Method for dynamic optimisation of service quality in a datatransmission network

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 Method for dynamic optimization of the quality of service in a data transmission network.

DESCRIPTION The present invention relates to a method and a device for dynamic optimization of the quality of service in a packet mode data transmission network, said network comprising a plurality of sources Si and a plurality of destinations Di connected to a transit network. via a plurality of access networks, each source being capable of sending a maximum bit rate Lmax, s, x, and each destination being capable of receiving a maximum bit rate Lmax, x, a, said sources each comprising a means for classifying and controlling the transmitted data rate.

This method and this device can be implemented regardless of the geographic extent of the network, regardless of the speed conveyed by the latter and regardless of the number of users of this network.

Packet mode telecommunications networks are characterized by the fact that the information conveyed is conveyed in groups called packets, essentially made up of:

- a header containing the information for routing the packet in the network;

- data to be transmitted.

The addressing information contained in the headers is used to identify information flows between end applications.

These surit packets conveyed through the network, and borrow the most varied means of transmission and switching at the option of this network.

CA 02398366 2002-07-24 WO 01/58094 2 PCT / FRO1 / 00324 The technology currently mainly used for these telecommunications networks in packet mode is the IP protocol (Internet Protocol).

This protocol is used from end to end, and can be conveyed on very diverse transmission networks.

Figure 1 gives an example of such a network.

Users can be either individual users, agencies, companies (with their own internal local network), etc.

The transit network represents the central part, generally with a large capacity and covering a large territory (the whole world in the case of the Internet network).

This network is generally shared by a multitude of users and / or private networks.

Access networks are generally medium or slow speed, and shared between users located in a limited geographical area.

The local loop, wire link, optical, radio, etc. between the user and the access service provider is considered hereafter as part of the access network.

Figure 2 shows different possible cases of access networks.

The writing conventions are as follows:

- For Carrier networks (in English): carrier of large amounts of information over long distances; it also provides interconnection with other carriers, thus allowing, in the case of the Internet network, interoperability between the users of the various ISP (Internet Service Provider) Internet service providers.

IAP: (Internet Access Provider in English) network access provider; it collects traffic on behalf of the ISP, the latter provides CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 3 typically to its users various authentication servers, web site hosting, pricing, messaging, etc. as well as access to the transit network.

Local Loop: (local loop in English) link (wired, optical, radio, ...) connecting the user to the network.

TELCO: telephone operator, often owner of the local loop.

- For equipment:

CPE: (Customer Premices Equipment in English) user equipment connected to the network (generally an access router).

MUX: multiplexer / demultiplexer (there are many kinds: telephone, xDSL, SDH, etc. ...).

NAS: (Network Access Server, in English) network access server; it can also be an access router.

A: Router (or switch).

We can see that there are many possible configurations.

Each of the devices (CPE, MUX, NAS, R, etc.) corresponds to a function of concentration of traffic and of pooling of telecommunications resources.

With the prodigious development of information exchanges through telecommunications networks, it is becoming essential for operators to ensure quality of service to their customers.

Quality of Service is made up of all the relevant characteristics affecting the transfer of information between two given points of a network.

We find in particular:

- quality of access to the service - availability of the service;

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 4 - the time to put back into service in the event of a failure.

- quality of the information transfer service:

- the time taken to transfer information between the source and the destination;

- variation of the information transfer delay (jitter);

- degradation of the information conveyed (losses, errors).

A major problem stems from the fact that the geographical scope, the strong mutualisation of infrastructure equipment between a large number of users, the variety of flows exchanged and the complexity of the architectures deployed make it very difficult to predict and guarantee the Quality of Service. on such networks.

The flow that it is possible to pass between two given users, the information transfer delay, the variation in time of this delay (jitter) and the associated loss rate are fundamental elements of this Quality of Service.

Only their mastery makes it possible to deploy critical professional services (transport of voice, images, critical data, electronic commerce1 * _. That, etc. ...).

A common way to improve the quality of service is to oversize the capacity of the network.

However, given the importance of the cost ci'investment and use of these networks, it is desired to use these networks as much as possible, and such a very expensive solution is therefore of limited use.

Devices (protocols, transmission, switching, routing equipment, etc.), depending on the nature of the different networks, can be implemented to manage these Quality elements of CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 Service.

They are generally based on mechanisms of priority and reservation of resources on demand (ATM, RSVP on IP, ...) or on configuration (ATM, DiffServ on IP, ...).

These devices are generally limited in scope to only part of the network.

Constantly changing, they have difficulty interoperating.

In all cases, the result is strongly dependent on the behavior of the source users: transmission rate, traffic regularity, traffic matrix, etc. ...

This behavior is very difficult to predict, due to the wide variety of applications using networks (voice transport, images, file transfer, consultation of databases, etc.), the multiplicity of users brought together and the wide range of their needs.

In all cases also, the result is highly dependent on the engineering rules and the configuration of the multiple parameters of the network.

These rules are very difficult to determine, in particular because of the size of the networks, the wide variety of technologies implemented at a given time (non-homogeneous fleet) and the multiplicity of organizations (service access operators, point-of-presence operators, long-haul carriers, etc.) involved from one end of the road to the other.

Furthermore, the Quality of Service is mainly linked to the congestion of the various network elements used by the information during its transfer.

Although there is an infinite number of gradations, we can schematize the cases of operations encountered by these two modes:

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 6 - either there is no resource reservation, and the network does its best to relay the information to the recipient;

- or there is a resource reservation, and the quantity of information injected into the network is more or less statistically controlled.

In all cases, temporary queued storage systems (memories), located at each multiplexing, concentration or switching point, make it possible to process the incoming simultaneities of the packets.

The instantaneous rate of memory occupancy encountered by a packet and the management policy (priority, number of queues, emptying rule, rejection, etc.) implemented at the level of each queue determine the time passed through a package in this device, as well as its possible rejection.

The transfer delay between two network points is due:

- the sum of the crossing times of lianas, cables, optical fibers, satellite links, etc.

used; this delay is generally fixed, and essentially depends on the medium and the distance traveled by the information.

- the sum of the crossing times of the queues in the different equipment; this delay is due globally to the instantaneous load encountered by each packet and to the policies for managing these queues.

Too high an instantaneous load causes the information packet to be rejected (loss). This phenomenon explains the loss of the packets.

This transfer delay is therefore sensitive to the instantaneous load, to seasonal variations (day / night cycles, peak hours, etc.), to changes in configuration (paths taken in the network, etc.).

The variation in the transfer delay is CA 02398366 2002-07-24 WO 01/58094 PCT / FRO1 / 00324 7 commonly of the order of (or even much greater than) its average value, whether in the short term (instantaneous congestion) long term (user activity cycle).

A more detailed analysis of these congestion phenomena makes it possible to segment the problem.

By examining backwards the path taken by a packet, we can go to the following three segments:

- delay, variation of delay and losses due to the load of the access network towards a given user: the access network is generally slower than the transit network; the sum of the flows towards this user can therefore exceed the capacity of this network.

The behavior here is essentially linked to the behavior of the sources generating packets to the user in question.

- delay, variation of delay and losses due to the load of the transit network are to be attributed to the result of the behavior of a multitude of sources.

It is only slightly correlated with the traffic destined for a given user.

We can see the transit network as a propagation medium whose characteristics (delay, variation of delay, losses) are 1) slowly variable relative to the periods of transmission of the packets, and 2) independent of the transmission of the packets of a given source.

- delay, variation in delay and losses due to the load on a user's access network (towards the transit network) are linked in particular to the quantity of information sent by the user in question.

A known method of controlling congestion in packet mode telecommunications networks is described in American patent US Pat. No. 5,936,940 (Adaptive rate-based congestion control in packet networks).

The principle proposed in this patent is CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 8 illustrated by FIG. 3 and consists in differentiating the network delay which represents the idle delay (transmission lines, etc.) and the queue delay which depends on the injected traffic and is representative of the congestion state of the system.

Test packets are time stamped relative to local times when they are sent and received.

The estimate of the queue delay is based on the difference in the inter-arrival and inter-departure intervals between consecutive test packets.

The recurring estimation formula and its bounding less than zero make it possible to overcome the minimum delay encountered, which represents the fixed part of the delay, the network delay.

This absolute information is used to slave the sources.

The estimated queue time is then used as an indicator to classify the state of the network into a few categories.

Depending on the state of the network, an output limiter on the source is modulated with a non-linear function.

A drawback of this method stems from the fact that it is based on relative measurements, giving an idea of ​​the variation in congestion, but not allowing the severity of this congestion to be known.

This makes real optimization very difficult.

In addition, there are problems with the initialization of the devices.

In addition, these partial or indirect measurements only give a fragmentary view of the situation of the observed network and are not linked to quantifiable objectives of the Quality of Service perceived by the users of the network.

Furthermore, in modern networks, test packets are not representative of useful packets because of the wide variety of paths, especially in queuing devices.

Also, CA 02398366 2002-07-24 WO 01/58094 PCT / FRO1 / 00324 9 the injection of test packets does not allow a sufficient number of measurements and can even generate an exaggerated load. Note, moreover, that the proposed mechanisms assume that all the sources of the network have identical behavior, which is not the case in large networks already deployed.

The object of the invention is a method and a device making it possible to optimize the use and the Quality of Service of networks in packet mode, more particularly, networks operating on the IP protocol.

Another object of this invention is a method and a device making it possible to dynamically manage the congestion of the access network arriving to a user, and particularly, when the congestion point on this access network is not located on the network. access line itself, and when the level of congestion varies over time.

The management of this congestion makes it possible to increase the load while controlling delays, jitter and losses.

Another object of this invention is a method and a device making it possible to dynamically manage the congestion of the access network when leaving a user, and particularly (but not only) when the congestion point on this access network is not is not located on the access line itself, and when the level of congestion varies over time.

Managing this congestion makes it possible to increase the load while controlling delays, jitter and losses.

Another object of this invention is a method and a device for dynamically managing the congestion of the transit network, and when the level of congestion varies over time.

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 management of this congestion making it possible to use other means of telecommunications wisely.

To achieve these goals, the method according to the invention comprises the following steps:

5 a- measure the absolute transfer delay, the jitter, the throughput of said data exchanged and the losses between each source S and each destination D, and perform the following sub-steps as a function of the measurements obtained in this step a:

10 al- for each destination D, distribute the maximum flow Lmax, x, d between the different sources;

a2- for each destination D, dynamically determine the limit of the congestion point of the access network optimizing the maximum throughput Lmax, x, d;

a3- for each source S, dynamically determine from the measurements obtained in step a, the limit of the congestion point of the access network optimizing the maximum overall throughput Lmax, S, x and the maximum throughput towards each destination Lmax, her ;

a4- for each source S, determine the effect of the congestion of the transit network; and b — controlling the classification and flow control means as a function of the results of sub-steps a1, a2, a3 and a4.

Preferably, step a1 comprises the following operations:

- Set the bit rate Lmax, X, d on the one hand, from the bit rate of the access line, and on the other hand, from step a2;

- sharing said maximum throughput Lmax, x, a between the active sources of the sub-network so as to define for each source a first maximum throughput value CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 11 Lmax, s, a, said sharing being carried out taking into account configuration data and any dynamic reservations of throughput Rs, d by all or part of the sources.

- transmitting to each of the sources said first value Lmax, s, a = Step a2 comprises the following operations:

- Calculate a first delay value representing a delay weighting between each of the sources and said destination;

- Calculate a first jitter value representing a jitter weighting between each of the sources and said destination;

- Calculate a first value of losses representing a weighting of the losses between each of the sources and said destination;

- determine the curve of these weighted values ​​as a function of the bit rate received by this destination;

- Set a maximum flow rate value Lmax, x, d as a function of said curves.

Step a3 comprises the following operations:

- Calculate a first delay value representing a delay weight between said source and each of said destinations;

- Calculate a first jitter value representing a jitter weighting between said source and each of said destinations;

- Calculate a first value of losses representing a weighting of the losses between said source and each of said destinations;

determining the curve of these weighted values ​​as a function of the bit rate emitted by said source (overall and to each of the destinations);

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 12 - Fix a maximum overall flow value Lmax, s, x and a second maximum flow value per direction Lmax, S, d as a function of said curves.

Step a4 comprises the following operations:

- compare the measurements of delay, jitter, losses and throughput, and the results of steps a1, a2 and a3 with predefined values ​​of delay, jitter, losses and throughput;

- establish a diagnosis on the quality of service based on this comparison.

According to a second characteristic of the invention, the method further comprises a step consisting in transmitting to the flow control means, the smallest value among said first and second maximum flow rate values ​​Lmax, s, d = This method is implemented. work by a device comprising:

means for measuring the absolute transfer delay, the jitter, the data rate exchanged and the losses between each source S and each destination D;

- Means for distributing, for destination D, the maximum flow Lmax, x, d between the different sources;

means for dynamically determining, for each destination D, the limit of the congestion point of the access network optimizing the maximum throughput Lmax, x, a;

means for dynamically determining, for each source S, the limit of the congestion point of the access network globally and towards each destination;

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 13 - nuclei for determining, for each source S, the effect of the congestion of the transit network towards each destination;

- means of classification and flow control; and - means for controlling the flow control means.

Preferably, said device further comprises a module M1, responsible for dynamically determining a first maximum quantity of information Lmax, S, a that a source S can send to a destination D, a module M2, responsible for dynamically determining the maximum bit rate. global Lmax, x, d that the destination D can receive to have optimal use of the access line while controlling the Quality of Service, an M3 module, responsible for dynamically determining the maximum overall throughput and by destination that the source S can send to have optimal use of the access line while controlling the Quality of Service, an M4 module, responsible for dynamically determining the characteristics of the transit network and to control a network selector and a module M5, dynamically setting the value of limiters of the source S as a function of the values ​​coming from the modules M1 and M3.

The parameters of the module M1 are: the flow measurements;

the maximum value Lmax, x, d that the destination D can receive from all the sources Si, as determined by the module M2;

- possible reservations of flow Rs, d between each source S and destination D.

The module M2 calculates, from the Quality of Service measurements, the Quality of Service variables CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 14 weighted between all the sources Si and provides the maximum throughput LmaX, x, a that destination D can receive from the network.

The module M3 calculates, from the Quality of Service measurements, weighted Quality of Service variables between all the destinations Di and provides the maximum bit rate that the source S can send to the network and the maximum bit rate that the source S can send. to each destination Di.

The module M4 establishes a diagnostic on the quality of service between the source S and each destination Di so as to control the selector to direct all or part of the traffic to another network.

Thanks to this method and to the device according to the invention, the following advantages are obtained:

- dynamic adaptation to variations in the Quality of Service characteristics of transit, incoming access and outgoing access networks;

- dynamic adaptation on various quality criteria such as delay, jitter and losses.

Combinations of criteria are also possible;

- all the users of the networks used do not need to obey the same mechanisms.

This point facilitates the introduction and use of the device in already deployed networks;

- dynamic maximization of throughput within absolute and exact (and not only relative) limits of transit time, jitter and loss;

- dynamic maximization of the quality of service (minimization of delay, jitter and losses) within absolute and exact limits for a given throughput ;

- dynamic optimization of the combination (Quality of Service; throughput) according to criteria adapted to each user flow;

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 - optimized use of several networks, depending on the load presented, the effective quality of service, the prices observed, the quality of service objective , and 5 price objectives (choice of network, choice of type of service offered, etc.).

Other features and advantages of this invention will emerge clearly from the following description, taken by way of non-limiting example, with reference to the attached figures:

FIG. 1 represents a general diagram of a transmission network in which the method according to the invention is implemented;

FIG. 2 represents different possible cases of access networks;

FIG. 3 illustrates a method for calculating the queue delay according to the prior art;

FIG. 4 schematically represents a curve for optimizing the quality of service according to the invention;

FIG. 5 diagrammatically represents a network in which a device according to the invention is implemented;

FIG. 6 represents a particular example of optimization curves according to the invention;

- Figure 7 schematically shows traffic limiters equipping the device according to the invention.

In the remainder of the description, we will call MRTTsd: the average value of the round trip delay for a pair SD of users.

Lmax, s, d: maximum limit value of the flow from a source S to a destination D.

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 16 LmaX, s, X: maximum limit value of the flow rate of a source S, all destinations combined.

Lmax, x, d: maximum limit value of the total flow to a destination D, all sources combined.

Ds, d: actual flow from a source S to destination D.

DX, d: total actual flow to destination D, all sources combined.

Ds, x: total real flow transmitted by the source S, all destinations combined.

Rs, d: rate reservation between source S and destination D.

Method for dynamic optimization of the quality of service in a data transmission network 2 in packet mode, said network comprising a plurality of sources Si and a plurality of destinations Di connected to a transit network 4 via a plurality of networks of access 6, each source being capable of sending a maximum bit rate Lmax, S, x, and each destination being capable of receiving a maximum bit rate Lmax, x, d, said sources each comprising a means of classification 8 and control 10 of the transmitted data rate.

The elements of Quality of Service (throughput, delays, jitter, loss) have their origin in particular in the congestion of the various parts of the network 2.

These phenomena are grouped into 3 categories:

Category Cl: Collective characteristics.

This category includes phenomena due to the joint activity of the sources of the sub-network.

This is particularly, but not only, congestion at outbound access to destinations.

CA 02398366 2002-07-24 WO 01/58094 PCT / FRO1 / 00324 17 Category C2: individual characteristics. 01 groups together in this category the phenomena due to the activity of each source, independently of the others.

This is particularly, but not only, congestion at the access entering the network.

Category C3: the characteristics of the environment.

The transit network 4 is seen as a propagation medium whose characteristics change relatively slowly, and relatively independently of the activity of the sources of the subnetwork.

The proposed method is based on the following points.

- use of precise, numerous and absolute measurements;

- the network considered does not need to be homogeneous;

- segmentation of the delay into different parts, and - control and optimization mechanisms using these segmented measures.

In a preferred embodiment, the method according to the invention comprises the following steps:

a- measure the absolute transfer delay, the jitter, the throughput of said data exchanged and the losses between each source S and each destination D, and perform the following sub-steps as a function of the measurements obtained in this step a:

al- for each destination D, distribute the maximum flow LmzJ ,, x, a between the different sources;

a2- for each destination D, dynamically determine the limit of the congestion point of the CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 18 access network 6 optimizing the maximum throughput Lmax, x, d;

a3- for each source S, dynamically determine from the measurements obtained in step a, the limit of the congestion point of the access network 6 optimizing the maximum overall throughput Lmax, s, x and the maximum throughput towards each destination Lmax , s, d;

a4- for each source S, determine the effect of the congestion of the transit network 4;

and b — controlling the means of classification 8 and of flow control 10 as a function of the results of sub-steps a1, a2, a3 and a4.

The object of a module M1 12 is to limit the traffic to a given source so as to control collective congestion and the sharing of access between the sources of the subnetwork.

For a given destination D, a limit Lmax, x, d is set from.

- the speed of the access line (Lmax, x, d <= speed of the access line);

- module M2 14, which searches for the best current value of Lmax, x, d;

- configuration parameters.

This rate Lmax, x, d must be shared between the active sources of the sub-network.

To do this, the M1 module can use:

an instantaneous local measurement of the flow rates D, s, d for each of the sources;

- dynamic reservations of flow Rs, d made by all or part of the sources;

- static configuration data (for example the speed of the access line, a fixed limit CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 19 by subscription with the operator of the access network, a priori weighting, etc.).

One possible principle consists for the destination D in communicating to each of the sources a first value Lmax, S, a that the source S cannot exceed in order to send information to the destination D (programming of a limiter 10).

Preferably, for a given destination D, the control module M1 12 comprises the following steps:

- set a limit Lmax, x, d from, on the one hand, the speed of the access line, and on the other hand, the module M2 14;

- share the limit Lmax, x, a between the active sources of the sub-network according to the local and instantaneous measurements of the flows Ds, d for each of the sources, the dynamic reservations of the flow Rd made by all or part of the sources, the static data configuration ;

- transmit to each source S the first limit value Lmax, s, a of flow rate to destination D.

If at a given time, or by construction of the access network 6, the collective congestion point to destination D is not the access line of said access network 6, the maximum usable speed is lower than that of the access line to destination D.

In addition, in this case, there is a good chance that this maximum flow rate will be variable over time, for example because it is due to a concentration stage in which other sources not belonging to the subnetwork are involved.

It is therefore necessary to find and dynamically adapt the limit Lmax, x, a to be used by the module CA 02398366 2002-07-24 WO 01/58094 PCT / FROI / 00324 Ml 12. This is the function of module M2. 14, implemented in each of the destinations.

The principle of the discovery of this limit consists in considering the relation:

5 QUALITY OF SERVICE = F (COLLECTIVE LOAD) Too great a load leads to a degradation of the Quality of Service parameters (delay, jitter, losses).

Too low a load leads to poor use of the access network, and therefore to a waste of costly resources.

FIG. 4 gives the typical appearance of such a function.

We identify 3 zones Zone 1. the quality of service is relatively independent of the bit rate; access is not congested and is probably underutilized.

Zone 2: the quality of service begins to deteriorate significantly with the speed; access is at the limit of congestion.

20 Zone 3 the quality of service deteriorates sharply for a small increase in throughput.

Access is congested.

The ideal point of equilibrium is in zone 2 high access speed and controlled quality of service.

The M2 14 module periodically performs:

a calculation of a weighted value between the sources for each of the different quality of service parameters (weighted delay, weighted jitter, weighted losses);

a determination of the curve of the weighted values ​​as a function of the measured load Dx, d.

This search is made from instantaneous measurements, and may require varying the limit Lmax, x, d around a temporary operating point, to obtain better visibility of the functions.

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 21 - a setting of a limit Lmax, x, a according to the curves and key configurable thresholds, in particular for the quality of se ~ --vice (by example maximum delay, or maximum loss).

If at a given moment, the first authorized flow Lmax, s, a between the source S and the destination D leads to a degradation of the flows coming from this source (for example because the access network 6 from the source S to the transit network 4 becomes congested), this source must limit its throughput otherwise the quality will deteriorate.

It is therefore necessary to find and dynamically adapt the limit for the bit rate Ds, d between the source S and the destination D. This is the function of the module M3 16, implemented in each of the sources.

A method that can be used for the M3 16 module is similar to that described for the M2 14 module.

Throughput limit values ​​are determined towards each destination Lmax, s, a, within limits possibly given by dynamic throughput reservations Rs, d made to all or part of the destinations, static configuration data and the throughput of the line d 'access.

These flow rate limit values ​​are determined from the curve of the Quality of Service values ​​towards each destination as a function of the load measured overall Ds, x and towards each destination Ds, a.

An M5 module 17 controls the classification means 8 and the flow control means 10 by dynamically setting the value of limiters 10 of the source S as a function of the values ​​coming from the modules M1 12 and M3 16.

FIG. 5 illustrates an example of a network in which the method is implemented.

CA 02398366 2002-07-24 WO 01/58094 PCT / FROl / 00324 22 Figure 6 gives an example of determining the limit of the flow rate Lmax, s, a optimum as a function of the relative positioning of the limit Lz, max, s, d determined by module M3 16 and the limit Ll, max, s, d authorized by module M1 12.

The choice of the optimum bit rate depends on the quality of service objectives for the flows considered.

The method can also take into account an overall criterion on transmission from the source, by calculating a weighted function of the Quality of Service parameters for all the flows on transmission (in fact, if the individual congestion takes place in the access network 6 to the transit network 4, all the flows emitted by the source participate in it).

In this case, a search is first made for what is the best value for the limit of the overall transmission rate Lmax, s, x, which will lead to programming a global limiter 10 on transmission, then a search for the best is carried out. flow rate to each destination D Lmax, s, a, which will lead to programming an individual limiter 10 for the flows to each destination.

FIG. 5 illustrates the optimization method according to the invention in a data stream transmission network in packet mode.

The device intended to implement this method comprises:

means 20 for measuring the absolute transfer delay, the jitter, the data rate exchanged and the losses between each source S and each destination D;

- Means for distributing, for each destination D, the maximum bit rate L, ax, x, a between the different sources;

CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 23 - means for dynamically determining, for each destination D, the limit of the congestion point of the access network 6 optimizing the maximum throughput Lmax, x, d ;

means for dynamically determining, for each source S, the limit of the congestion point of the access network 6 globally and towards each destination;

means for determining, for each source S, the effect of the congestion of the transit network 4;

- means of classifying the data packets sent according to their destination and the quality of service required;

- means of controlling the flow;

means for controlling said classification means and said control means; and - network selection means between each source S and each destination D.

FIG. 7 describes an arrangement of flow control means 10 constituted by global limiters and limiters by destination at the output of each source.

The transit network 4 also has an impact on the quality of service of the flows between sources and destinations.

By assumption, this quality is not correlated with the bit rate of the sources.

This impact varies over time (for example depending on the user activity cycle: day / night, peak hours, etc.), at a slow rate relative to that of collective and individual congestion phenomena.

It is important to know the Quality of Service intrinsic to the transit network 4, in particular to determine the best overall Quality of Service CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 24 that can be achieved from Source to Destination, and to refine the results of modules M1 12, M2 14 and M3 16.

Knowledge of this quality of service is provided by a module M4 22, which establishes a diagnosis of the quality of service between the source S and each destination Di so as to control the selector 24 to direct all or part of the traffic to another network. .

As the final quality of service cannot exceed that of the transit network, we have the following cases, with respect to a given quality objective:

1.

Quality of the transit sufficiently superior to the objective: the optimization modules M1 12, M2 14 and M3 16 are effective.

2.

Transit quality inferior or too close to the objective: the optimization modules M1 12, M2 14 and M3 16 are not sufficient to achieve this objective.

Other arrangements may therefore have to be made.

In the event that the quantity of traffic to be handled, the optimization methods, or even the specific quality of the transit network 4, do not make it possible to obtain a sufficient quality of service, other networks 26 could be implemented. communication e:, be sources and destinations by controlling selector 24, for example:

- the use on the same access link of a better quality service at a higher price;

- access to another service operator (instead of or in addition to the first);

- a direct link (for example through the telephone network).

The precise knowledge of the quality that it is possible to obtain thus makes it possible to manage the compromises CA 02398366 2002-07-24 WO 01/58094 PCT / FR01 / 00324 quality / price according to criteria specific to the users.