FR2830149A1 - Digital switching network digital word transmission bandwidth allocation having input module stream numbers determination and input/output coupler weighting with global weighting arbitration determining bandwidth split. - Google Patents

Abstract

The block transfer bandwidth separation process has an input module (ITMi) determination of the number of active streams for each output module (OTMj) destination crossing the switching network. There is a step of determining the weighting from each input/output couple, followed by a global weighting arbitration function. Finally, the bandwidth is assigned based on the previous steps.

Description

<Desc / Clms Page number 1>

The invention relates to a method for sharing the available switching bandwidth for transferring data blocks across the switching network of an asynchronous switching node. The invention relates to a method for sharing the available switching bandwidth for transferring data. , grouped in the form of data blocks, through a switching network which interconnects the input terminal modules and the output terminal modules of an asynchronous switching node, in a telecommunications system. The various input and output terminal modules will simply be referred to as input modules and output modules in the remainder of this document.

This switching node is supposed to process flows of data blocks corresponding to certain services without full guarantee of quality of service, that is to say services without any guarantee of quality of service (in English: "best-effort services" "), or services with a limited quality of service guarantee (for which bandwidth capacity has been allocated by prior reservation without this being sufficient to support them in full). In this case, such a flow risks entering into competition, at least for part of its traffic, with other flows of this type to access the same destination output module, and of creating a congestion situation. In order to regulate the transfer rates of the data blocks of such flows to each output module in such a node, various known methods of flow control can be used.

One of these known methods is based on a method making it possible to carry out a weighted sharing dynamically and tending to be equitable of the bandwidth globally available for each subset of the different streams of active data blocks whose data blocks must be transferred to the same destination output module given from the different input modules, and this for each output module. Such a method of sharing available bandwidth makes it possible to dynamically allocate, to each of the different input modules, a fraction of the bandwidth available to access each output module, and thus ensuring a high level of performance in terms of traffic flow to each output module through the switching network of the asynchronous switching node.

A dynamic bandwidth sharing method of this type is described in an article entitled "Buffering and Flow Control for Statistical Multiplexing in an ATM Switch" by T. WORSTER & al which was published in April 1995 in volume n01 pages 273-277 of the report of the Switching Symposium 155 95.

<Desc / Clms Page number 2>

 According to the method described, in order to carry out a weighted sharing dynamically and aiming to be equitable in the globally available switching bandwidth at a node, provision is made for: - an evaluation of the volume of traffic on hold in temporarily stored in input queues that each input module comprises, for each of the output modules likely to be addressed to it; a determination, as a function of the volume of traffic on standby observed, of a so-called required bandwidth parameter, for each pair formed by a given input module and a given destination output module; a collection of these required bandwidth parameters relating to each destination output module, from the various input modules; - sharing aimed at an equitable distribution of the switching bandwidth available to each destination output module, as a function of these different required bandwidth parameters relating to the different input modules; - the supply to each input module of a shared bandwidth parameter, allocated to each of the pairs of modules that it forms with the various output modules of the node.

The implementation of such a method at the level of an internal flow control mechanism of an asynchronous switching node makes it possible to share the available switching bandwidth, from the values of determined required bandwidth parameter values as a function of the volume of waiting traffic observed at the queues of the node's input modules.

The band sharing thus achieved has the effect of favoring the most active flows, that is to say having a larger volume of data on hold, but it does not make it possible to take into account different levels of quality of service. for the respective flows of data blocks. Thus, the transfer capacity offered to a flow requiring a high quality of service can therefore be penalized by the presence of at least one other flow requiring a lower quality of service, when the latter is more active than it. Such a method of sharing available bandwidth is therefore fair with regard to the volumes of traffic waiting in the various respective input modules, and it may be suitable for arbitrating the traffic of data block flows which are homogeneous in term of quality of service. On the other hand, it may not be fair in the case of data block flows requiring different levels of quality of service characterized by different categories of respective relative priority.

<Desc / Clms Page number 3>

To overcome this drawback, the invention therefore proposes another type of sharing method, weighted and aiming to be more equitable with regard to the quality of service, the switching bandwidth available for transferring data blocks, for example by cells or packets, through a switching network of an asynchronous switching node which interconnects input modules, provided with buffer memories allowing temporary storage of the data blocks, and output modules. Each data block, to be transferred, belongs to a flow of data blocks to which are assigned, at the level of an input module, at least two associated parameters: one of these parameters specifies the module receiving the flow, l the other is a relative priority weight parameter of the stream. An internal flow control mechanism dynamically ensures a balanced and tending to be equitable sharing of the bandwidth generally available for the different flows of data blocks, for transferring them from the input modules to the different destination output modules through the node switching network.

According to a characteristic of the invention, the method provides: - a classification of the different streams of data blocks by relative priority category, as a function of their respective relative priority weight parameters, a predetermined category weight parameter value being assigned to each relative priority category; - a supply, to each input module and by the internal flow control mechanism, of shared bandwidth parameter values, each parameter being relative to one of the pairs of modules composed of an input module and d a destination output module, these shared bandwidth parameter values allocated by pair of modules being obtained by taking into account the value of the category weight parameter assigned to each stream, and by performing a succession of steps comprising: a first step, at the level of each input module, of determining the number of active flows for each recipient output module and for each category of relative priority, an active flow being a flow for which there is at least one stored data block ready to be transferred across the switching network; a second step of determining, for each input module, a relative priority weight parameter value per pair, for each pair of modules, this parameter value being obtained for a given pair of modules by performing the sum , for all priority categories

<Desc / Clms Page number 4>

 relative, of the products which are respectively calculated, category by category, by multiplying the value of the category weight parameter corresponding to one of the categories of relative priority, by the number of active flows determined for this category at the level of the input module of this torque and for the output module of this torque; a third step of determining, for each output module, a global relative priority weight parameter value, this parameter value being obtained by performing the sum of the relative priority weight parameter values per couple respectively calculated for each of the pairs of modules, called couples concerned, in which the output module considered is addressed; a fourth step of sharing, for each output module, the bandwidth available to access this output module, from the various input modules of said pairs of modules concerned, in which this output module is intended , this weighted and equitable sharing being carried out in proportion to the relative priority weight parameter values per couple calculated for the couples concerned, the results of this sharing for each output module making it possible to determine for each input module said values of shared bandwidth parameter thus allocated to each of the pairs of modules for the transfer of the data blocks of the set of data block flows relating to this pair of modules.

Thus, the value of the bandwidth parameters allocated to each of the pairs of input and output modules, for transferring data blocks belonging to the flow group relating to this pair, results from a weighted sharing of the width of band available to access the torque output module. This sharing calculation is performed, for each module couple relating to a given output module, by dividing its available bandwidth according to a weighting proportional to the relative priority weight parameter values per couple calculated for all the couples of modules concerned, which therefore allows this sharing to be fair with regard to the different levels of quality of service relating to the different flows intended for this output module.

According to another characteristic of the invention, it is further provided that, when sharing the available bandwidth of a given output module according to said fourth step, the calculation of the shared bandwidth allocated to each pair of modules take into account a maximum shared bandwidth value which constitutes an upper limit value of the value of this bandwidth

<Desc / Clms Page number 5>

 shared allocated to each pair, this upper limit value being obtained as a function of values of a desired bandwidth parameter, which are determined by the different input modules for each of the pairs of modules relating to a given output module, by evaluating the volume of data blocks awaiting transfer for all of the flows active in this input module, which are intended for the output module of this pair.

Thus, also taking into account a maximum shared bandwidth value to determine the shared bandwidth allocated to a pair of modules, a weighted sharing with a higher value makes it possible to avoid this bandwidth being allocated to a couple of modules, excessively compared to the desired bandwidth for this couple which is sufficient to sell the volume of data blocks awaiting transfer for the flows of data blocks intended for the output module of this couple. This makes it possible to carry out a balanced and equitable sharing with respect to the different levels of quality of service relating to the different streams, while achieving greater efficiency in using the bandwidth available to access each output module.

The invention, its characteristics and its advantages are explained in the following description in conjunction with the figures listed below.

FIG. 1 presents a schematic diagram relating to an asynchronous node allowing the implementation of the sharing method according to the invention.

FIG. 2 illustrates the groups of flows coming from the various input modules, which are intended for a given output module, in a switching node as illustrated in FIG. 1.

FIG. 3 presents a block diagram relating to an input module of a switching node as illustrated in FIGS. 1 and 2.

The list below gives the terminology of the abbreviations used, with their English translation: AB Available bandwidth
Available bandwidth AF Arbitration function
Arbiter function GF Allocation factor
Grant factor CW Category weight
Class weight DBV Volume of data blocks

<Desc / Clms Page number 6>

Need for bandwidth OTM Module de sortie
Output termination module Q File d'attente
Queue RAB Largeur de bande disponible restante
Remaining available bandwidth RGRPW Poids de priorité relative gobal restant
Remaining global relative priority weight RPC Catégorie de priorité relative
Relative priority class RPW Poids de priorité relative
Relative priority weight RS Sélection par rang
Rank selection SBG Largeur de bande partagée EXB volume data block Excess bandwidth
Excess bandwidth FG Stream group
Flow group FSG Flow subgroup
Flow sub-group GTI Allocation type indicator
Grant type indicator GUEB Overall unallocated excess bandwidth
Global unatiocated excess bandwidth CRPW Relative priority weight per couple
Torque relative priority weight ITM Input module
MSG input termination module Maximum shared bandwidth
Maximum shared banswidth NAF Number of active streams
Number of active flows NFB Desired bandwidth

Need for bandwidth OTM Output module
Output termination module Q Queue
RAB tail Remaining available bandwidth
Remaining available bandwidth RGRPW Remaining global priority weight
Remaining global relative priority weight RPC Relative priority category
Relative priority class RPW Relative priority weight
Relative priority weight RS Selection by rank
Rank selection SBG Shared bandwidth

<Desc / Clms Page number 7>

Shared bandwidth grant GRPW Overall relative priority weight
Global relative priority weight UEB Unallocated excess bandwidth Unallocated excess bandwidth L Sum [of NFB values]
Sum [of NFB values].

In addition, the following index ratings are generally used: - The index 'i' designates any input module, such as ITMi; - The index'j 'designates any output module, such as OTMj; - The index '' denotes a considered or given output module, such as OTMd; - The index 'q' denotes any weight of relative priority assigned to a flow, such as RPWq; - The index 'k' designates any category of relative priority, such as RPCk The sharing method according to the invention is intended to be implemented at the level of an asynchronous switching node 1 in a telecommunications system.

This node 1 comprises a plurality of n1 input modules, such as tTM1, tTMi, ITMnl in FIG. 1, which are interconnected with a plurality of n2 output modules, such as OTM1, OTMj, OTMn2, via d 'an asynchronous switching network 2, according to a conventional arrangement for those skilled in the art.

Such a separate representation of n1 input modules and n2 output modules can correspond either to an embodiment of physically separate input and output modules, or to a logical separation of two types of functions, input and output. , realized in mixed terminal modules grouping these two types of functions, in which case the numbers of modules n 1 and n2 are generally equal. The asynchronous switching network 2 ensures the transfer of a data block transmitted by a given input module and intended for a given output module, according to any internal routing mode, for example of the connectionless type or with connection.

In this switching node, each data block belongs to a flow of data blocks, which corresponds to a communication service through a plurality of nodes of the telecommunications system. In a node, all

<Desc / Clms Page number 8>

 data blocks belonging to the same flow are transferred from an input module to a destination output module. Each input module can process a plurality of data block flows intended for different output modules.

We consider here data block flows corresponding to services without full guarantee of quality of service. In this case, such a flow, when it is active, risks entering into competition, at least for part of its traffic, with other flows of this type, simultaneously active, to access the same destination output module. , and create a congestion situation. A flow is considered as active, in an input module, when there is at least one data block, belonging to this flow, waiting in this module to be transferred to the output module recipient of this flow.

The object of the method according to the invention is to dynamically carry out, at a given time, a balanced and tending to be equitable sharing of the available switching bandwidth, at the switching network 2 of node 1, between the different flows then active which are intended for the same given output module. It is necessary to regulate the data transfer rates of the blocks of such active flows to each output module in this node, insofar as the capacity for transferring data blocks of services without full guarantee of quality of service through the switching network is physically limited by characteristics specific to this network. Also, for a given OTMd output module, its available bandwidth ABd for services without full guarantee of quality of service, is at least limited by the physical capacity of the access interface of the switching network to this module; moreover, it may even be less than this physical capacity if part of it has been reserved by pre-allocation to services with full guarantee of quality of service.

In each input module ITMi, each subset of the active flows which are intended for the same output module OTMd, constitutes a group of active FGid flows.

Such a flow group FGij therefore relates to a pair of modules formed by an input module ITMi and an output module OTMj, designated ITMi / OTMj. FIG. 2 illustrates such groups of active flows FG 1 d, ... FGid, ... FGn2d of data blocks to be transmitted to the same destination output module OTMd, from different input modules tTMi, .. . tTMi, ... tTMnt.

Each input module, such as ITMi in FIG. 3, can be considered as comprising a buffer memory 3 structured in a plurality of queues Q in which are temporarily stored the data blocks to be transmitted to their respective destination output module . In an ITMi input module, after

<Desc / Clms Page number 9>

 processing by known input interface functions (not shown in FIG. 3), the received data blocks are distributed to the different queues, such as Q1, ..., Qi, ..., Qn2. Each queue Qj stores the data blocks relating to the various flows which are intended for the same output module OTMj, that is to say all the flows belonging to the same group of flows FGij. Thus, according to a known principle, all the data blocks of the flow group FGij, intended for the same output module OTMj, are temporarily stored in the queue Qj corresponding to this flow group for this recipient module. Then all the traffic of data blocks of an input module, intended for the different output modules, is stored in the different queues Q1, ..., Qi, ..., Qn2 respective by module recipient output OTM1, ..., OTMj, ..., OTMn2. To remove a data block (or a data block segment) from a queue and supply it to the switching network, a multi-server queue 4 uses one of these service cycles to read and retrieve the corresponding data from the queue. According to a known technique, such a multi-queue server distributes its service cycles to the different queues, in proportion to the data output rate authorized for each queue Qj, therefore in proportion to the bandwidth allocated to this queue Qj to transfer the data blocks intended for the corresponding OTMj output module. This allocated bandwidth corresponds at least to the shared bandwidth SBGij which is dynamically allocated to a pair of ITMi / OTMj modules, by the flow control mechanism of the invention; this allocated bandwidth may possibly also include a preallocated bandwidth which has been reserved for services with guaranteed quality of service. Thus, with regard to the available bandwidth ABd of a given output module OTMd, the homologous queues Qd, relating to this same destination output module, in the various input modules ITMi of the node, will be served at a data rate corresponding to the shared bandwidth SBGij which is respectively allocated to them as a fraction of this available bandwidth, so that the overall data rate transmitted, through the switching network 2, to this module given output does not exceed the capacity of this available bandwidth.

The operation of all the queues of an input module is controlled by a queue management unit 5, here assumed to be assigned to this input module, as shown diagrammatically in FIG. 3.

<Desc / Clms Page number 10>

 The constitution of the output modules of node 1 is not detailed here insofar as each output module is capable of being produced conventionally in this field, with a plurality of registers or memories making it possible to receive the blocks of data for each of the flows to which it is addressed at the node level. The blocks thus received are then transmitted to the output interface functions of this module.

The various control functions required to ensure the operation of the switching node 1 are of any type and are considered to be known. The various corresponding means are not developed here insofar as they are known to those skilled in the art. The control functions are therefore not shown in the figures, with the exception of the queue management unit 5 of an input module, indicated in FIG. 3, which includes part of the characteristic means exploited for the implementation of the sharing method according to the invention, here considered as constituting a specific flow control mechanism which is specific to node 1. This queue management unit 5 is provided with material means, in particular processors and memories, as well as suitable software means, in particular, to allow the implementation of the sharing method according to the invention.

To allow the implementation of the available bandwidth sharing method according to the invention, at least two parameters are provided to characterize each flow of data blocks at an ITMi input module.

A first parameter is provided to designate the output module OTMj recipient, at the node, of the stream then considered. A second parameter RPWq corresponds to a relative priority weight, assigned to each flow as a function of one or more criteria, at least one of these criteria being relative to the quality of service requested for the data transmission that is ensured the flow considered. This relative priority weight parameter RPWq is intended to be taken into account during the available bandwidth sharing operations in order to improve its fairness.

The implementation of the method according to the invention more particularly involves classifying, in each ITMi input module, the different flows of data blocks into groups of FGid flows having for the same given output module OTMd, this classification is systematically carried out in each input module ITMi for the various output modules OTMj of the node. According to this classification, each of the groups of flows FGij therefore brings together all of the flows of blocks of

<Desc / Clms Page number 11>

 data transferred between a couple of modules formed by any ITMi input module and any OTMj output module, such as ITMi / OTMj couple.

In addition, in each input module ITMi, within each group of flows FGid relating to the same output module OTMd given, the flows are classified, according to their respective relative priority weights RPWq, into several subgroups FSGidk flows by relative priority category RPCk. In the switching node, a predetermined category weight parameter value CWk is assigned to each relative priority weight category RPCk. In the general case where the relative priority weight values RPWq assigned to the streams do not coincide with the predetermined category weight values CWk for the different possible relative priority weight categories RPCk, pre-established conversion rules are used for determining which category of relative priority weight RPCk should be assigned a flow as a function of the relative priority weight RPWq which is assigned to it; an example of a conversion rule is that a given relative priority weight category RPCk, and therefore its corresponding predetermined category weight CWk, are assigned to any stream having a relative priority weight RPWq whose value is within an interval of values [RPWk / min; RPWk / max] predetermined for the relative priority category RPCk. In a particularly simple example where any possible value of relative priority weight RPWq assigned to a stream always coincides with one of the possible values of weight of category CWk predetermined in the switching node for the different categories of relative priority weight RPCk, the weight parameters RPWq and CWk then constitute, in this particular case, one and the same parameter, and the flows are classified, within each group of flows FGij relating to the same output module OTMj into subgroups of flows FSGijq directly according to their own relative priority weight values RPWq.

The internal flow control mechanism ensures the supply of parameters of shared bandwidth SBGij allocated to each of the flow groups FGij of an input module ITMi having a same output module OTMj as recipient.

Such a parameter SBGij is therefore relative to each of the pairs of input and output modules, such ITMi / OTMj, comprising this input module, such ITMi, and an output module, such OTMj, which is intended for the blocks FGij flow group data from this input module.

The determination of the shared bandwidth parameter values is carried out taking into account the value of the category weight parameter CWk

<Desc / Clms Page number 12>

 assigned to each stream. It is obtained following a succession of processing steps.

As part of a first step, it is planned to determine at a given instant, at the level of each input module ITMi of the node, the number NAFijk of streams which are then active for each subgroup of streams FSGijk. In the present description, a stream of data blocks is considered to be active as long as there is, for this stream, at least one block of data waiting in an input module, more precisely in the queue Qi relating to the OTMj output module concerned. This determination of the number of active flows for per FSGijk flow subgroup in an ITMi input module is therefore carried out selectively, taking into account the two parameters assigned to each flow, namely its recipient output module OTMj and its category. of relative priority RPCk.

As part of a second step, provision is made to determine, for each input module ITMi of the node, a value of relative priority weight parameter by pair CRPWij, for each of the flow groups FGij corresponding to the pairs of modules ITMi / OTMj input and output. This value of relative priority weight parameter by couple CRPWij is obtained by performing the sum, in the group of flows FGij corresponding and for all the categories of relative priority RPCk, of the products respectively established for each of these categories, c ' that is to say in each FSGijk stream subgroup by category, by multiplying the value of the corresponding category weight parameter CWk by the number of active NAFijk streams determined in the first step for this category. The calculation of the value of each relative priority weight parameter by pair CRPWij is therefore the sum of the products of parameter values CWk and NAFijk carried out for each subgroup of flows FSGijk relating to a relative priority category RPCk, ie for example for 8 possible relative priority categories RPC1 ... RPC8:
CRPWij = CW1 x NAFijl + ... + CWk x NAFijk + ... + CW8 x NAFij8 The results, obtained by implementing the second stage, are used within the framework of a third stage during which it is planned to determine, for each output module OTMj, a weight parameter value of overall relative priority GRPWj. The calculation of this overall relative priority weight parameter for each given OTMd output module involves the prior grouping together of the values of relative priority weight parameter values per pair relating to this output module considered OTMd and relating to the various output modules. ITMi input. The value of this global relative priority weight parameter GRPWd is determined by a function which performs, for at least one output module OTMd

<Desc / Clms Page number 13>

 given, the sum of the relative priority weight values CRPWid, respectively calculated within the framework of the second step, for each of the pairs of modules ITMi / OTMd in which the output module OTMd considered is destination. The means used to perform this function are hereinafter called the arbitration function of the flow control mechanism. This calculation of the value of the parameter GRPWd is carried out for each of the output modules OTMj; it therefore concerns all the pairs of input and output modules ITMi / OTMj at the node.

The method according to the invention provides a fourth step for calculating the sharing of the available bandwidth ABi for accessing each destination output module OTMj making it possible to determine the shared bandwidth SBGij allocated to each input module ITMi for accessing this output module. This calculation is performed, for each pair of ITMi / OTMj modules, taking into account the parameter values obtained during the three preceding steps.

The available bandwidth ABd, which can be used to access a given OTMd output module, is then shared between all the ITMi input modules of the pairs of ITMi / OTMd modules, referred to as the couples concerned, where this OTMd module is included in as output module recipient of at least one active flow.

For each of the output modules OTMj, the weighted and equitable sharing of its available switching bandwidth ABj is thus carried out for the input modules ITMi of the couples concerned, in proportion to the relative priority weight parameter values by torque CRPWij calculated. for these couples concerned, relative to the value of the global relative priority weight parameter GRPWj calculated for this output module OTMj. The available switching bandwidth ABj of an output module OTMj is therefore divided, for the various input modules ITMi, according to a weighting proportional to the values of relative priority weight parameter by torque CRPWij.

Such sharing is performed for each of the OTMj output modules; it therefore concerns all the pairs of input and output modules ITMi / OTMj at the node. The result of these sharing processes for the various output modules OTMj of the node is reflected in the supply of individualized values of shared bandwidth parameter SBGij allocated to each of the pairs of modules whose input modules ITMi share the same available bandwidth ABj to access the same output module OTMj.

In each ITMi input module, the queue management unit 5, shown in FIG. 3, constitutes the main interface level between the

<Desc / Clms Page number 14>

 flow control and the classic node functions, through which input and output data for the sharing process of the invention are exchanged dynamically, namely, for each pair of ITMi / OTMj modules: - on the one hand, as input data, the numbers of active flows NAFijk by category of relative priority RPCk for this pair (the weights of category CWk by category being supposed to be predetermined and known by the means of flow control mechanism), - and on the other hand , as output data, the shared bandwidth SBGij allocated to this pair, which is used to determine the data output rate authorized to serve the queue Qi relative to the output module OTMj in the input module ITMi.

The other input data of the sharing method, namely the available bandwidth ABi for each output module OTMj, are supplied to the arbitration function which performs the sharing for the corresponding output module.

In a preferred form of implementation, the value of the shared bandwidth parameter SBGij allocated to an input module ITMi for accessing the output module OTMj of a given pair of modules ITMi / OTMj, is derived from a calculation. According to this calculation, the value of the shared bandwidth parameter SBGij, which is allocated to the flow group FGij corresponding to a pair of modules ITMi / OTMi, is proportional on the one hand to the value of the available bandwidth ABj for access the output module OTMj of this torque, and secondly the value of the relative priority weight parameter CRPWij, calculated for this torque, and inversely proportional to the value of the overall relative priority weight parameter GRPWj, calculated for the OTMj output module considered.

According to an example of the simplest formulation of this weighted sharing, the value of the shared width parameter SBGij, allocated to the group of active FGij flows relating to a pair of modules ITMi / OTMj, is calculated by:
SBGij = CRPWij x {ABj / GRPWj} Different embodiments are possible for the practical implementation of the four-step process defined above. To obtain a dynamic flow control mechanism, it is possible to execute its process of sharing available bandwidth according to different scenarios: for example by output module, by groups of output modules, or for all the output modules of the switching node, and this periodically, or alternatively for certain modules whose traffic loads change significantly.

<Desc / CRUD Page number 15>

On the other hand, as regards the means which ensure the execution of the various stages defined above, their implementation is conceivable in a more or less centralized manner, or else in a distributed manner in the terminal modules of the node. It is natural to install at least the means executing the first step defined above, in each input module which locally supervises the activity of the flows of data blocks, for example in the queue management unit. 5 indicated in FIG. 3. On the other hand, at least the third step must be carried out for each output module by an arbitration function of the flow control mechanism, and the corresponding means can be distributed in each output module, or much more or less centralized by group of output modules. This arbitration function, not shown in the figures, is provided with hardware means, in particular with processors and memories, as well as suitable software means, in particular, to allow the implementation of the sharing method according to the invention.

According to the invention, two possible implementation variants are provided according to the type of parameters which are provided by the input modules to the arbitration functions to allow the grouping necessary for the execution of the third step defined above. , namely the grouping of all of the relative priority weight parameter values by couple CRPWid relative to each given OTMd output module.

According to a first variant, each input module ITMi itself performs, during the second step defined above, the calculation of the value of the relative priority weight parameter by pair CRPWij, for each of the pairs ITMi / OTMj of which it is part. These relative priority weight parameter values by couple CRPWij are then communicated to the arbitration functions of the flow control mechanism. An AF arbitration function then performs, at least for a given output module OTMd given from the different ITMi / OTMd couples concerned, a simple collection of the relative priority weight parameter values by couple CRPWid concerning this output module, which are transmitted by the input modules. This arbitration function thus ensures their grouping for all of the ITMi input modules of the node.

According to the second variant envisaged, the value of the weight parameter of category CWk predetermined for each category of relative priority RPCk, is chosen to be identical for all the input modules. After having carried out the first step, each ITMI input module then supplies, to the arbitration functions of the flow control mechanism, the numbers of active NAFijk flows determined by category of

<Desc / Clms Page number 16>

 relative priority RPCk, for each recipient output module OTMj of one of the pairs JTMi / OTMj to which it belongs. An arbitration function AF ′ of this mechanism performs, at least for a recipient output module OTMd given from the various ITMi / OTMd couples concerned, a collection of the numbers of active flows NAFidk relating to this output module, which are transmitted by the modules input.

Then, this arbitration function performs the second step defined above of determining the relative priority weight parameter values by couple CRPWid, which allows it then to ensure the grouping of these CRPWid values for all the modules d ITMi input of the node.

According to the invention, two solutions are also provided concerning the fourth step defined above, which differs by their choice as regards the type of parameters transmitted back to the input modules by the arbitration functions of the flow control mechanism. , following the sendings made by these modules to these arbitration functions.

According to a first of these two solutions, an arbitration function AF "of the flow control mechanism ensures, for at least one given output module OTMd, the completion of the calculation of the sharing of its available bandwidth ABd for the groups FGid flow of each ITMi input module destined for this output module. This AF arbitration function "determines, for each couple TMi / OTMd of this output module, the value of the shared bandwidth parameter SBGid allocated to each ITMi input module of these couples, and it then transmits, to the input module of each of these couples concerned, the shared bandwidth parameter SBGid allocated to this couple.

de sortie. Cette fonction d'arbitrage AF"catcute e ! ! e-même ! a va ! eur de la largeur de bande partagée SBGid allouée à chaque couple concerné TMi/OTMd ayant un module de sortie OTMd donné comme destinataire, c'est-à-dire à chaque groupe de flux FGid correspondant, de manière que les valeurs de paramètre de largeur de To completely perform this calculation of the available bandwidth sharing, this AF arbitration function "performs beforehand, for at least one given OTMd output module, the grouping of relative priority weight parameter values by couple CRPWid, which allows, according to the third step defined above, to determine a global relative priority weight parameter value GRPWd for this output module. Then it performs, according to the fourth step defined above, the weighted and equitable sharing of the width of ABd band available to access this given OTMd output module, taking into account the relative priority weight parameter values by couple CRPWid and global GRPWd relative to this module

Release. This AF arbitration function "catcutes e!! Itself! Has! The value of the shared bandwidth SBGid allocated to each couple concerned TMi / OTMd having an output module OTMd given as recipient, that is to say tell each group of corresponding FGid streams, so that the width parameter values of

<Desc / Clms Page number 17>

 shared band SBGid calculated, be transmitted to the ITMi input modules these couples.

According to the second of the two solutions considered above, an arbitration function AF ′ ″ of the flow control mechanism performs only part of the calculation of the sharing of available bandwidth ABd, for at least one output module OTMd given, for the FGid stream groups of each ITMi input module intended for this output module, in the sense that it does not calculate the shared bandwidth values SBGid itself according to the fourth step defined above In the case of this second solution, the arbitration function AF '"then transmits, to all the input modules ITMi, common sharing data relating to a given output module OTMd, which are identical for all the modules input. These common sharing data relate on the one hand to the bandwidth ABd available to access this output module, and on the other hand to the overall relative priority weight parameter GRPWd, the arbitration function of which has calculated the value for this output module according to the third step defined above.

Each input module ITMi then itself completes the calculation, according to the fourth step defined above, of determining the shared bandwidth SBGid which is allocated to it, for each group of flows FGid of this input module ITMi to destination of this OTMd output module.

To perform this partial calculation of the available bandwidth sharing, this AF "'arbitration function of the flow control mechanism ensures, as in the previous solution, for at least one given OTMd output module, the grouping of relative priority weight parameter values by pair CRPWid, this grouping enabling it to determine, according to the third step defined above, a global relative priority weight parameter value GRPWd for this given output module.

In a first variant according to the invention, for each given OTMd output module for which it provides sharing of available bandwidth, this arbitration function AF "'of the flow control mechanism transmits, to all the input modules , common sharing data which is the value of the global relative priority weight parameter GRPWd which it has calculated, and the value of the available bandwidth ABd which has been considered to be exploitable for accessing this output module.

In a second variant according to the invention, for each given OTMd output module for which it provides sharing of available bandwidth, this function AF "′ of arbitration of the flow control mechanism transmits, to all the modules

<Desc / Clms Page number 18>

 input, common sharing data formulated by the value of a sharing rate parameter obtained by dividing the available bandwidth ABd, to access this output module, by the value of the global parameter GRPWd relative to this same output module.

In one or other of these variants, each ITMi input module itself completes the calculation of the weighted and equitable sharing of bandwidth, according to the fourth step defined above, for the group of flows FGid bound for a given OTMd output module, taking into account: on the one hand, the common sharing data transmitted by this output module, which provide the ratio obtained by dividing the available bandwidth ABd by the value of the priority weight parameter relative global
GRPWd, this ratio being either provided directly by the sharing rate received in the case of the second variant, or obtained by calculation from the values of the two parameters transmitted in the case of the first variant; on the other hand, the value of the relative priority weight parameter by pair of modules CRPWid, which is specific to this input module ITMi for the corresponding flow group FGid, and which has been calculated by this input module according to the second step defined above.

These elements allow each input module ITMi to calculate the shared bandwidth parameter value SBGij which is allocated to it, for each pair of modules that it forms with an output module OTMj, that is to say for each of the flow groups FGij, pairs of corresponding ITMi / OTMj modules.

In the case of certain traffic flow configurations coming from the input modules and intended for a given OTMd output module, the result of a sharing carried out as described so far, that is to say strictly by weighting proportional to the relative priority weight parameter values per couple CRPWid, may be such that, for certain pairs of ITMx / OTMd modules, the shared bandwidth SBGxd allocated is greater than the bandwidth which would be sufficient to carry the traffic in waiting in the corresponding ITMx input modules, whereas, for other pairs of ITMy / OTMd modules, the allocated shared bandwidth SBGyd is less than the bandwidth which would be sufficient to carry the waiting traffic in the modules corresponding ITMy input. Although it is fair with regard to the quality of service requested for the respective data block flows, such a sharing result is not optimal, insofar as it does not provide the best possible use of the width of available band ABd to access this destination output module, if

<Desc / Clms Page number 19>

 at least one pair of ITMx / OTMd modules exists, for which the allocated shared bandwidth SBGxd is greater than the bandwidth which would be sufficient to carry the traffic waiting in the corresponding ITMx input modules. In order to remedy this drawback, additional means are introduced in the remainder of the description, which allow weighted sharing to be carried out while achieving better efficiency of use of this available bandwidth ABd.

According to the invention, it is also provided that, during the balanced and equitable sharing of the available bandwidth ABd of a given output module OTMd, the calculation of the shared bandwidth parameter SBGid allocated to each pair of modules ITMi / OTMd concerned, according to the fourth step defined above, also takes into account the value of a parameter of maximum shared bandwidth MSBid which constitutes an upper limit of this shared bandwidth allocated to each of these couples concerned.

The value of this maximum shared bandwidth parameter MSBid is established as a function of values of a desired bandwidth parameter NFBid, which are determined by the input modules ITMi respectively for each of the pairs ITMi / OTMd relating to a same recipient OTMd output module given. The value of this desired bandwidth parameter NFBid is obtained by evaluation, in each ITMi input module, of the volume of DBVid data blocks awaiting transfer to the OTMd output module of this ITMi / OTMd pair, therefore for the group of active FGid streams of this input module which are intended for this output module.

Thus, such a weighted sharing also taking into account a value of maximum shared bandwidth MSBij to determine the shared bandwidth SBGij allocated to a pair of ITMi / OTMj modules, makes it possible to prevent this bandwidth SBGij from being allocated excessively compared to the desired bandwidth NFBid which is sufficient to drain the volume of data blocks DBVij awaiting transfer for the corresponding group of streams FGij.

In a preferred embodiment, it is chosen that the value of the maximum shared bandwidth parameter MSBij, for a pair of ITMi / OTMi modules, is equal to the value of the desired bandwidth parameter NFBij, established for this couple .

In the foregoing, two types of solution have been described concerning the fourth step defined above, which differs by their choice in terms of the type of parameters transmitted back to the input modules by the arbitration functions. As explained, these two types of solutions also differ in the type of function which

<Desc / Clms Page number 20>

 performs the calculation of the shared bandwidth value SBGij allocated to each of the pairs of modules ITMi / OTMi, namely by an arbitration function for at least one output module OTMj in the context of the first solution, or else by each ITMi module as part of the second solution. According to the invention, two types of methods, corresponding respectively to the framework of each of these two solutions, are also provided concerning the taking into account of a maximum shared bandwidth value MSBij to determine the shared bandwidth value SBGij allocated to a couple of modules.

According to the invention, in the context of the second solution described above, during the calculation performed by each ITMi input module according to said fourth step, to determine the value of the shared bandwidth parameter SBGid allocated to each pair of modules ITMi / OTMj that it forms with an OTMj output module, this input module also takes into account a maximum shared bandwidth value MSBij. In this case, the input module itself determines this upper limit value assigned to this pair of modules, as a function of the desired bandwidth parameter value NFBij which it has established for this pair of modules.

According to the invention, for each pair of ITMi / OTMd modules, each ITMi input module also transmits, to the arbitration function which performs said third step for a given OTMd output module belonging to this pair, the value of the parameter of the desired bandwidth NFBid that he established for this pair of modules. Thus, an arbitration function has the desired bandwidth parameter values NFBid which it receives from the ITMi input modules.

If this is the case, according to the invention, in the context of the first solution described above, during the calculation, performed by an arbitration function at least for a given OTMd output module to share its bandwidth ABd according to said fourth step, which determines the value of the shared bandwidth parameter SBGid allocated to each pair of ITMi / OTMd modules that it forms with this output module OTMd, the arbitration function also takes into account a value of maximum shared bandwidth MSBid which constitutes an upper limit of this allocated shared bandwidth, which it determines according to the desired bandwidth parameter values NFBid which it receives from the input modules of these pairs of modules.

According to the invention, to check whether it is useful to take account of these upper limit values MSBid of allocated bandwidth, the arbitration function determines beforehand whether the available bandwidth ABd of a given output module OTMd

<Desc / Clms Page number 21>

 is sufficient or not to satisfy all the requests for desired bandwidth NFBid supplied by the ITMi input modules and relating to the pairs of modules concerned ITMi / OTMd. To this end, the arbitration function performs a comparison between the value of this available bandwidth ABd and the value of the sum Ld of these desired bandwidth parameter values NFBid received from these input modules and concerning this output module. The result of this comparison is used to determine the value of the shared bandwidth parameter SBGid which is allocated to each of the input modules of the couples concerned.

When the result of this comparison indicates that this available bandwidth ABd is greater than or equal to the sum Ld of these desired bandwidth parameter values NFBid, it is possible to satisfy each of the desired bandwidth requests supplied by the ITMi input modules for pairs of ITMi / OTMd modules having this given OTMd output module as recipient. In this case, there is therefore no reason to limit, by an upper limit value MSBid, the value of the shared bandwidth allocated to each of these pairs of modules.

According to the invention, when the result of this comparison, for a given OTMd output module, indicates that its available bandwidth ABd is greater than or equal to the sum Ed of these values of desired bandwidth parameter values NFBid established and provided by the ITMi input modules of the pairs of ITMi / OTMd modules having this OTMd output modules as recipient, all the shared bandwidth parameter values SBGid allocated to these couples are at least equal to the respective width parameter parameters desired band NFBid relative to the corresponding couples.

Furthermore, according to the invention, when the result of this comparison indicates that this available bandwidth ABd turns out to be greater than the sum Ed of these values of desired bandwidth parameter values NFBid, this calculation plans to redistribute, to the pairs of modules concerned, an excess bandwidth EXBd obtained by difference between the values of this available bandwidth ABd and of this sum Ld. Then, the shared bandwidth SBGid, which is allocated to each ITMi / OTMd pair, is equal to the desired bandwidth NFBid for this couple, increased by an additional allocated bandwidth which is a part of this bandwidth surplus EXBd redistributed to each of the ITMI input modules. It is possible to provide different types of sharing of this excess bandwidth EXBd.

<Desc / Clms Page number 22>

In a preferred form of the invention, this excess bandwidth EXBd is distributed in equal parts, by dividing its value by the number n 1 of input modules ITMi of the node. In this case, the shared bandwidth SBGid allocated to a pair of modules concerned ITMi / OTMd is:
SBGid = NFBid + {ABd - Ld} / n 1.

According to a variant of this preferred form, this distribution in equal parts is restricted to only the input modules which have supplied a non-zero value of desired bandwidth NFBid.

On the other hand, when the result of this comparison indicates that this available bandwidth ABd is less than the sum Ld of these desired bandwidth parameter values NFBid, it is not possible to satisfy all the bandwidth requests desired NFBid provided by the ITMi input modules and relating to the pairs of ITMi / OTMd modules having this OTMd output module given as recipient. In such a case, as mentioned above, the result of a sharing which would be carried out strictly by proportional weighting with the relative priority weight parameter values by couple CRPWid, may not be optimal in terms of efficiency of use of the available bandwidth ABd.

This is the case when such a calculation would result in allocating, to at least one pair of ITMx / OTMd modules, a shared bandwidth SBGxd which is greater than the bandwidth which would be sufficient to carry its waiting traffic in a corresponding ITMx input module.

However, the result of such sharing is optimal, in terms of efficiency of use of the available bandwidth ABd, in the particular case where it results in allocating, to none of the pairs of ITMi / OTMd modules, a width of shared band SBGid which is greater than the sufficient bandwidth to carry its waiting traffic in a corresponding ITMi input module.

According to the invention, when the result of this comparison, for a given OTMd output module, indicates that this available bandwidth ABd is less than the sum Ld of these desired bandwidth parameter values NFBid, provision is also made two weighted calculation methods with upper limit concerning the fourth step defined above, which differs by the choice of the method used to redistribute the unallocated excess bandwidth to the pairs of modules whose shared bandwidth parameter value calculated by simple proportional weighting turns out to be greater than the maximum shared bandwidth value MSBid established for this couple.

<Desc / Clms Page number 23>

According to a first calculation mode, when the result of said comparison, for a given output module OTMd, indicates that the available bandwidth ABd is less than this sum Ld, this calculation also provides for classifying in advance the different pairs of modules concerned ITMi / OTMd, in two disjoint subsets of pairs of modules. The first subset includes the pairs of ITMx / OTMd modules whose shared bandwidth parameter value calculated by simple proportional weighting is greater than the maximum shared bandwidth value MSBxd established for this couple, the width of shared band SBGxd actually allocated to such a torque then being reduced by this upper limit value MSBxd. The second subset includes the pairs of ITMy / OTMd modules whose shared bandwidth parameter value calculated by simple proportional weighting proves to be less than the maximum shared bandwidth value MSByd established for this pair.

Then, this calculation determines, for each of the ITMx / OTMd couples of the first subset, an unallocated excess bandwidth value UEBxd which is equal to the difference between the shared bandwidth parameter value calculated by simple weighting and the value MSBxd maximum allocated shared bandwidth. Then it performs the sum of these unallocated excess bandwidth values UEBxd; the value of this sum corresponds to the overall unallocated excess bandwidth GUEBd for this OTMd output module.

Then, this calculation redistributes, to each pair of ITMy / OTMd modules of the second subset, a part of this excess overall bandwidth GUEBd, by adding this part to the shared bandwidth parameter value calculated by simple weighting, to determine the amount of shared bandwidth SBGyd that is actually allocated to these pairs of modules of the second subset.

In a preferred form of the invention, this overall unallocated excess bandwidth GUEBd is redistributed in equal parts, by dividing its value by the number Nd 'of input modules ITMy relative to the pairs of modules ITMy / OTMd of the second sub - together, that is:
SBGyd = CRPWyd x {ABd / GRPWd} + GUEBd / Nd 'According to a variant of this preferred form, this distribution in equal parts is restricted to only the ITMy input modules which have provided a non-zero value of desired bandwidth NFByd.

According to such a first weighted calculation method with upper limit, the sharing carried out, fairly with respect to the respective qualities of service, provides a

<Desc / Clms Page number 24>

 better efficiency of use of the available bandwidth ABd of each output module, insofar as it makes it possible to save an overall unallocated excess bandwidth GUEBd of the pairs of modules of the first ITMx / OTMd subset which have enough by relative to their desired bandwidth, to redistribute it to the pairs of ITMy / OTMd modules of the second subset which would not have enough. However, despite the improvement it provides, such redistribution is not yet optimal, because it can lead to the fact that, among the pairs of ITMy / OTMd modules of the second subset, the shared bandwidth SBGyd, which is actually allocated following this redistribution, ie greater than their desired bandwidth for some of these couples, but less than their desired bandwidth for some other of these couples.

An optimal redistribution is achievable using the second calculation method. According to this second mode, when the result of said comparison, for a given OTMd output module, indicates that the available bandwidth ABd is less than this sum Xd, to determine the values of shared bandwidth SBGid allocated to each of the pairs of ITMi / OTMd modules concerned having this OTMd output module as recipient, a recursive calculation is used for the Nd ITMu input modules having reported a non-zero value of desired bandwidth NFBud. For the other ITMv input modules having reported a zero value of desired bandwidth NFBvd, the shared bandwidth SBGvd, allocated to each ITMv input module of the ITMv / OTMd pairs is assumed to be of zero value. It can be noted that in practice, if it is desired to prevent any SBGvd bandwidth from being allocated to such ITMv input modules, provision is made for it to be possible to ensure that, for each module ITMv input that has no traffic waiting for the group of flows intended for a given OTMd output module, the desired bandwidth value NFBvd, provided by such a module, that is, by convention, nevertheless equal to a predetermined minimum value. Then these ITMv input modules are considered to be ITMu type modules for bandwidth sharing, which is carried out, in this case, for all the input modules.

According to this recursive calculation, a prior classification is carried out, in ascending order 1, ..., n, ..., Nd, of the values of a selection parameter by rank RSud which is associated with each of these modules d ITMu input. The value of this selection parameter by rank RSud is determined, for each of these input modules ITMu, by dividing the value of its desired bandwidth NFBud by that of its relative priority weight parameter by pair CRPWud, that is:

<Desc / Clms Page number 25>

RSud = NFBud / CRPWud.

(It should be noted that at this stage, the previous restriction of performing sharing only for ITMu type input modules whose NFBud value is not zero, is justified for all ITMv input modules, since their zero NFBvd value would result in a zero value of their rank selection parameter).

Then a sequence of weighted sharing calculation steps with upper limit is carried out successively for each of these Nd ITMu input modules, in ascending order, of rank 1, ..., n, ..., Nd , values of their selection parameter by associated RSud rank.

According to the invention, this sequence of Nd weighted sharing calculation steps with upper limit uses two recursive parameters which are updated at each rank step 1, ..., n, ..., Nd, given, relating to an ITMu input module classified at rank n according to the value of its selection parameter by rank RSud: A first recursive parameter provides the value of the remaining available bandwidth RABdn at rank n, including the initial value RABd 1 is equal to that of the available bandwidth ABd of the output module; a second recursive parameter supplies the value of the overall relative priority weight remaining RGRPWdn at rank n, whose initial value RGRPWdl is equal to that of the overall relative priority weight parameter GRPWd determined for this output module.

Then, each of the Nd weighted sharing calculation steps with upper limit, of a given rank z and relating to the input module ITMw which has been classified at this rank z according to its selection parameter by rank RSwd, determines d firstly the value of the shared bandwidth parameter SBGwd allocated to the pair of modules relative to this input module ITMw, and secondly the new values of these first and second recursive parameters, updated for the step next of row following z + 1 relative to another input module.

This shared bandwidth value SBGwd, calculated during a step of calculating given rank z, is equal to the smaller of two values: - the first value being that of the desired bandwidth NFBwd provided by this module d 'ITMw entry; - the second value being the result of the product of the value of the first recursive parameter RABdz by that of the relative priority weight parameter by couple
CRPWwd of this ITMw module, divided by the value of said second recursive parameter RGRPWdz.

<Desc / Clms Page number 26>

The value of the shared bandwidth parameter SBGwd, allocated to a pair of modules ITMw / OTMd relative to the input module ITMw which is processed during the step of calculating given rank z, is therefore obtained by the formula:
SBGwd = Min [NFBwd; CRPWwd x {RABdz / RGRPWdz}] This mathematical expression reflects well a weighted calculation with upper limit, according to which the shared bandwidth results, either from a sharing by proportional weighting, equal to:
CRPWwd x {RABdz / RGRPWdz} (if <NFBwd) is shared by upper limit equal to:
NFBwd (if <CRPWwd x {RABdz / RGRPWdz}) A step of calculating given rank z also includes determining the new values of these first and second recursive parameters, which are updated for the following calculating step, of rank according to z + 1, relating to another input module.

The new value RABdz + 1 of the first recursive parameter, for the next rank z + 1, is equal to its present value RABdz, for the given rank z, reduced by the value of the shared bandwidth parameter SBGwd which has just been allocated at this given rank z.

The new value RBFWdz + 1 of the second recursive parameter, for the next rank z + 1, is equal to its present value RBFWdz, for the given rank z, minus the value of the parameter relative weight priority by torque CRPWwd of the input module ITMw classified at this given rank z.

An interesting property of such a recursive computation is that it ensures, in the case where the available bandwidth ABd turns out to be less than the sum Ed of these values of parameter of desired bandwidth parameter NFBud, that the bandwidth shared SBGud allocated to different pairs of modules is less than or equal to, but never greater than, their desired bandwidth. Conversely, in the case where the available bandwidth ABd turns out to be greater than the sum Ld of these desired bandwidth parameter values NFBud, it is easy to carry out a sharing such as the shared bandwidth SBGud allocated to the different couples of modules is greater than or equal to, but never less than, their desired bandwidth NFBud. For example, it is then possible to perform a simple sharing such that shared bandwidth SBGud is equal to the desired bandwidth NFBud increased by an additional bandwidth obtained by equal redistribution, for Nd input modules, the overall unallocated excess bandwidth GUEBd, that is:

<Desc / Clms Page number 27>

GUEBd = ABd-Ed), so an additional bandwidth value equal to {ABd-Xd} / Nd.

This means that, for all of the input modules, the two types of situation where the allocated shared bandwidth is either greater or less than the desired bandwidth are strictly exclusive during a sharing operation given.

Also, according to the invention, this observation makes it possible to provide that the arbitration function, which performs the sharing of the available bandwidth ABd for a given output module OTMd, provides, to each input module ITMu, data sharing system comprising: - On the one hand, a GTlud allocation type indicator relating to the value of the shared bandwidth parameter SBGud, allocated to the pair of ITMu / OTMd modules corresponding to this input module. This indicator is intended to specify whether this shared width value is equal to, greater than or less than the value of the desired bandwidth parameter NFBud established by this input module.

 - On the other hand, a common parameter of allocation factor GFd, which here is advantageously of identical value for all the input modules and whose value constitutes, depending on the value of the allocation type indicator GTlud, that is an increase factor, a reduction factor in bandwidth. Although the two situations of increase and reduction are exclusive [as explained above], the use of an allocation type indicator value GTlud specific to each ITMu input module is justified in the case ABd <Zd, for which the recursive computation leads to allocating a shared bandwidth SBGud which is less than their desired bandwidth NFBud for at least part of these input modules, but which can be equal to NFBud for the other modules .

Then, to determine the value of the shared bandwidth parameter SBGud allocated to the pair of modules that constitute this input module ITMu and the given output module OTMd, each input module ITMu uses this sharing data, namely l allocation type indicator GTlud and the common allocation factor parameter GFd, which it receives from the arbitration function performing the sharing of the available bandwidth ABd for this output module, taking into account the values of its own parameters of relative priority weight per couple CRPWud and desired bandwidth NFBud.

Depending on the value of the GTlud allocation type indicator, an ITMu output module thus determines the value of the shared bandwidth parameter SBGud, in

<Desc / Clms Page number 28>

- Si GTlud indique [valeur SBGud] = [valeur NFBud], alors : SBGud = NFBud - Si GTlud indique [valeur SBGud] < [valeur NFBud], alors : SBGud = CRPWud x GFd, où : GFd = ABd/GRPWd - Si GTlud indique [valeur SBGud] > [valeur NFBud], alors : SBGud = NFBud + GFd, où : GFd = t ABd-Y-d Nd Ainsi l'utilisation de telles données de partage, permise par l'observation faite plus haut, a pour avantage de réduire le volume des données de partage que transmet chaque fonction d'arbitrage vers les différents module d'entrée. taking into account the values of its NFBud and CRPWud parameters, as follows:

- If GTlud indicates [SBGud value] = [NFBud value], then: SBGud = NFBud - If GTlud indicates [SBGud value] <[NFBud value], then: SBGud = CRPWud x GFd, where: GFd = ABd / GRPWd - If GTlud indicates [SBGud value]> [NFBud value], then: SBGud = NFBud + GFd, where: GFd = t ABd-Yd Nd Thus the use of such sharing data, permitted by the observation made above, has for advantage of reducing the volume of sharing data transmitted by each arbitration function to the different input modules.

As indicated above, the implementation of the flow control mechanism uses hardware and / or software means known in themselves, and the functions performed are likely to be more or less centralized or well distributed, at least in part. in the queue management units 5 of the input modules, as required and in particular for the functions mentioned above which are provided at the level of the mechanism, according to techniques familiar to those skilled in the art.

Claims (22)

CLAIMS 1. Method for sharing the available switching bandwidth for transferring data blocks through the switching network of an asynchronous switching node, this network interconnecting input terminal modules, provided with buffers allowing a temporary storage of the data blocks and of the output terminal modules, in which, on the one hand, each data block belongs to a stream of data blocks to which are assigned, at the level of an input module, at least two associated parameters, one of these parameters specifying the output module recipient of the flow, the other being a relative priority weight parameter (RPWq) of the flow, and in which, on the other hand, an internal control mechanism flow dynamically ensures a balanced and tending to be equitable sharing of the bandwidth generally available for the different flows of data blocks transferred from the input modules to the di different destination output modules, said method being characterized in that it provides: - a classification of the different data block flows by relative priority category (RPCk) as a function of their respective relative priority weight parameters (RPWq), a predetermined category weight parameter value (CWk) being assigned to each relative priority category (RPCk); - a supply, to each input module by the internal flow control mechanism, of shared bandwidth parameter values (SBGii), each parameter being relative to each of the pairs of modules (ITMi / OTMj) composed of an input module (ITMi) and a destination output module (OTMj), these shared bandwidth parameter values (SBGij) allocated by pair of modules being obtained by taking into account the value of the weight parameter of category (CWk) assigned to each stream, and by performing a succession of steps comprising: - a first step, at the level of each input module (ITMi), of determining the number of active streams (NAFijk), for each recipient output module (OTMj) and for each relative priority category (RPCk), an active stream being a stream for which there is at least one data block ready to be transferred through the switching network; a second step, for each input module (ITMi), of determining a relative priority weight parameter value per couple (CRPWij), for each pair of modules (ITMi / OTMj), this parameter value being obtained for a given pair of modules by performing the sum, for all of the relative priority categories (RPCk), of the products which are respectively calculated, category by category, by multiplying the value of the category weight parameter <Desc / Clms Page number 30>  (CWk) corresponding to one of the relative priority categories (RPCk), by the number of active streams (NAFijk) determined for this category at the input module (ITMi) of this pair and for the output module (OTMj) of this couple; a third step, for each output module (OTMj), performed by an arbitration function for at least one given output module (OTMd), of determining a value of overall relative priority weight parameter (GRPWd) , this parameter value being obtained by performing the sum of the relative priority weight parameter values per couple (CRPWijd) respectively calculated for each of the pairs of modules (ITMi / OTMd), said couples concerned, in which this output module considered (OTMd) is recipient; a fourth step, for each output module (OTMj), of sharing the available bandwidth (ABd) to access an output module (OTMd) given from the various input modules (ITMi) of said pairs of modules concerned (ITMi / OTMd), in which this output module is intended, this sharing, weighted and equitable, being performed in proportion to the relative priority weight parameter values by couple (CRPWid) calculated for the couples concerned, the results of this sharing for each given output module (OTMd) making it possible to determine, for each input module (ITMi), said shared bandwidth parameter values (SBGid) thus allocated to each of the pairs of modules (ITMi / OTMd) concerned for the transfer of the data blocks of all the data block flows relating to this pair of modules. 2. Method according to claim 1, characterized in that it provides that the determination of the value of shared bandwidth (SBGij) allocated to a pair of modules (ITMi, OTMj) is proportional to the available bandwidth ( ABj) to access the torque output module, proportional to the value of the relative priority weight parameter (CRPWij) calculated for this torque, and inversely proportional to the value of the overall relative priority weight parameter GRPWj calculated for the module exit of this couple. 3. Method according to claim 1, characterized in that it provides for the determination of the value of the relative priority weight parameter by couple (CRPWij) specific to a pair of modules (ITMi / OTMj), to be carried out by the input module (ITMi) of this couple, and that this input module transmits this parameter value per couple, which it determined within the framework of the second determination step, to an arbitration function (AF) flow control mechanism, which performs, at least for a recipient output module (OTMd) different <Desc / Clms Page number 31>  couples concerned (ITMi / OTMd), a collection of relative priority weight parameter values per couple (CRPWid) concerning this output module and supplied by all the input modules (ITMi) of the node. 4. Method according to claim 1, characterized in that it provides that, for each relative priority category (RPCk), the value of the category weight parameter (CWk) which is assigned to this category, is chosen identical for all input modules, and that each input module (ITMi) provides, for each recipient output module (OTMj), the numbers of active flows (NAFijk) determined during the first step by relative priority category (RPCk ), and transmits them to an arbitration function (AF ') of the flow control mechanism, which performs, at least for a destination output module (OTMd) of the different couples concerned (ITMi / OTMd), said second step of determination of the values of relative priority weight parameter by couple (CRPWid), after a collection of the numbers of active streams (NAFidk) relative to this output module and supplied by the set of input modules (ITMi) of the node. 5. Method according to claim 1, characterized in that it provides that an arbitration function (AF ") of the flow control mechanism performs, for at least one given output module (OTMd): - the determination according to said third step, of the value of overall relative priority weight parameter (GRPWd) for this output module, from the values of relative priority weight parameter by couple (CRPWid) relating to the various input modules ( ITMi) of the node, and - the sharing, according to said fourth step, weighted and equitable of the available bandwidth (ABd) to access this output module, taking into account the relative priority weight parameter values per couple (CRPWid) and global (GRPWd) relative to this output module, to calculate the value of the shared bandwidth (SBGid) allocated to each pair of modules (ITMi / OTMd) having this output module as recipient, in order to transmit , to the e modules input (ITMi) of these pairs, these shared bandwidth parameter values (SBGid) that it has calculated. 6. Method according to claim 1, characterized in that it provides that: - an arbitration function (AF '") of the flow control mechanism performs, for at least one given output module (OTMd), the determination, according to said third step, of the value of overall relative priority weight parameter (GRPWd) for this output module, from the values of relative priority weight parameter by torque (CRPWid) relating to the various input modules (ITMi) of the node, so <Desc / Clms Page number 32>  to transmit to them common sharing data, identical for all the input modules, which relate to the value of this global parameter (GRPWd) and to the value of the available bandwidth (ABd) which can be used to access this module given output; and each input module completes the weighted and equitable sharing calculation of the available bandwidth (ABd) of a given output module (OTMd) for transferring the data blocks intended for this output module, taking into account counts on the one hand said common sharing data received, relating to the values of this global parameter (GRPWd) and of available bandwidth (ABd), and on the other hand the value of the relative priority weight parameter by pair of modules (CRPWid), which is specific to it, for calculating the shared bandwidth parameter value (SBGid) allocated to the pair of modules that it forms with said given output module. 7. Method according to claim 6, characterized in that said common sharing data transmitted to all input modules (ITMi), for a given output module (OTMd), are the value of the global parameter (GRPWd) and that of the available bandwidth (ABd) relating to this output module. 8. Method according to claim 6, characterized in that said common sharing data transmitted to all input modules (ITMi), for a given output module (OTMd), are formulated by the value of a rate parameter sharing, which is obtained by dividing the available bandwidth (ABd) to access this output module, by the value of the global parameter (GRPWd) relative to this same output module. 9. Method according to claim 1, characterized in that it also provides that, when sharing available bandwidth (ABd) of an output module (OTMd) given according to said fourth step, the calculation of the shared bandwidth (SBGid) allocated to each pair of modules concerned (ITMi / OTMd) takes into account a maximum shared bandwidth value (MSBid) which constitutes an upper limit value of the value of this allocated shared bandwidth at each pair of modules, this upper limit value being obtained as a function of values of a desired bandwidth parameter (NFBid), which are determined by the different input modules (ITMi) for each of the pairs of modules ( ITMi / OTMd) relating to a given output module (OTMd), by evaluation of the volume of data blocks (DBVid) awaiting transfer for all the active flows in this input module (ITMi), which are intended to the output module (OTM d) of this couple. <Desc / Clms Page number 33> 10. Method according to claim 9, characterized in that, for each pair of modules, said upper limit value (MSBid) of the shared bandwidth allocated to each pair is equal to said desired bandwidth value relating to this couple (NFBid). 11. Method according to claims 6 and 9, characterized in that each input module (ITMi), when it calculates, according to said fourth step, the value of the shared bandwidth parameter (SBGij) allocated to each pair of modules (ITMi / OTMj) which it forms with an output module (OTMi), takes account of an upper limit value (MSBij) of this shared bandwidth, which it determines according to the parameter value of desired bandwidth (NFBij) that he established for this pair of modules. 12. Method according to claim 9, characterized in that, for each pair of modules (ITMi / OTMd), each input module (ITMi) also transmits to the arbitration function which performs said third step for a module given output (OTMd) belonging to this pair, the value of the desired bandwidth parameter (NFBid) that it has established for this pair of modules. 13. Method according to claims 5 and 12, characterized in that it provides, when sharing, performed by an arbitration function, the available bandwidth (ABd) for at least one output module (OTMd) given, to calculate the value of the shared bandwidth parameter (SBGid) allocated to each of the pairs of modules concerned (ITMi / OTMd) having this output module as recipient, also taking into account a bandwidth value maximum shared (MSBid), established for this couple and constituting an upper limit of this allocated shared bandwidth, which is determined by the arbitration function as a function of the desired bandwidth parameter values (NFBid) established and transmitted by the input modules of these pairs of modules. 14. Method according to claim 13, characterized in that, when determining, by the arbitration function, shared bandwidth parameter values (SBGid) allocated to the different pairs of modules concerned (ITMi / OTMd) having as recipient a given output module (OTMd) whose shared bandwidth (ABd) is shared, this calculation also provides for determining beforehand whether or not this available bandwidth is sufficient for all the input modules, by making a comparison between this available bandwidth (ABd) and the sum (Ed) of the desired bandwidth parameter values (NFBid) supplied by the input modules. <Desc / Clms Page number 34>  15. Method according to claim 14, characterized in that, when the result of said comparison indicates that the available bandwidth (ABd) is greater than or equal to said sum (bd), this calculation provides that the value of the parameter of shared bandwidth (SBGid), allocated to each pair of modules (ITMi / OTMd) having this output module (OTMd) as recipient, is at least equal to the value of the desired bandwidth parameter (NFBid) supplied by the input module of this couple. 16. Method according to claim 15, characterized in that, when the result of said comparison indicates that the available bandwidth (ABd) is greater than said sum (bd), this calculation provides for redistribution, to the pairs of modules concerned , an excess bandwidth (EXBd) obtained by difference between the values of this available bandwidth (ABd) and this sum (lad), so that the shared bandwidth (SBGid), which is allocated to a couple of modules (ITMi / OTMd), is equal to the desired bandwidth (NFBid) for this couple, increased by an additional allocated bandwidth which is a part of this excess bandwidth (EXBd) redistributed to the different modules d 'Entrance. 17. Method according to claim 16, characterized in that said excess bandwidth (EXBd) is redistributed in equal parts to the only input modules which have provided a non-zero value of desired bandwidth NFBid. 18. Method according to claim 14, characterized in that, when the result of said comparison indicates that the available bandwidth (ABd) is less than said sum (eu), this calculation also provides: - the prior classification of different pairs of modules concerned (ITMi / OTMd), in two disjoint subsets of pairs of modules: on the one hand, a first subset comprising the pairs of modules (ITMx / OTMd) including the bandwidth parameter value shared calculated by simple proportional weighting turns out to be higher than the upper limit value (MSBxd) established for this couple, the shared bandwidth (SBGxd) actually allocated to such a couple then being equal to this upper limit value (MSBxd), and on the other hand, a second subset comprising the pairs of modules (ITMy / OTMd) whose shared bandwidth parameter value calculated by simple proportional weighting is found to be infer ior to the upper limit value (MSByd) established for this couple; - a determination, for each of the pairs (ITMx / OTMd) of the first subset, of a value of unallocated excess bandwidth (UEBxd) which <Desc / Clms Page number 35>  is equal to the difference between the shared bandwidth parameter value calculated by simple weighting and the upper limit value (MSBxd) of allocated bandwidth; - an accumulation of the values of excess unallocated bandwidth (UEBxd) of each of the pairs of the first subset, the sum of these values constituting a value of overall unallocated excess bandwidth (GUEBd) for this output module ( OTMd); and - a redistribution, for each pair of modules (ITMy / OTMd) of the second subset, of part of this overall unallocated excess bandwidth (GUEBd), this part being added to the shared bandwidth parameter value calculated by simple weighting, to thus determine the amount of shared bandwidth (SBGyd) which is allocated to these pairs of the second subset. 19. Method according to claim 18, characterized in that said redistribution, at each pair of modules (ITMy / OTMd) of said second sub-assembly, of part of this excess overall bandwidth (GEXBd), is carried out according to an equal distribution by dividing this overall excess bandwidth by the number of couple of modules (Nd ') identified in this second subset. 20. Method according to claim 14, characterized in that, when the result of said comparison indicates that the available bandwidth (ABd) is less than said sum (bd), this calculation of the values of shared bandwidth (SBGid ), allocated to each of the pairs of modules (ITMi / OTMd) concerned having this output module (OTMd) as recipient, is performed recursively for the [Nd] input modules (ITMU) having supplied a non-zero value of desired bandwidth (NFBud), and in that it comprises: - a preliminary step of classification of these [Nd] input / input modules (ITMu) 1 in ascending order [1, .... n ,. ... Nd] of the values of a selection parameter by rank (RSud) associated with each of these input modules, each associated value (RSud) being determined, for each of these input dl modules (ITMu) 1 by dividing the value of its desired bandwidth NFBud by the value of its relative priority weight parameter by co uple CRPWud; and - a sequence of weighted sharing calculation steps with upper limit, carried out successively for each of these [Nd] input modules (ITMu), and in ascending order [1, ..., n, .. ., Nd] of the values of their associated selection parameter by rank (RSud). 21. Method according to claim 20, characterized: <Desc / Clms Page number 36>  relative by couple (CRPWwd) of the input module (ITMw) relative to this given rank H.

 - in that said sequence of weighted sharing calculation steps with upper limit uses two recursive parameters, updated at each step of rank [1, ..., n, ..., Nd] given, relating to a input module (ITMu) classified at a rank [n] determined according to its selection parameter by rank (RSud): a first recursive parameter providing the value of the remaining available bandwidth (RABdn), including the initial value (RABdl) is equal to that of the available bandwidth (ABd) of the output module, and a second recursive parameter providing the value of the remaining overall relative priority weight (RGRPWdn), whose initial value (RGRPWdl) is equal to that of the global relative priority weight parameter (GRPWd) determined for this output module; - and in that each of the [Nd] weighted sharing calculation steps with upper limit of said sequence, of a given rank [z] and relating to an input module (ITMw) classified at this rank, determines: d firstly, the value of the shared bandwidth parameter (SBGwd) allocated to the pair of modules (ITMw / OTMd) relative to this input module (ITMw), which is equal to the lesser of two values, the first value being that of the desired bandwidth (NFBwd), and the second value being the result of the product of said first recursive parameter (RABdz) by the relative priority weight parameter by torque (CRPWwd) of this input module, divided by the value of said second recursive parameter (RGRPWdz); and - on the other hand, the new values of said first and second recursive parameters, updated for the next step of calculating rank [z + 1] relating to another input module, the new value (RABdz + 1) of the first recursive parameter, for the next step [z + 1], being equal to its value (RABdz) present, for the current step of given row [z], reduced by the value of the shared bandwidth parameter ( SBGwd) which has just been allocated to this given rank [z], and the new value (RGRPWdz + 1) of the second recursive parameter, for the next step [z + 1], being equal to its present value (RGRPWdz), for the current step of given row [z], reduced by the value of the priority weight parameter 22. Method according to claim 20, characterized in that the arbitration function, which performs the sharing of the available bandwidth (ABd) for a given output module (OTMd), supplies each input module (ITMu), sharing data including an allocation type indicator (GTlud) indicating whether the value <Desc / Clms Page number 37>  of the shared bandwidth parameter (SBGud), allocated to the pair of modules (ITMu / OTMd) corresponding to this input module, is equal, greater or less than the value of the desired bandwidth parameter (NFBud) established by this input module, as well as a common allocation factor parameter (GFd) which is identical for all the input modules and whose value constitutes, depending on the value of the allocation type indicator (GTlud ), either an increase factor or a reduction in bandwidth factor, and in that each input module (ITMu) uses this sharing data (GTlud; GFd) received from the arbitration function performing the sharing the available bandwidth for a given output module (OTMd), taking into account its own parameters of relative priority weight per couple (CRPWud) and desired bandwidth (NFBud), to determine the value of the parameter allocated shared bandwidth (SBGud) e to the pair of modules that constitute this input module and the given output module.