit R _bound is adopted as the guaranteed band R, it is assumed that the delay time becomes smaller than necessary than the required delay D _req . In other words, there is a problem that the guaranteed bandwidth R is largely estimated. This is not only a waste of network resources, but also wasteful in terms of cost if, for example, the network fee structure increases in proportion to the bandwidth.

[Problem 2] Low accuracy with R = R _prob estimated from traffic statistics. If the approximate value R _prob calculated by the approximation formula is adopted as the guaranteed band R so that the required delay D _req is satisfied with the probability of α using the traffic statistics, the approximation between the actual traffic and the adopted traffic model is obtained. And the degree of approximation of the guaranteed bandwidth calculation formula, the deviation of the obtained delay time from the required delay D _req increases.

[Problem 3] When negotiating (re-declaring) flow setting parameters, there is no R selection guideline.
If the flow is not accepted, it is necessary to negotiate RSpec parameters with the network, but usually the RSpec parameters adopted by the application are fixed once decided, and even if they can be changed, the range and change guidelines that can be changed Since there is nothing, the user has to change to an appropriate value by trial and error.

[Problem 4] The first flow setting parameter has an estimation error. There is an error in estimating the RSpec parameters at the time of setting the flow, and there is a possibility that the usage situation of the user may change, so that the parameters initially set are not always optimal. In addition, there is a demand to modify this parameter to the optimal value as much as possible during the operation of the application.However, even if the parameter can be changed, there is no changeable range or change guideline. You have to change it.

[Problem 5] The statistical properties of traffic vary depending on how the user uses it. The statistical properties of the traffic for estimating the approximate value R _prob vary depending on the environment in which the application is used and the usage of the user.

[Problem 6] The parameters of the MMPP are not measurable statistics. When a generally used MMPP is adopted as a traffic model, it is necessary to convert the measurable statistics into model parameters (duration of each high load / low load state and load of each state). Further, when solving this, there is a problem that it takes time to calculate the numerical solution because Laplace transform and iterative calculation are necessary.

[Problem 7] It is impossible to cope with a case where the burst has no characteristic length. In LAN (Local Area Network) traffic and the like, bursts have no characteristic length (or burst length exists over a very wide range),
It is known to have a statistical property that when the time unit is changed, the variance increases exponentially with increasing time unit. On the other hand, in an MMPP or the like having a finite burst length, it is known that the dispersion converges to a constant value if the time unit increases to some extent. Therefore, there is a problem that a traffic model such as MMPP cannot express the above-described statistical characteristics. Therefore, a problem to be solved in the case of traffic having no characteristic length in bursts is an issue.

The present invention has been made in view of the above points, and an object of the present invention is to provide a method for determining parameters of an IP quality assurance service by an application described below. 1. Eliminating excessive or under-allocation eliminates waste of resources and eliminates quality degradation due to band estimation errors. 2. Formulate the change standard of the guaranteed bandwidth when the application negotiates the flow setting parameter.

3. To be able to modify the flow setting parameters during the operation of the application to the optimal values as possible. 4. The statistical traffic characteristic parameter of the application can be changed according to the actual statistical characteristic that fluctuates depending on the operating environment of the application and the usage of the user, and the estimation error due to the usage condition is eliminated. 5. The parameters of the IP quality assurance service are determined using a traffic model that includes a measurable statistic as a model parameter and can easily calculate a numerical solution. 6. Provided is a method for determining parameters of an IP quality assurance service capable of coping with a traffic input having no characteristic length of a burst.

[0022]

[Means for Solving the Problems] To solve the above problems,
In order to achieve this, the invention according to claim 1 is applied to an application flow
Scheduling mechanism that can guarantee bandwidth every time
Network to which a router having a network is connected in a network
Work between interconnected devices via a connection
The application is responsible for defining traffic limits.
Declaring Hick Parameters, End-to-End Packet
Request delay D as transfer delay time_req Claiming the guarantee of I
When using the P quality assurance service, the IP quality assurance service
Of guaranteed bandwidth R, which is a quality parameter that defines the service
IP quality assurance service by application that determines
In the method for determining a traffic parameter,
Assuming traffic limited by the meter,
Request delay D_req End-to-end packet transfer delay
Upper limit value R of the band necessary to guarantee the interval_bound And the said
End-to-end packet transfer delay time is the required delay
D_req User or the application
So that it is less than the value α that does not hinder the application.
Application's own statistical traffic characteristics and
And end-to-end expressed by the traffic parameters.
Calculated from the approximate expression of the packet transfer delay time distribution
Band approximation R_probAnd the relational expression R = (1-
β) R _bound+ ΒR_prob Guaranteed by (0 ≦ β ≦ 1)
It is characterized in that the band R is estimated.

Further, according to the second aspect of the present invention, a router having a scheduling mechanism capable of guaranteeing a bandwidth for each application flow is connected to a network.
An application operating between terminals connected to each other via a P network declares a traffic parameter defining an upper limit of traffic and requests a guarantee of a required delay D _req as an end-to-end packet transfer delay time. When using the IP quality assurance service,
In an IP quality assurance service parameter determination method by an application that determines a value of a guaranteed bandwidth R, which is a quality parameter that defines a P quality assurance service, when an initial setting value of the guaranteed bandwidth R is not accepted by the IP network. , Assuming traffic limited by the traffic parameters, the required delay D
and the upper limit value R _bound bandwidth needed to guarantee end packet transfer delay of the end - - End _req probability that end packet transfer delay time exceeds the required delay D _req is, to the user or the application An approximate value R of a band calculated from an approximate expression of an end-to-end packet transfer delay time distribution expressed by the statistical traffic characteristics and the traffic parameters held by the application itself so as to be less than a value α which does not cause a problem. Based on _prob , the guaranteed bandwidth R of the quality parameter required for the IP network is R _prob ≦ R ≦ R
Within the scope of the _{bound (R bound -R pro b)} / n (n
> 0), and the quotation is reduced, and negotiation is performed with the IP network.

According to a third aspect of the present invention, a guaranteed bandwidth R at the start of an application is determined according to the method of estimating a guaranteed bandwidth R according to the first aspect, and when the guaranteed bandwidth R is not accepted by the IP network. According to a second aspect of the present invention, the guaranteed bandwidth R is sequentially changed and reset according to the negotiation method of the guaranteed bandwidth R.

The invention according to claim 4 is an application for
Scheduling that guarantees bandwidth for each application flow
I connected to a router with a
Terminals connected to each other via the P network
Applications that run between
Declare the specified traffic parameters and end-
Required delay D as the packet transfer delay time_req Guarantee
Request to use the IP quality assurance service,
P is a quality parameter that defines the quality assurance service.
IP products by the application that determines the value of the certification band R
In the method for determining the parameters of the quality assurance service,
Assuming traffic limited by traffic parameters
If the request delay D_req End-to-end packet
The upper limit R of the bandwidth required to guarantee the transfer delay time
_bound And the end-to-end packet transfer delay time
Is the request delay D_req Probability that exceeds
It will be less than the value α that does not hinder the application
As such, the statistical tokens held by the application itself
Traffic characteristics and traffic parameters
Approximate expression of end-to-end packet transfer delay time distribution
Approximate value R of the band calculated from_probAnd based on said
While running the application, end-to-end
Packet transfer delay time D^*And the transfer delay time D^*But
The request delay D _reqProbability Pr (D^*> D_req)But
Pr (D^*> D_req) <The smallest value that satisfies α
Thus, the quality parameter required for the IP network
The guaranteed bandwidth R of_prob≤R≤R_bound Increase or decrease in the range
It is characterized by being adjusted.

Further, the invention described in claim 5 provides the invention according to claims 1 to
3. The guaranteed bandwidth R at the start of the application is determined according to the method of estimating the guaranteed bandwidth R according to any one of the items 3 and 3, while the application is operated.
The method is characterized in that the guaranteed bandwidth R is sequentially changed in accordance with the method of adjusting the guaranteed bandwidth R described above. According to a sixth aspect of the present invention, in the invention according to any one of the first to fifth aspects, a traffic is measured at the time of operation of the application, and a parameter X _k expressing the statistical traffic characteristic is measured at the time of measurement. Based on the set value X _0k and the statistical value X ^* _k calculated from the measured traffic data,
The application is characterized in that the application has a procedure for updating and holding X _k = (1−γ _k ) X _0k + γ _k X ^* _k (0 ≦ γ _k ≦ 1).

[0027] Further, the invention according to claim 7 provides the invention according to claims 1 to
6. The invention according to any one of items 6, wherein an asymptotic dispersion constant c ² and a normalized third moment m ₃ are adopted as the statistical traffic characteristics, and are statistically defined by these and an average rate r of the traffic parameters. The above-mentioned approximate value R _prob is calculated using an approximate expression of an end-to-end packet transfer delay time distribution obtained by analyzing a GI / D / 1 queuing system of general distributed traffic input.

[0028] Further, the invention according to claim 8 is based on claims 1 to
6. The variance-average ratio a and the self-similarity parameter H as the statistical traffic characteristics.
And an approximate expression of the end-to-end packet transfer delay time distribution obtained by analyzing the queuing system of the FBM traffic input statistically defined by these and the average rate r of the traffic parameters. The method is characterized in that the approximate value R _prob is calculated.

Further, according to the ninth aspect of the present invention, a router having a scheduling mechanism capable of guaranteeing a bandwidth for each application flow is connected to a network.
An application operating between terminals connected to each other via a P network declares a traffic parameter defining an upper limit of traffic and requests a guarantee of a required delay D _req as an end-to-end packet transfer delay time. When using the IP quality assurance service,
In a parameter determination program of an IP quality assurance service that determines a value of a guaranteed bandwidth R, which is a quality parameter that defines a P quality assurance service, when traffic limited by the traffic parameter is assumed, the end of the request delay D _req An upper limit value calculation process for calculating an upper limit value R _bound of a band required to guarantee an end packet transfer delay time, and the end-to-end packet transfer when traffic limited by the traffic parameter is assumed. The end-to-end packet represented by the statistical traffic characteristics and the traffic parameters held by the application such that the probability that the delay time exceeds the required delay D _req is less than a value α that does not hinder the user or the application. From the approximate expression of the transfer delay time distribution, Value calculation processing for calculating the approximate value R _prob of the relation R = (1−β) R _bound + βR
It is characterized by causing a computer in the terminal to execute estimation processing for estimating the guaranteed bandwidth R by _prob (0 ≦ β ≦ 1).

[0030] Further, according to the tenth aspect of the present invention, a terminal having terminals connected to each other via an IP network to which a router having a scheduling mechanism capable of guaranteeing a bandwidth for each application flow is connected in a network. When the application operating under the IP service declares a traffic parameter that defines the upper limit of traffic and requests the guarantee of the required delay D _req as the end-to-end packet transfer delay time to use the IP quality assurance service, In the parameter determination program of the IP quality assurance service for determining the value of the guaranteed bandwidth R, which is a quality parameter defining the guaranteed service, a request reception for detecting that the first set value of the guaranteed bandwidth R has not been accepted by the IP network. Judgment processing, limited by the traffic parameters Assuming traffic, an upper limit value calculation process for calculating an upper limit value R _bound of a band required to guarantee an end-to-end packet transfer delay time of the request delay D _req is limited by the traffic parameter. Assuming traffic, the statistical information held by the application is such that the probability that the end-to-end packet transfer delay time exceeds the required delay D _req is less than a value α that does not hinder the user or the application. Approximate value calculation processing for calculating an approximate value R _{prob of a} band from an approximate expression of an end-to-end packet transfer delay time distribution expressed by traffic characteristics and the traffic parameters; and a guaranteed band R of a quality parameter required for the IP network. in the range of _{R prob ≦ R ≦ R bound (} R bound -R prob) / n (n> 0 And estimated process is reduced by the step size to negotiate with the IP network of the is characterized by causing a computer to execute within the terminal.

The invention according to claim 11 is the same as the claim 9
The guaranteed bandwidth R at the time of starting an application is determined according to the estimation processing described in claim 10, and when it is detected in the request acceptance determination processing in claim 10 that the guaranteed bandwidth R is not accepted by the IP network, the request bandwidth is determined. The present invention is characterized in that a computer in the terminal executes a process of sequentially changing the guaranteed bandwidth R and resetting it in accordance with the estimation process.

The invention according to claim 12 is an application
Scheduling that guarantees bandwidth for each application flow
Router with networking mechanism is connected to the network
Ends connected to each other via an IP network
The application that runs between end-to-end traffic limits
Declare the traffic parameters that specify
Request delay D as the packet transfer delay time of_req No
When using the IP quality assurance service by requesting a certificate,
A quality parameter that defines the IP quality assurance service
Parameter of the IP quality assurance service that determines the value of the guaranteed bandwidth R
In the meter determination program, the traffic parameter
Assuming traffic limited by data,
Delay D_req End-to-end packet transfer delay time
Upper limit value R of the band required to guarantee_bound Calculate
Upper limit calculation process, limited by the traffic parameters
End-to-end traffic
Packet transfer delay time is the required delay D_req Cross over
Probability hinders the user or the application
Is less than the value α.
The statistical traffic characteristics to be maintained and the traffic
End-to-end packet transfer delay expressed in parameters
From the approximate expression of the time distribution, the approximate value R of the band_probCalculate near
Similarity calculation process and the operation of the application
End-to-end packet transfer delay time D^*Measure
And the transfer delay time D^*Is the request delay D _reqSure to cross
Rate Pr (D^*> D_req) Is Pr (D^*> D_req) <Α
The IP network so as to have the smallest value.
The guaranteed bandwidth R of the quality parameter required for_prob≤R
≤R_bound Adjustment process to adjust by increasing or decreasing in the range of
Characterized in that it is executed by a computer in the terminal.
ing.

The invention according to claim 13 is the same as the claim 9.
13. The terminal according to claim 11, wherein the terminal determines a guaranteed bandwidth R at the time of starting the application according to the estimation processing according to any one of claims 11 to 11, and sequentially changes the guaranteed bandwidth R according to the adjustment processing according to claim 12, while operating the application. It is characterized by being executed by a computer in the inside. According to a fourteenth aspect of the present invention, in the invention according to any one of the ninth to thirteenth aspects, a traffic measurement process for measuring traffic at the time of operation of the application, and a parameter X representing the statistical traffic characteristic.
The _k, statistics X ^* _k calculated from the data of the set value X _0k a measured traffic at the measurement point based on, X _k
= (1−γ _k ) X _0k + γ _k X ^* _k (0 ≦ γ _k ≦ 1), and the update processing to be held by the application is executed by the computer in the terminal.

The invention according to claim 15 provides the invention according to claim 9
15. The method according to any one of items 14 to 14, wherein
Outflow processing, asymptotic variance determination as the statistical traffic characteristics
Number c ^Two And normalized third moment m_Three Adopt this
And the average rate r of the traffic parameters
Therefore, G of general distribution traffic input statistically defined
End obtained by analyzing I / D / 1 queuing system
-Using the approximate expression of the end packet transfer delay time distribution
The approximate value R_probIs calculated. Ma
The invention according to claim 16 is any one of claims 9 to 14.
In the invention described in the paragraph, the approximate value calculation process is
Variance-average ratio a and self-similarity as statistical traffic characteristics
The parameter H is adopted, and these and the traffic parameters are used.
Statistically defined by the average rate r of data
Analysis of the queuing system for FBM traffic input
Of End-to-End Packet Transfer Delay Time Distribution
Approximate value R_probIs characterized by calculating
doing.

[0035]

Embodiments of the present invention will be described below with reference to the drawings. Here, the present invention can be realized by software running on a terminal, and the software runs on a program-controlled computer provided in the terminal. That is, a computer program for realizing the present invention (a program for determining parameters of the IP quality assurance service) is provided by being stored in various recording media. Ware will be controlled.

As a recording medium, for example, a magnetic recording medium such as a floppy disk, a hard disk, a magnetic tape, a magnetic card, a magnetic drum, a semiconductor recording medium such as a memory chip, an IC (integrated circuit) card, and a CD-ROM (compact disk, Optical recording media such as read-only memory, optical disk, optical card, MO (magneto-optical)
Various types of media such as a disk are conceivable, and other various types of information storage media may be used.

Hereinafter, a processing procedure performed by a computer in the terminal based on a command of a program operating on the terminal will be described. [Method 1] upper limit R to a guaranteed bandwidth R of RSpec estimated from TSpec <upper limit value R _bound and stochastic approximation R guaranteed by the combination of _prob band R determination method>
_{If bound} , the delay time obtained between the end-end terminals becomes smaller than necessary. In other words, there is a problem that _R = R boun _d becomes excessive allocation (Problem 1). Further, there is a problem that the accuracy is low when R = R _prob which is stochastically estimated from the statistics of application traffic using approximation (problem 2). In order to solve these problems, a method of estimating the guaranteed bandwidth R using the upper limit value R _bound and the stochastic approximation value R _prob is proposed.

The state of the network Adspec (C _tot , D _tot ) through which the flow passes between the end-end terminals is previously owned by the application, and the traffic parameters of the IP quality assurance service declared by the application.
Assuming traffic limited by TSpec (average rate r, peak rate p, burst size b, maximum packet size M), the bandwidth required to guarantee the end-to-end packet transfer delay time of request delay D _req To the upper limit (R _bound = R)

(Equation 2) It is calculated by: On the other hand, if Adspec puts the R _bound = p unless the application is notified.

Next, assuming a traffic model generated by the application, the probability Pr (W> D) that the end-to-end packet transfer delay time W exceeds the required delay D _req
req ) can be approximately solved using a parameter indicating the statistical characteristics of the model assuming the average rate r of TSpec and the guaranteed rate R. Therefore, the application itself holds statistical traffic characteristics other than TSpec, and α _{denotes the} probability that the probability that the end-to-end packet transfer delay time exceeds the required delay D _req will not hinder the user or the application, and > D _req ) =
From α, an approximate value R _prob = R of the band necessary to stochastically guarantee the end-to-end packet transfer delay time of the required delay D _req can be calculated.

Then, by combining the upper limit value R _{bound of the} band and the probabilistic approximation value R _prob , the equation R = (1−β) R _bound
The guaranteed band R is estimated by + βR _prob (0 ≦ β ≦ 1). However, the value of β is fixedly embedded in the application, selected by the user himself based on experience, or calculated using the optimum value of the guaranteed bandwidth R by the method of [Method 3] described later. become. As described above, according to this method, the upper limit value R _bound based on TSpec that can become excessive estimate is assumed, between the low accuracy of the stochastic approximation R _prob estimated from statistics of the traffic RSpe
By setting the guaranteed bandwidth R of c, it is possible to eliminate over- or under-allocation. As a result, it is possible to eliminate the waste of the resource amount and eliminate the quality deterioration due to the estimation error of the bandwidth.

[Method 2] <Method Including Selection Mechanism of Guaranteed Band R in Parameter Negotiation at Flow Setting> According to [Method 1], the guaranteed band R of RSpec is set to the upper limit R
_By making the estimation smaller than the _{bound and} larger than the low-precision stochastic approximation R _prob, it is possible to eliminate the waste of the resource amount and eliminate the quality deterioration due to the estimation error of the bandwidth. However, when the flow is set with the initially determined parameters and the flow is not accepted with that parameter value,
It does not include how to change parameter values when resetting. That is, the upper limit R _bound or the stochastic approximation R _prob
In addition to the problem of the estimation error when estimating the guaranteed bandwidth R by using the parameter alone, there is a problem that there is no parameter selection guideline such as a changeable range of the guaranteed bandwidth R when negotiating the flow setting parameters (Problem 3). . Therefore, we propose a method that includes a mechanism for updating the parameter value when the flow is not accepted and the network is negotiated with the network and reset.

The state of the network through which the flow passes between the end-end terminals, Adspec (C _tot , D _tot ) is owned by the application in advance, and the traffic parameter of the IP quality assurance service declared by the application.
Assuming traffic limited by TSpec (average rate r, peak rate p, burst size b, maximum packet size M), the bandwidth required to guarantee the end-to-end packet transfer delay time of request delay D _req To the upper limit (R _bound = R)

(Equation 3) It is calculated by: On the other hand, if Adspec puts the R _bound = p unless the application is notified.

Next, assuming a traffic model generated by the application, the probability Pr (W> D) that the end-to-end packet transfer delay time W exceeds the required delay D _req
req ) can be approximately solved using a parameter indicating the statistical characteristics of the model assuming the average rate r of TSpec and the guaranteed rate R. Therefore, the application itself holds statistical traffic characteristics other than TSpec, and α _{denotes the} probability that the probability that the end-to-end packet transfer delay time exceeds the required delay D _req will not hinder the user or the application, and > D _req ) =
From α, an approximate value R _prob = R of the band necessary to stochastically guarantee the end-to-end packet transfer delay time of the required delay D _req can be calculated.

Then, assuming traffic limited by the traffic parameter (TSpec) declared by the application, the upper limit value (R) of the bandwidth required to guarantee the end-to-end packet transfer delay time of the request delay D _{req. The} probability that the end-to-end packet transfer delay time exceeds the required delay D _req is less than a value (α) that does not hinder the user or the application by utilizing the statistical traffic characteristics held by the application and the application itself. , TSpec and the approximate value (R _prob ) of the band calculated from the approximate expression of the end-to-end packet transfer delay time distribution that can be expressed using the statistical traffic characteristics, the first set value is accepted by the network. If no, the guaranteed bandwidth R of RSpec requesting the network in the range of R _prob ≦ R ≦ R _bound (R _bound -R _prob) /
The parameter is determined by negotiating with a step size of n (n> 0).

Note that the RSpe first requested to the network
The guaranteed bandwidth R of c is given by R = (1−β) R shown in [Method 1].
There is also a method of declaring with _bound + βR _prob (0 ≦ β ≦ 1), and when the value is not accepted, the value is gradually reduced to a value accepted with the above-mentioned step size. As described above, in the present method, the change criterion of the guaranteed bandwidth R in the parameter negotiation at the time of setting the flow is based on the upper limit R _bound and the stochastic approximation R
Formulated using _prob . As a result, once the RSpec parameters adopted by the application are determined, they are fixed, and even if they can be changed, there is no changeable range or change guideline, and the user has to change to an appropriate value by trial and error. Is eliminated.

[Method 3] <Method of Determining Guaranteed Bandwidth R Including Update Procedure Based on Measurement of Packet Transfer Delay D ^* > According to [Method 1], the guaranteed bandwidth R of RSpec is changed to the upper limit R
_By making the estimation smaller than the _{bound and} larger than the low-precision stochastic approximation R _prob, it is possible to eliminate the waste of the resource amount and eliminate the quality deterioration due to the estimation error of the bandwidth. However, since there is a fluctuation due to a change in the usage situation of the user and there is an estimation error of the stochastic approximate value R _prob , the value is not always optimal in [Method 1] (Problem 4). [Method 2] proposes a method including negotiation with a network. However, in this case, an estimation error is also included, and the optimum value is not obtained. Therefore, in order to eliminate the estimation error of the guaranteed bandwidth R estimated at the start of the application, a method including a mechanism for updating R so as to approach an optimal value by measuring traffic during operation of the application is proposed.

The state of the network through which the flow flows between the end-end terminals Adspec (C _tot , D _tot ) is previously owned by the application, and the traffic parameters of the IP quality assurance service declared by the application
Assuming traffic limited by TSpec (average rate r, peak rate p, burst size b, maximum packet size M), the bandwidth required to guarantee the end-to-end packet transfer delay time of request delay D _req To the upper limit (R _bound = R)

(Equation 4) It is calculated by: On the other hand, if Adspec puts the R _bound = p unless the application is notified.

Next, assuming a traffic model generated by the application, the probability Pr (W> D) that the end-to-end packet transfer delay time W exceeds the required delay D _req
req ) can be approximately solved using a parameter indicating the statistical characteristics of the model assuming the average rate r of TSpec and the guaranteed rate R. Therefore, the application itself holds statistical traffic characteristics other than TSpec, and α _{denotes the} probability that the probability that the end-to-end packet transfer delay time exceeds the required delay D _req will not hinder the user or the application, and > D _req ) =
From α, an approximate value R _prob = R of the band necessary to stochastically guarantee the end-to-end packet transfer delay time of the required delay D _req can be calculated.

Then, assuming traffic limited by the traffic parameter (TSpec) declared by the application, the upper limit (R) of the bandwidth required to guarantee the end-to-end packet transfer delay time of the request delay D _{req. bound} ), and the probability that the end-to-end packet transfer delay time exceeds the required delay D _req will be less than a value (α) that does not hinder the user or the application by utilizing the statistical traffic characteristics held by the application itself. , TSpec and the approximate value (R _prob ) of the band calculated from the approximate expression of the end-to-end packet transfer delay time distribution that can be expressed using the statistical traffic characteristics, while operating the application, measured packet transfer delay D ^*, Make the transfer delay time exceeding the required delay D _req There _{^{Pr (D *> D req)}} <
The guaranteed bandwidth R of RSpec required for the network is set so that R _prob ≦ R ≦ R _bound so as to satisfy α.
The optimum parameter is determined by adjusting the value by increasing or decreasing the range.

In this case, the range of increase or decrease is (R _bound -R
_prob ) / n (n> 0). Also, at the start of the application, the guaranteed bandwidth R of the RSpec required for the network is represented by R = (1-
β) R _bound + βR _prob (0 ≦ β ≦ 1), and there is also a method of measuring the transfer delay time D ^* at the time of operating the application and gradually changing the value to an appropriate value in the above-mentioned step size. Further, in order to update β included in the calculation formula of R to an appropriate value, the optimum value of R found here is used and calculated assuming that this is equal to “(1−β) R _bound + βR _prob ”. There is also a method of adopting β as a new value.

As described above, in the present method, the end-to-end packet transfer delay time D ^* is measured, and if the guaranteed bandwidth R is not optimal, the upper limit value R _bound and the stochastic approximation value R _prob
The guaranteed bandwidth R according to the change width standard formulated using
Has changed. As a result, when the estimation error of the RSpec parameter at the time of flow setting is corrected to the optimum value during the operation of the application, there is no changeable range or change guideline even if the parameter can be changed. However, the problem of having to change to an appropriate value by trial and error is solved.

[Method 4] <Method including update of statistical traffic characteristic parameter of application by traffic measurement> According to [Method 1], the guaranteed bandwidth R of RSpec is changed to the upper limit R
_By making the estimation smaller than the _{bound and} larger than the low-precision stochastic approximation R _prob, it is possible to eliminate the waste of the resource amount and eliminate the quality deterioration due to the estimation error of the bandwidth. However, since there is a fluctuation due to a change in the use situation of the user and there is an estimation error of the probabilistic approximate value R _prob , the optimal value is not always obtained in [method 1]. [Method 2] proposes a method including negotiation with a network. However, in this case, an estimation error is also included, and the optimum value is not obtained. Further, in [method 3], in order to eliminate the estimation error of the guaranteed bandwidth R estimated at the start of the application, a method including a mechanism for updating the guaranteed bandwidth R to an optimum value while measuring the traffic during the operation of the application is included. Suggested.

However, unless the initial estimation error is reduced, the updating process cannot be executed efficiently. Therefore, in order to reduce the initial estimation error as much as possible (Issue 5), the statistical traffic characteristic for calculating the stochastic approximation R _prob that causes this error is flexibly changed according to the operating environment of the application. A method including processing is proposed.

That is, the traffic is measured at the time of operation of the application, and the statistical traffic characteristic held by the application itself (the parameter expressing this is X _k
Based with the to), the set value of the measurement time points (X _0k) and the calculated statistical value from the measurement data ^{_{(X * k), X k}} = (1-
γ _k ) X _0k + γ _k X ^* _k (0 ≦ γ _k ≦ l) is updated and held. Then, when the next application is executed, the guaranteed bandwidth R is estimated by calculating the probabilistic approximate value R _prob using the updated parameters. Note that this method 4 is used in combination with each of [method 1] to [method 3]. As described above, according to the present method, the statistical traffic characteristic parameter of the application for estimating the probabilistic approximate value R _prob is updated based on the measurement. As a result, the guaranteed bandwidth R follows actual statistical characteristics that fluctuate depending on the operating environment of the application and the usage of the user, and it is possible to eliminate estimation errors due to usage conditions.

[Method 5] <Method of Using GI / D / 1 Queuing System as Statistical Traffic Characteristics> In [Method 1] to [Method 4], the stochastic approximate value R
_It is necessary to assume some traffic model to calculate _prob . However, the commonly used MMP
In P, there is a problem that it is not easy to convert a measurable statistic into a model parameter, and it takes time to solve as a numerical value (problem 6). Therefore, a probabilistic approximation R is calculated using a traffic model that includes a measurable statistic as a model parameter and can easily calculate a numerical solution.
We propose a method of calculating _prob and estimating the guaranteed bandwidth R based on it.

Here, a general distribution input model defined by an average rate (r), an asymptotic dispersion constant (c ² ), and a normalized third moment (m ₃ ) is adopted as a traffic model. Here, since the average rate is the same as that of TSpec, the statistical traffic characteristics to be held by the application are the asymptotic dispersion constant c ² and the normalized third moment m ₃ . When the GI / D / 1 queuing system is analyzed using these characteristic parameters and the average rate r of TSpec, the approximate expression of the end-to-end packet transfer delay time distribution is expressed by the following expression (AWBerger and W. Whit)
t, “Maximum Values in Queueing Processes,” Proba
bility in the Engineering and Information Science
s, Vol. 9, 1995.).

(Equation 5) However,

(Equation 6)

Using this approximate expression, Pr (W>
R _prob = R is calculated so that D _req ) = α. This probabilistic approximate value R _prob is used in any of [method 1] to [method 4] to estimate the guaranteed band R.
In particular, the procedure for updating the statistical traffic characteristics when combined with [Method 4] is as follows.

That is, when the GI / D / 1 queuing system is used, asymptotic dispersion constant c ² and standardized third moment m ₃ exist as statistical traffic characteristics held by the application. The application measures the average EA (t), the variance VarA (t), and the normalized third moment m ^* ₃ of the traffic amount A (t) input to the network from the start to the time t, and asymptotically measures The dispersion constant

(Equation 7) Is calculated by Then, the set value c ₀ ² , m at the time of measurement
₀₃ and the measurement statistics c ^* 2 calculated from the data, m ^* ₃ ^{Based, c 2 = (1-γ} c2) c 0 2 + γ c2 c * 2, m 3 = (1-γ
_m3 ) Update to m ₀₃ + γ _m3 m ^* ₃ and hold.

As described above, in the present method, the GI / D / 1 queuing system using the measurable statistic as a parameter is employed, and the approximate expression of the end-to-end packet transfer delay time distribution obtained therefrom is used. Probabilistic approximation R _prob
Is estimated. Thus, as in the case of estimating the probabilistic approximation R _prob using the MMPP, the measurable statistic is converted into its model parameters (each time duration in the high load / low load state and the load in each state). This solves the problem that the Laplace transform and the iterative calculation are required to solve the trouble and the calculation takes time.

[Method 6] <As Statistical Traffic Characteristics
Fractional Brownian Motion (Method of Using FBM Traffic Model) In [Method 1] to [Method 4], the stochastic approximation R _prob
It is necessary to assume some traffic model in order to calculate. However, the commonly used MMPP
Then, there is a problem that it is not easy to convert a measurable statistic into a model parameter, and it takes time to solve as a numerical value (Problem 6).

Further, in LAN traffic and the like, the burst has no characteristic length (or the burst length exists over a very wide width), and when the time unit is changed,
It is known to have a statistical property that the variance increases exponentially with increasing time units. If the application traffic has this property, the MM
In the traffic model shown in PP and [Method 5], there is a problem that such statistical characteristics cannot be expressed (Problem 7). Therefore, for traffic having no characteristic length in the burst, a measurable statistic is included as a model parameter, and a probabilistic approximate value R _prob is calculated using a traffic model that can easily calculate a numerical solution. Based on this, a method for estimating the guaranteed bandwidth R is proposed.

Here, the variance-average ratio (a) and the self-similarity parameter (H) are adopted as the statistical traffic characteristics held by the application, and are statistically defined by these and the average rate (r) of TSpec. The FBM traffic model is adopted as input traffic. Analyzing the queuing system with this traffic input, the end-to-end packet transfer delay time distribution is obtained as follows (I. Norros, “A Storage Model with Self-similar In
put, “Queueing Systems, Vol. 16, 1994.).

(Equation 8)

Using this, Pr (W> D _req ) = α
An approximate value R _prob = R of the band is calculated so that This probabilistic approximate value R _prob is used in any of [method 1] to [method 4] to estimate the guaranteed band R.
In particular, the procedure for updating the statistical traffic characteristics when combined with [Method 4] is as follows.

That is, the application measures the average EA (t) and the variance VarA (t) of the traffic amount A (t) input to the network from the start to the time t, and calculates the average at a plurality of t. Find the variance.
Then, the relational expression assumed in this model:

(Equation 9) The relation:

(Equation 10)And based on the statistics measured for some t
, Using least square approximation^*= H and a^*= A
calculate. And the set value a at the time of measurement₀ , H ₀And measurement
Statistical value a calculated from constant data^*, H^*Based on
a = (1-γ_a) A ₀+ Γ_aa^*, H = (1−γ_H) H₀+ Γ
_HH^*Update and hold.

As described above, in the present method, the FBM traffic model using the measurable statistic as a parameter is employed, and the probabilistic approximation is obtained by using the approximate expression of the end-to-end packet transfer delay time distribution obtained therefrom. The value R _prob is estimated. Thus, as in the case of estimating the probabilistic approximation R _prob using the MMPP, the measurable statistic is converted into its model parameters (each time duration in the high load / low load state and the load in each state). This solves the problem that the Laplace transform and the iterative calculation are required to solve the trouble and the calculation takes time. Also, since the FBM traffic model has a statistical property that the variance increases exponentially with the increase of the time unit when the time unit is changed, it cannot cope with the traffic input having no characteristic length of the burst. Problems can be solved.

[Embodiment of Method 1] FIG. 3 shows an embodiment for realizing a parameter estimation method according to [Method 1].
Shown in In this figure, the same components as those in FIGS. 1 and 2 are denoted by the same reference numerals. When an application sets a flow, first, the upper limit R _bound calculation unit 2
However, the request delay D _req and IP
The traffic parameter TSpec (average rate r, peak rate p, burst size b, maximum packet size M) of the quality assurance service is received, and the state of the network Adspec (C _tot , C _tot ,
D _tot ) is received from the network 4. Then, based on this, the calculation formula shown in [Method 1] is:

[Equation 11] Calculates the upper limit value R _bound = R of the band required to guarantee the end-to-end packet transfer delay time of the requested delay D _req , and passes the result to the R calculation unit 7. On the other hand, if Adspec
If _bound has not been notified to the application, R _bound =
Put p.

At the same time, the stochastic value R_probThe calculation unit 6
Request delay D from application body 3 _reqAnd allow
Receiving the probability α and the average rate r of TSpec
Based on the statistical properties of the traffic
Next, the end-to-end packet transfer delay shown in [Method 1] is described.
Delay time W is required delay D_reqProbability Pr (W>
D_req): Pr (W> D)_req) = Α
Stochastic approximation R_prob= R is calculated and the result is sent to the R calculation unit 7
hand over. Here, assuming a traffic model, Pr
(W> D_req) Is the model parameter and the guaranteed bandwidth R
And TSpec is a function of the average rate r. In particular, stochastic values
R_probStatistical properties of traffic held by the calculation unit 6
Is the asymptotic dispersion constant c assuming GI^TwoAnd standardized 3rd mode
Ment m_ThreeAnd assuming FBM, the variance-average ratio a
And the self-similarity parameter H.

Finally, the R calculation unit 7 combines the upper limit value R _bound and the stochastic approximation value R _prob to obtain a relational expression shown in [Method 1]: R = (1−β) R _bound + βR _prob (0 ≦ β ≤
Estimate the guaranteed bandwidth R by 1) and return it to the application main body 3. However, the value of β is fixedly embedded in the application, selected by the user himself based on experience, or calculated using the optimum value of the guaranteed bandwidth R by the method of [Method 3]. . The application requests the network 4 to set the flow, using the guaranteed bandwidth R as RSpec. When the flow is set, data is transmitted from the transmitting terminal 5.

[Embodiment of Method 3] FIG. 4 shows an embodiment for realizing a parameter estimating method according to Method 3. In this figure, the same components as those in FIGS. 1 to 3 are denoted by the same reference numerals. When the application sets a flow, first, the upper limit R _bound calculation unit 2 sends a request delay D _req and a traffic parameter TSpec (average rate r, peak rate p, burst size b) of the IP quality assurance service from the application main body 3. , The maximum packet size M), and the state of the network Adspec (C _tot , D _tot ) through which the flow passes between the end-end terminals
From the network 4. Then, based on this, the calculation formula shown in [Method 1] is:

(Equation 12) Calculates the upper limit value R _bound = R of the band required to guarantee the end-to-end packet transfer delay time of the requested delay D _req , and passes the result to the R calculation unit 17. On the other hand, if Adsp
If ec is not notified to the application, R
Let _bound = p.

At the same time, the stochastic value R_probThe calculation unit 6
Request delay D from application body 3 _reqAnd allow
Receiving the probability α and the average rate r of TSpec
Based on the statistical properties of the traffic
Next, the end-to-end packet transfer delay shown in [Method 1] is described.
Delay time W is required delay D_reqProbability Pr (W>
D_req): Pr (W> D)_req) = Α
R_prob= R is calculated, and the result is passed to the R calculation unit 17. here
Assuming a traffic model, Pr (W>
D_req) Is the model parameter, guaranteed bandwidth R and TSpe
It is a function of the average rate r of c. In particular, the stochastic value R_prob
The statistical property of the traffic held by the calculation unit 6 is G
Assuming I, the asymptotic dispersion constant c^TwoAnd standardized third moment
Tom_ThreeAssuming FBM, the variance average ratio a and the self
This is the similarity parameter H.

Then, the R calculating section 17 calculates the upper limit R
Combining the _bound and the stochastic approximation R _prob [Method 1]
R = (1−β) R _bound + βR _prob (0 ≦
The guaranteed bandwidth R is estimated based on β ≦ 1) and returned to the application main body 3. However, the value of β is fixedly embedded in the application, selected by the user himself based on experience, or guaranteed bandwidth R by the method of [Method 3].
(That is, recalculation of the guaranteed bandwidth R shown below). The application requests the network 4 to set the flow, using the guaranteed bandwidth R as RSpec. When the flow is set, data is transmitted from the transmitting terminal 5.

Thereafter, the packet is monitored by the traffic measuring unit 8 and the packet is analyzed by the delay time D ^* calculating unit 9 to calculate the end-to-end delay time D ^* . For example, if Real-time Transport Protocol (RTP) is used, the delay time D ^{* of the} packet can be calculated by subtracting the time of the time stamp of the packet header from the current time. The R calculation unit 17 recalculates the guaranteed bandwidth R by the method shown in [Method 3] in consideration of the difference between the delay time D ^* and the required delay D _req , and passes the result to the application main body 3. The application requests the network 4 to change the flow band with the guaranteed band R as RSpec.

[Embodiment of Method 4] FIG. 5 shows an embodiment for realizing a parameter estimating method according to [Method 4].
Shown in In this figure, the same components as those in FIGS. 1 to 4 are denoted by the same reference numerals. When an application sets a flow, first, the upper limit R _bound calculation unit 2
However, the request delay D _req and IP
The traffic parameter TSpec (average rate r, peak rate p, burst size b, maximum packet size M) of the quality assurance service is received, and the state of the network Adspec (C _tot , C _tot ,
D _tot ) is received from the network 4. Then, based on this, the calculation formula shown in [Method 1] is:

(Equation 13) Calculates the upper limit value R _bound = R of the band required to guarantee the end-to-end packet transfer delay time of the request delay D _req , and passes the result to the R calculation unit 7. On the other hand, if Adspec
If _bound has not been notified to the application, R _bound =
Put p.

At the same time, the stochastic value R_probThe calculation unit 6
Request delay D from application body 3 _reqAnd allow
Receiving the probability α and the average rate r of TSpec
Based on the statistical properties of the traffic
Next, the end-to-end packet transfer delay shown in [Method 1] is described.
Delay time W is required delay D_reqProbability Pr (W>
D_req): Pr (W> D)_req) = Α
R_prob= R is calculated, and the result is passed to the R calculation unit 7. here
Assuming a traffic model, Pr (W> D_req)
Is the average of the model parameters and the guaranteed bandwidth R and TSpec.
It is a function of the average rate r. In particular, the stochastic value R_probCalculator
The statistical properties of the traffic held by GI
If asymptotic dispersion constant c^TwoAnd normalized third moment m_Threeso
If FBM is assumed, the variance-average ratio a and the self-similarity
Parameter H.

Then, the R calculation unit 7 combines the upper limit value R _bound and the stochastic approximation value R _prob to obtain a relational expression shown in [Method 1]: R = (1−β) R _bound + βR _prob (0 ≦ β ≦
Estimate the guaranteed bandwidth R by 1) and return it to the application main body 3. However, the value of β is fixedly embedded in the application, selected by the user himself based on experience, or calculated using the optimum value of the guaranteed bandwidth R by the method of [Method 3]. . The application requests the network 4 to set the flow, using the guaranteed bandwidth R as RSpec. When the flow is set, data is transmitted from the transmitting terminal 5.

Thereafter, traffic is measured by the traffic measuring unit 8, and the traffic statistics calculating unit 10 estimates a statistic representing the characteristics of the measured traffic by the method shown in [Method 4], and uses this to calculate a probabilistic value. An R _prob calculation unit 6 calculates a probabilistic approximate value R _prob . The R calculation unit 7 holds the statistics for use in the next flow setting.

[Embodiment in which Method 3 and Method 4 are Combined] FIG. 6 shows an embodiment in which the parameter estimation methods in Method 3 and Method 4 are combined. In this figure, the same components as those in FIGS. 1 to 5 are denoted by the same reference numerals. The operation in this case is described above [Embodiment 3 of Method 3]
And [Embodiment of Method 4] is clear from the description, and the description is omitted here.

[Application Example to Real Traffic] In order to evaluate the effect of applying the present invention to actual application traffic, a simulation is performed using traffic data measured on Ethernet. Here, assuming an IP network in which the transmitting and receiving terminals are connected via one router, the packet transmission from the router is W
Assume that FQ scheduling is followed. It is assumed that video conferencing (H.261 encoding) and Internet telephony (Intel version) are adopted as applications, and data of all traffic (LAN traffic) measured on Ethernet is flown in the background. The measurement traffic data used in the simulation has a set of a packet generation time (unit: seconds) and its size (unit: bytes) arranged in the order of generation. FIG. 7 shows the statistics of the measurement data of each application traffic.

From the measured traffic data, the TSpec
Estimate (p, r, b, M). In addition, the user has D _req
Assuming that a delay time of ≦ 1 second is required, the upper limit value R _{bo und} of the guaranteed bandwidth can be calculated according to [ _Prior Art 1]. Further, the stochastic guaranteed band R _prob is set such that D _req ≒ 1 second, for example, with a probability of α _{例 え ば} 10 ⁻³ [Prior Art 2].
It can be calculated by applying the method of [Method 5] or [Method 6] to the method of [1]. Here, it is assumed that [method 6] is applied. Further, in order to evaluate under an unfavorable condition, the link speed μ transmitted from the passing router is set to be slightly larger than the sum of the average rate of the application traffic of interest and the average rate of the LAN traffic.
These numerical values used in the actual simulation are shown in FIG.

First, the effect of applying this parameter determination method to the traffic of a video conference will be evaluated. If the guaranteed band R = R _bound , if the guaranteed band R = R _prob , β
FIG. 9 shows the distribution of packet transfer delay time between end-to-end terminals for the method of [Method 1] with = 0.9. FIG. 3 also shows a case without bandwidth guarantee for reference.

As can be seen from the figure, in the case of R = R _bound where the upper limit value is the guaranteed band, most of the packets have a delay time of 0.2 seconds or less despite the required delay of 1 second. ing. On the other hand, in the case of R = R _prob where the band calculated stochastically based on the statistical characteristics of the application traffic is the guaranteed band, about 5% of the packets exceed the required delay of 1 second. On the other hand, in the method of [Method 1] in which the upper limit value R _bound and the probabilistic approximation value R _prob are combined, since the distribution is intermediate between the two, excessive bandwidth allocation is not performed and the required delay time is reduced to a realistic value. Ratio (1 in this case)
0 ^-3 or less).

Next, the effect of applying the present invention to Internet telephone traffic will be evaluated. Here, when the guaranteed bandwidth R = R _bound , when the guaranteed bandwidth R = R _prob ,
FIG. 10 shows the packet transfer delay time distribution between end-to-end terminals for the method of [Method 1] with β = 1.0. FIG. 3 also shows a case without bandwidth guarantee for reference.

As can be seen from the figure, in the case of R = R _bound where the upper limit is the guaranteed band, most of the packets have a delay time of 0.1 second or less even though the required delay is 1 second. ing. Also, in the case of R = R _prob where a band calculated stochastically based on the statistical characteristics of application traffic is a guaranteed band, the delay time of most packets is 0.3 seconds or less. On the other hand, the upper limit R _bound
Combining stochastic approximation R _prob with the method of [method 1], beta = 1 and since the are and has become the same distribution as for R = R _prob, R = as when you select R _bound excess bandwidth Not assigned. As described above, since the estimation error of the stochastic approximate value R _prob may be large,
With the method of [Method 1] alone, the guaranteed bandwidth R determined first does not always become an optimal value. Therefore, in such a case, in order to approach the optimum value, it is necessary to adopt a procedure for updating the guaranteed bandwidth R based on the measurement of [Method 3].

[0084]

As described above, according to the first or ninth aspect of the present invention, the guaranteed bandwidth of the quality parameter required for the IP network is not affected by the probability that the packet transfer delay time exceeds the required delay. Since the bandwidth is set between the approximate value of the bandwidth without the bandwidth and the upper limit of the bandwidth required for guaranteeing the required delay, it is possible to eliminate over- or under-allocation, and to reduce the estimation error of the bandwidth without wasting resources. This has the effect of eliminating quality deterioration due to

According to the second or tenth aspect of the present invention, when the initial setting of the guaranteed bandwidth R is not accepted by the IP network, the guaranteed bandwidth of the quality parameter required for the IP network is changed to the packet transfer delay time. Between the approximate value of the band in which the probability of exceeding the required delay does not hinder the application and the upper limit value of the band required for guaranteeing the required delay, with a step size obtained by equally dividing the approximate value and the upper limit value by n Since it is reduced, it is possible to formulate a change criterion of the guaranteed bandwidth in the parameter negotiation at the time of setting the flow. Therefore, once the quality parameter adopted by the application is determined, it is fixed, and even if it can be changed, there is no changeable range or change guideline, and the user has to change to an appropriate value by trial and error. Has an effect.

According to the fourth or twelfth aspect of the present invention, the packet transfer delay time D ^* is measured while operating the application, and the probability that this value exceeds the required delay is a value α which does not hinder the application or the like. The probability that the packet transfer delay time exceeds the required delay so that the packet transfer delay time exceeds the required delay is between the approximate value of the band that does not hinder the application and the upper limit value of the band required for guaranteeing the required delay. The guaranteed bandwidth of the required quality parameter is increased or decreased. As a result, the estimation error of the quality assurance parameter at the time of setting the flow can be corrected to an optimal value as much as possible during the operation of the application. Therefore, even if such parameters can be changed, there is no changeable range or change guideline, and the problem that the user or the application has to change to an appropriate value by trial and error is solved.

According to the invention of claim 6 or claim 14, traffic is measured during operation of the application,
Based on the statistical value calculated from the measured traffic data, the parameter expressing the statistical traffic characteristics of the application is updated, and an approximate value of the subsequent band is calculated. The guaranteed bandwidth of the quality assurance parameter follows actual statistical characteristics that fluctuate depending on the operating environment of the application and the usage of the user, and thus has the effect of eliminating estimation errors due to usage conditions.

According to the present invention, a packet transfer delay time distribution obtained by analyzing a GI / D / 1 queuing system of general distribution traffic input using a measurable statistic as a parameter. The approximate value of the band is calculated using the approximate expression. As a result, as in the case of estimating an approximate value of a band using MMPP, it is necessary to convert measurable statistics into MMPP model parameters, and it is necessary to perform Laplace transform and iterative calculation when solving the problem. This has the effect of solving the problem that the calculation takes time.

According to the present invention, an approximate expression of a packet transfer delay time distribution obtained by analyzing an FBM traffic input queuing system using measurable statistics as a parameter is used. To calculate the approximate value of the band. As a result, as in the case of estimating an approximate value of a band using MMPP, it is necessary to convert measurable statistics into MMPP model parameters, and it is necessary to perform Laplace transform and iterative calculation when solving the problem. This has the effect of solving the problem that the calculation takes time. In addition, the FBM traffic model has a statistical property that when the time unit is changed, the variance increases exponentially with the increase in the time unit. There is also an effect that problems that could not be solved can be solved.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a conventional parameter estimation method proposed in an Internet draft.

FIG. 2 is a block diagram illustrating a conventional parameter estimation method for stochastically guaranteeing a packet transfer delay.

FIG. 3 is a block diagram illustrating a parameter estimation method of [method 1] according to an embodiment of the present invention.

FIG. 4 is a block diagram illustrating a parameter estimation method of [method 3] in the embodiment.

FIG. 5 is a block diagram illustrating a parameter estimation method of [method 4] in the embodiment.

FIG. 6 is a block diagram illustrating a parameter estimating method when [method 3] and [method 4] in the embodiment are combined.

FIG. 7 is a chart showing statistics of measurement data of each application traffic used in a simulation for evaluating the effect of the present invention.

FIG. 8 is a table showing numerical values of simulation conditions for performing the simulation.

FIG. 9 is a graph showing a packet transfer delay time distribution in a case where the present invention is applied to video conference traffic.

FIG. 10 is a graph showing a packet transfer delay time distribution when the present invention is applied to Internet telephone traffic.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Receiving terminal 2 Upper limit R _bound calculating unit 3 Application main body 4 Network 5 Transmitting terminal 6 Stochastic value R _prob calculating unit 7, 17 R calculating unit 8 Traffic measuring unit 9 Delay time D ^* calculating unit 10 Traffic statistic calculating unit

Claims (16)

[Claims] 1. A router having a scheduling mechanism capable of guaranteeing a bandwidth for each application flow via an IP network connected to a network,
An application operating between mutually connected terminals declares a traffic parameter that defines an upper limit of traffic, requests a guarantee of a required delay D _req as an end-to-end packet transfer delay time, and uses an IP quality assurance service. In this case, in the method of determining the parameter of the IP quality assurance service by the application that determines the value of the guaranteed bandwidth R, which is a quality parameter that defines the IP quality assurance service, when traffic limited by the traffic parameter is assumed, end of the request delay D _req - the upper limit value R _bound bandwidth needed to guarantee end packet transfer delay of the end - the probability that the end packet transfer delay time exceeds the required delay D _req is, the user Or it is less than the value α that does not hinder the application To the application end is expressed by the statistical traffic characteristics and the traffic parameter held by itself - on the basis of the approximate value R _prob bands calculated from the approximate expression of the end packet transfer delay time distribution, relation R = (1-β) R _bound + βR _prob (0 ≦ β ≦
1) A method for determining parameters of an IP quality assurance service by an application, wherein the guaranteed bandwidth R is estimated by: 2. A router having a scheduling mechanism capable of guaranteeing a bandwidth for each application flow via an IP network connected to a network.
An application operating between mutually connected terminals declares a traffic parameter that defines an upper limit of traffic, requests a guarantee of a required delay D _req as an end-to-end packet transfer delay time, and uses an IP quality assurance service. In the method for determining a parameter of the IP quality assurance service by an application for determining a value of the guaranteed bandwidth R, which is a quality parameter defining the IP quality assurance service, the first set value of the guaranteed bandwidth R is accepted by the IP network. Otherwise, assuming traffic limited by the traffic parameters, the upper limit R _{bound of the} bandwidth required to guarantee the end-to-end packet transfer delay time of the request delay D _req , end - end of packet transfer delay is the required delay D _req The approximate expression of the statistical traffic characteristics held by the application itself and the end-end packet transfer delay time distribution represented by the traffic parameters so that the probability of exceeding the value is less than a value α that does not hinder the user or the application. _Is calculated based on the approximate value R _{prob of the} band, and the guaranteed band R of the quality parameter required for the IP network is set within the range of R _prob ≦ R ≦ R _bound (R
_A method for determining parameters of an IP quality assurance service by an application, wherein the estimation is performed by reducing the step size by _bound- R _prob ) / n (n> 0) and negotiating with the IP network. 3. The guaranteed bandwidth R at the start of an application according to the method of estimating the guaranteed bandwidth R according to claim 1.
And if the guaranteed bandwidth R is not accepted by the IP network, the guaranteed bandwidth R is sequentially changed and reset according to the method of negotiating the guaranteed bandwidth R according to claim 2. A method for determining parameters of the IP quality assurance service. 4. A router having a scheduling mechanism capable of guaranteeing a bandwidth for each application flow via an IP network connected to a network.
An application operating between mutually connected terminals declares a traffic parameter defining an upper limit of traffic, requests an assurance of a required delay D _req as an end-to-end packet transfer delay time, and uses an IP quality assurance service. When performing the parameter determination method of the IP quality assurance service by the application that determines the value of the guaranteed bandwidth R, which is a quality parameter that defines the IP quality assurance service, when traffic limited by the traffic parameter is assumed, end of the request delay D _req - the upper limit value R _bound bandwidth needed to guarantee end packet transfer delay of the end - the probability that the end packet transfer delay time exceeds the required delay D _req is, the user Or it will be less than the value α that does not hinder the application To the application end is expressed by the statistical traffic characteristics and the traffic parameter held by itself - on the basis of the approximate value R _prob bands calculated from the approximate expression of the end packet transfer delay time distribution, the application While operating, the end-to-end packet transfer delay time D ^* is measured, and the probability Pr (D ^* >) of the transfer delay time D ^* exceeding the required delay D _req is measured.
D _req) are formed so that the smallest value that satisfies _{^{Pr (D *> D req)}} <α, the guaranteed bandwidth R quality parameter for requesting the IP network is increased or decreased in a range of R _prob ≦ R ≦ R _{bou nd} A method for determining parameters of an IP quality assurance service by an application, wherein the parameters are adjusted. 5. The guaranteed bandwidth R according to claim 4, wherein the guaranteed bandwidth R at the start of the application is determined in accordance with the method for estimating the guaranteed bandwidth R according to claim 1, and the application is operated. A parameter determination method for an IP quality assurance service by an application, wherein the guaranteed bandwidth R is sequentially changed in accordance with the adjustment method. 6. A traffic value is measured during operation of the application, and a parameter X _k expressing the statistical traffic characteristic is changed to a set value X _0k at the time of measurement and a statistical value X calculated from data of the measured traffic. ^* _k and based _{on, X k = (1-γ} k) X 0k + γ k X * k (0 ≦ γ k ≦
The method for determining a parameter of an IP quality assurance service by an application according to any one of claims 1 to 5, wherein the application has a procedure for updating and holding the parameter. 7. An asymptotic dispersion constant c ² and a normalized third moment m ₃ are adopted as the statistical traffic characteristics,
An approximate expression of an end-to-end packet transfer delay time distribution obtained by analyzing a GI / D / 1 queuing system of general distribution traffic input statistically defined by these and an average rate r of the traffic parameters is used. 7. The method according to claim 1, wherein the approximate value R _prob is calculated.
Method for determining parameters of P quality assurance service. 8. A queuing system for an FBM traffic input, which employs a variance-average ratio a and a self-similarity parameter H as the statistical traffic characteristics, and is statistically defined by these and an average rate r of the traffic parameters. 7. The IP quality by the application according to claim 1, wherein the approximate value R _prob is calculated using an approximate expression of an end-to-end packet transfer delay time distribution obtained by analyzing the IP quality. How to determine warranty service parameters. 9. A router having a scheduling mechanism capable of guaranteeing a bandwidth for each application flow via an IP network connected to a network.
An application operating between mutually connected terminals declares a traffic parameter defining an upper limit of traffic, requests an assurance of a required delay D _req as an end-to-end packet transfer delay time, and uses an IP quality assurance service. In doing so, the IP that determines the value of the guaranteed bandwidth R, which is a quality parameter that defines the IP quality assurance service,
In the parameter determination program of the quality assurance service, assuming traffic limited by the traffic parameter, an upper limit value R _{bound of} a band required to guarantee an end-to-end packet transfer delay time of the request delay D _req. And the probability that the end-to-end packet transfer delay time exceeds the required delay D _req , assuming traffic limited by the traffic parameter, will not hinder the user or the application. Approximate value calculation for calculating an approximate value R _{prob of a} band from an approximate expression of an end-end packet transfer delay time distribution represented by the statistical traffic characteristics held by the application and the traffic parameter so as to be less than a non-existent value α. Processing and relational expression R = (1−β) R _bound + βR _{p rob} (0 ≦ β ≦
1) causing a computer in the terminal to execute an estimation process for estimating the guaranteed bandwidth R according to 1).
A recording medium on which a parameter determination program of a P quality assurance service is recorded. 10. An application that operates between terminals connected to each other via an IP network to which a router having a scheduling mechanism that enables bandwidth guarantee for each application flow is connected in a network, When using the IP quality assurance service by declaring the traffic parameter defining the upper limit and requesting the guarantee of the required delay D _req as the end-to-end packet transfer delay time, the quality parameter defining the IP quality assurance service is used. In a parameter determination program for an IP quality assurance service for determining a value of a guaranteed bandwidth R, a request acceptance determination process for detecting that the first set value of the guaranteed bandwidth R has not been accepted by the IP network; Assuming limited traffic An upper limit value calculation process for calculating an upper limit value R _bound of a band required to guarantee an end-to-end packet transfer delay time of the request delay D _req ; and assuming traffic limited by the traffic parameter. The statistical traffic characteristics and the traffic parameters held by the application such that the probability that the end-to-end packet transfer delay exceeds the required delay D _req is less than a value α that does not hinder the user or the application. Approximate value calculation processing for calculating an approximate value R _{prob of the} band from the approximate expression of the end-to-end packet transfer delay time distribution expressed by the following equation: R _prob ≦ R ≦ R Within the scope of the _bound (R
a parameter determination program for an IP quality assurance service, wherein a computer in the terminal executes an estimation process of negotiating with the IP network by decreasing the step size by _bound- R _prob ) / n (n> 0). Recording medium on which is recorded. 11. A guaranteed bandwidth R at the start of an application is determined according to the estimation processing according to claim 9, and it is detected in the request acceptance determination processing according to claim 10 that the guaranteed bandwidth R is not accepted by the IP network. In this case, a parameter determination program for the IP quality assurance service is recorded, wherein the computer in the terminal executes a process of sequentially changing and resetting the guaranteed bandwidth R according to the estimation process according to claim 10. recoding media. 12. An application that operates between terminals connected to each other via an IP network in which a router having a scheduling mechanism for enabling a bandwidth guarantee for each application flow is connected in a network manner. When using the IP quality assurance service by declaring the traffic parameter defining the upper limit and requesting the guarantee of the required delay D _req as the end-to-end packet transfer delay time, the quality parameter defining the IP quality assurance service is used. In a parameter determination program of an IP quality assurance service for determining a value of a certain guaranteed bandwidth R, assuming traffic limited by the traffic parameter, an end-to-end packet transfer delay time of the required delay D _req is guaranteed. Calculate the upper limit R _{bound of the} band required for Upper limit value calculation process, and assuming traffic limited by the traffic parameter, the probability that the end-to-end packet transfer delay time exceeds the required delay D _req is a value that does not hinder the user or the application. an approximate value calculation process of calculating an approximate value R _{prob of a} band from an approximate expression of an end-end packet transfer delay time distribution represented by the statistical traffic characteristics held by the application and the traffic parameters so as to be less than α. The end-to-end packet transfer delay time D ^* is measured while operating the application, and the probability Pr (D ^* >) of the transfer delay time D ^* exceeding the required delay D _req is measured.
D _req) are formed so that the smallest value that satisfies _{^{Pr (D *> D req)}} <α, the guaranteed bandwidth R quality parameter for requesting the IP network is increased or decreased in a range of R _prob ≦ R ≦ R _{bou nd} A computer in the terminal for executing an adjusting process for adjusting the parameter by adjusting a parameter of the IP quality assurance service. 13. The guaranteed bandwidth R at the start of an application is determined according to the estimation processing according to claim 9, and the guaranteed bandwidth R is determined according to the adjustment processing according to claim 12, while operating the application.
A recording medium storing a parameter determination program for an IP quality assurance service, which causes a computer in the terminal to execute a process of sequentially changing the parameters. 14. A traffic measurement process for measuring traffic during operation of the application, and a parameter X _k representing the statistical traffic characteristic.
And the statistical value X ^* _k calculated from the data of the measured set value X _0k at the measurement time traffic based on, X _k =
_{(1-γ k) X 0k} + γ k X * k (0 ≦ γ k ≦ 1) Update To characterized in that to execute the update processing to be retained in the application on a computer in the terminal claim 9 14. A recording medium on which the parameter determination program of the IP quality assurance service according to any one of Items 13 to 13 is recorded. 15. The approximation value calculation process adopts an asymptotic dispersion constant c ² and a normalized third moment m ₃ as the statistical traffic characteristics, and statistically calculates these values based on these and an average rate r of the traffic parameters. GI / D / 1 for defined general distribution traffic input
The IP according to any one of claims 9 to 14, wherein the approximate value R _prob is calculated using an approximate expression of an end-to-end packet transfer delay time distribution obtained by analyzing a queuing system. A recording medium that records a quality assurance service parameter determination program. 16. The approximate value calculation process adopts a variance-average ratio a and a self-similarity parameter H as the statistical traffic characteristics, and is statistically defined by these and an average rate r of the traffic parameters. 15. The method according to claim 9, wherein the approximate value R _prob is calculated by using an approximate expression of an end-to-end packet transfer delay time distribution obtained by analyzing a queuing system for FBM traffic input. A recording medium on which a parameter determination program of the IP quality assurance service described in the item is recorded.