WO2012004689A1

WO2012004689A1 - A method and system of bandwidth control

Info

Publication number: WO2012004689A1
Application number: PCT/IB2011/052249
Authority: WO
Inventors: Vincenzo Gesmundo; Mario Marchese; Maurizio Mongelli
Original assignee: Selex Communications SpA
Current assignee: Selex Elsag Datamat SpA
Priority date: 2010-05-24
Filing date: 2011-05-24
Publication date: 2012-01-12
Anticipated expiration: 2012-11-24
Also published as: RU2597227C2; IT1400169B1; RU2012155863A; GB2494350A; AU2011275478B2; GB201222815D0; GB2494350B; AU2011275478A1; MY158555A; ITTO20100429A1; BR112012029928A2

Abstract

A method for managing heterogeneous traffic flows accessing a communications network (WN) is described, in which an access node (G) of the network operates an aggregation of traffic flows (f_i ^L3) in at least a real transmission queue (Q_i ^L2) at a predetermined protocol layer (L2) of the network, including allocating a predetermined bandwidth ( θ_i ^max) to the transmission queue (Q_i ^L2) in order to guarantee a related quality of service (QoS_i ^L3) established according to a service level agreement. The method comprises: generating a virtual transmission queue (Q_iv ^L2) in the data plane (UP_L2) of the system for managing traffic flows (f_i ^L3) accessing the network, including a current replication of the aggregated traffic flow (/J ) at the real transmission queue (Q_i ^L2) measuring at least a parameter representative of the current quality of service (QoS_i ^L3oL2) obtained on the virtual transmission queue (Q_iv ^L2) in a plurality of predetermined successive observation horizons (OH_j(k)); and dynamically estimating the bandwidth need (θ*_i ) of the virtual transmission queue (Q_iv ^L2) complying with the quality of service as a function of the value assumed by the aforementioned parameter in each observation horizon (OH_i(k)); hence according to the estimated bandwidth need ( θ*_i) the bandwidth ( θ_i ^max ) allocated to the real transmission queue (Q_i ^L2) for the subsequent observation horizon (OH_i(k+1)) is modified or the traffic flow (f_i ^L2) towards the queue (Q_i ^L2) is limited. The method can be carried out at an access node of a communications network, such as a gateway device (G) adapted to interconnect two local and/or wide area networks (TN, WN) having different architecture, arranged for carrying out a process of aggregation of traffic flows (f_i ^L3) at a data link protocol layer (L2), or an edge router adapted to interconnect two portions of the same communication network (TN; WN), arranged for carrying out a process of aggregation of traffic flows (f_i ^L3) at a network protocol layer (L3).

Description

A METHOD AND SYSTEM OF BANDWIDTH CONTROL

The present invention concerns the field of communications and more in particular a method and a system for controlling the allocation of bandwidth to traffic flows conveyed at an access point of a communications network.

Specifically, the invention covers a method for managing heterogeneous traffic flows accessing a communications network according to the preamble of claim 1 , and a system for managing traffic flows accessing a communications network according to the preamble of claim 19.

In a communications network, the communication aspects between nodes are managed according to a predetermined paradigm based upon a protocol stack performing a layer-by- layer communication.

A suite of protocols implementing a protocol stack according to an established paradigm (for example the ISO/OSI model or the suite TCP/IP used in Internet) represents a group of standards that are connected to one another which define the reference architecture of a communications network and it is made up of different functional levels (or layers). For example, the well-known ISO/OSI reference model separates the various functionalities of a communication process in seven separate layers, a physical layer (LI ) and a link layer (L2) with reference to the network wiring, a network layer (L3), a transport layer (L4) and a session layer (L5) with reference to the logic communication of the data between network units, a presentation layer (L6) with reference to the method with which the data transmitted interacts with the network applications and an application layer (L7) with reference to the applications available for all the users of the network.

A plurality of parameters correlated to the network traffic (typically the loss rate of the data packet, the transmission delay, the available transmission bandwidth) determines the qual- ity of service (QoS) offered in managing the traffic on a communications network. This depends upon the performance achieved at each layer of the protocol stack and it is charac- terised by parameters based upon the functions implemented at different layers of the protocol stack and at the interfaces between its layers. For example, with reference to the OS! model, the quality of service comes from the configuration of the physical layer and of the link layer, which offer specific transportation services to the upper network layers. A service contract that provides for respecting predetermined parameters of quality of service is generally indicated as a Service Level Agreement (SLA).

In the operative field of a communications network, traffic flows generated by the upper network levels are managed at the network layer and link layer at the nodes of the network by processing modules implemented through application specific electronic devices or by processing and storage electronic devices programmed according to one or more code modules, which respectively form a control plane (used for managing the signalling information) and a user or data plane (used for transporting the user data). The data plane operates directly on the traffic flow under the management and supervision of the control plane, to forward the traffic flows to a physical interface adapted to convey the information along a transmission channel. Even when the network layer achieves efficient mechanisms for keeping a predetermined quality of service (for example, according to the protocols IP Int- Serv, IP DiffServ, MPLS) it is necessary for the lower layers to ensure the connection to the physical channel maintaining specific quality of service performance constraints. If this does not happen, making complex mechanisms for maintaining the quality of service at the upper layers could be insufficient and therefore completely useless.

Consequently, the quality of service requirements must "project" vertically along the protocol stack and must be satisfied by all layers of the protocol stack.

This means that the protocols at the link layer (the second layer of the protocol stack, hereafter identified in the abbreviated form L2) must achieve suitable aggregation mechanisms of logically distinct traffic flows from the upper layers to respect the service level agreement defined at the upper network layer (third layer of the protocol stack, or in its abbrevi- ated form L3). In some cases, in particular in radio environments (also called wireless), it should further be noted that the layer L2 acts in cooperation with the physical layer (LI ) through the application of specific solutions of the inter-layer type (known as cross-layer solutions).

The interaction between layers in this context qualifies a so called "mapping of the quality of service" or more frequently "QoS Mapping". The concept of mapping comes from the technological leap encountered at the access point of the network in which the operations of aggregation of data flow are carried out. An access point (node) of the network, in the following generally identified by a logic gateway device, can indeed interconnect two different network portions (or portions of the same network) in which corresponding different aggregation schemes of the data packets are applied to the traffic flows. Moreover, concerning the gateway, there can even be a modification in the encapsulation format of the data, dictated by the specific protocols used, for example when a portion of network is based upon an IP protocol, whereas another portion is based upon an ATM protocol.

In all these cases, suitable mechanisms must be adopted to calculate the exact bandwidth necessary for the traffic flows conveyed to layer L2 so as to ensure the quality of service defined at layer L3.

The problem of QoS Mapping specifically makes it necessary to tackle the technological aspect of the allocation of bandwidth to the traffic flows dealt with, or rather, of controlling the bandwidth assigned to the single traffic flows, in conditions of heterogeneous traffic. The problem in particular concerns controlling the bandwidth in the case in which different service classes are aggregated. Such an aggregation leads to the generation of heterogeneous channels from the point of view of the traffic sources and of the QoS requirements.

Since each traffic flow is forced to respect a specific guaranteed QoS, a coherent allocation of bandwidth must be foreseen on the network, both before, and after the aggregation operations of the traffic flows. In general, allocating bandwidths to the traffic flows before entering a gateway device of a network is controlled according to predetermined methods specific to the communication technology outside the network, which are not the subject of the present invention. The invention concerns the aspect correlated to the allocation of bandwidth to traffic flows aggregated at the gateway device for the propagation in the specific network.

Controlling the bandwidth in homogeneous conditions is a problem that is widely treated in scientific and patent literature.

On the contrary, the aspect of controlling the bandwidth in conditions of heterogeneous traffic is currently a matter that is still open for discussion and that has not received the necessary attention.

In the patent literature, it is possible to find different documents that tackle the problem of the adaptive control of the bandwidth, with the purpose of meeting the QoS requirements of transmission of data packets. The prior art, if applied to solve the problem of controlling the bandwidth in heterogeneous traffic conditions as is the subject of the present invention, leads however, to a sub-optimal use of the bandwidth resources available to a network.

Amongst the prior art documents concerning the problem of QoS Mapping, patent application EP 1 1 13 628 in particular should be mentioned, which concerns a mechanism for managing the quality of service on IP protocols for a wireless network. More specifically, this document proposes a layered architecture for controlling the quality of service on the entire protocol stack of communication of the network. The description, however, does not directly tackle the problem of estimation and allocation of the bandwidth and does not go into explicit details on how it is possible to optimise the controlling of the bandwidth which the different management levels of the quality of service in the protocol stack must perform.

Indeed, it should also be noted that if a portion of network is based upon a system of wireless communication, in generating the relative format of encapsulation of the data, specific mechanisms of channel degradation are applied. All these elements make the problem of determining the bandwidth at the gateway a very difficult task.

The inventors Mario Marchese and Maurizio Mongelli have tackled the aspects of QoS mapping, with particular reference to making interfaces between the network layers and the link layers of a communication protocol and the definition of algorithms of allocation of bandwidth to the link layer, respecting the QoS constraints.

The articles "Vertical QoS Mapping over Wireless Interfaces", IEEE Wireless Communications, Vol. 16, No. 2, April 1 , 2009, pages 37-43, "Neural Bandwidth Allocation Function (NBAF) Control Scheme at WiMAX MAC Layer Interface", International Journal of Communication Systems, Vol. 20, n. 9, December 12, 2006, pages 1059-1079 and "Optimal Bandwidth Provision at WiMAX MAC Service Access Point on Uplink Direction", 2007 IEEE International Conference on Communications, June 24, 2007, pages 80-85, describe a control law that acts upon queues of a link layer directly assigning them a transmission speed such as to respect the quality of service agreed upon.

The general purpose of the present invention is that of optimising the allocation of bandwidth resources in a communications network, and specifically at an access point of the network operating an aggregation of heterogeneous traffic flows, maintaining the quality of service agreed upon according to an established service level agreement of the services provided by the network.

More in particular, one puipose of the invention is that of offering a better coordination between protocols at different layers of a protocol stack on which a communications network is based, so as to allow a more effective allocation of bandwidth at an access point of the network operating an aggregation of heterogeneous traffic flows, and to ensure a predetermined quality of service with respect to an established service level agreement offered by the network.

The present invention also has the purpose of calculating in the most reliable way possible the exact bandwidth need required by an aggregated traffic flow at an access point to a communications network, so as to comply with a predetermined quality of service with the minimum amount of bandwidth possible.

According to the present invention, such purposes are achieved thanks to a method for managing traffic flows having characteristics that are claimed in claim 1.

Particular embodiments form the subject of the dependent claims, the content of which should be considered as an integral or integrating part of the present description.

A further subject of the invention is a system for managing traffic flows having the characteristics claimed in claim 19.

The invention also concerns a computer program or group of computer programs for performing the aforementioned method of managing traffic flows, as well as an access node of a communications network and a communications network comprising a system for managing traffic flows, as claimed.

In brief, the present invention is based upon the principle of modifying the structure of the control plane and of the data plane at the levels of the protocol stack involved in the aggregation of heterogeneous traffic flows, and for this reason it defines supplementary entities operating in the field of a protocol stack of a device for accessing a communications network (gateway), the role of which is that of controlling the calculation of the exact bandwidth need of the traffic flow entering the device.

In particular, these entities are represented by components or processing modules of the resource manager (RM) of the respective control plane at the layers L3 and L2 of the gateway device. The resource manager of layer L2 (hereafter, in brief, L2RM) acts so as to ensure the quality of service established at layer L3 and mapped on layer L2, calculating in real time the exact bandwidth need of the flows conveyed at layer L2 and consequently changing the corresponding allocation of the bandwidth resources. _ . In order to do this, differently from the prior art, the resource manager of the layer L2 applies a process of dynamically estimating the bandwidth based upon periodic measurements of the current quality of service applied to a virtual transmission queue (or traffic queue), which is a copy of the real transmission queue (or traffic queue) simultaneously managed by the data plane of the same layer. The bandwidth provided for forwarding the real traffic is initially oversized and is adapted periodically as a function of the outcome of the dynamic estimation obtained in a previous calculation time based upon the measurements carried out on the virtual transmission queue.

Advantageously, the transmission speed of the queues at the link layer is maintained within a safety threshold with respect to the transmission speed of the virtual queues so as to avoid possible imprecisions of the control law itself.

Non limiting examples of possible mathematical forms used to carry out the estimation process are provided in the following of the present description. The resource manager of the layer L2 exploits the result of the estimation process and consequently modifies the allocation of the bandwidth resources to the gateway device. In this context, the resource manager of layer L2 makes use of primitive communication specifications used to communicate the outcome of the process of modifying the bandwidth to the resource managers of the upper layers.

In the case in which the resource manager of layer L2 determines that not enough bandwidth resources are available to support the required quality of service, it informs the resource manager of the upper layer L3. The modalities in which the resource manager of layer L3 reacts to such communications is however outside the field of the present invention.

The entities of the resource manager of layer L2 and of the resource manager of layer L3 can be installed in the respective control planes, like for example the control plane IP at layer 3 or the control plane DVB at layer 2 without affecting the per se known original structure of such planes. These entities can be built by processing modules that are adapted to run computer programs, possibly in the form of program updates, whereby they are adapted to be loaded onto the control planes of the gateway so as to not interfere with the original architecture.

The invention advantageously has application in different embodiments, relative to different types of access points of communication networks in which an aggregation of the network traffic occurs, in any form, including gateway devices, routers or the like, that are adapted to carry out conversions of communication protocols between nodes of local and/or wide area networks having different architecture, in which different traffic flows entering the network are aggregated together and are forwarded to the nodes of the network, and it is independent from the embodiment of the device.

Possible examples comprise access points of wireless terrestrial networks (for example: Tetra, WiFi, WiMAX) or of satellite communication networks, which consider a technological leap between the third and second layer of the protocol stack. A further example at layer L2, which is not related to wireless technologies, is the encapsulation of the IP traffic over cabled Ethernet technologies operating according to model 802. lp (i.e. Ethernet with quality of service). The invention is of particular interest for mapping the quality of service in wireless environments where the bandwidth is a scarce resource, in comparison with cabled systems in which the optimisation of the bandwidth is not a critical problem and can be provided through a suitable oversizing of the resources (bandwidth and buffers of the network nodes) available.

In a different embodiment of the invention, in which there is an aggregation of heterogeneous traffic flows without changing the format of encapsulating data, and without using counter-measures for the channel degradation, examples of devices for accessing a communications network are represented by the edge router devices, for example operating in technological scenarios of traffic aggregation IntServ over DiffServ, IntServ over MPLS or DiffServ over MPLS. In this context, not only do gateways (also called edge routers) operate based upon operations of mapping the quality of service between layer L3 and layer L2, but also involve mapping operations exclusively referred to different network technologies and protocols operating at layer L3.

Further characteristics and advantages of the invention shall be shown in detail in the following detailed description, given as an example and not for limiting purposes, with reference to the attached drawings, in which:

fig. 1 is a schematic representation of an architecture for accessing a communications network, comprising a gateway device arranged between a portion of terrestrial network and a portion of wireless network;

fig. 2 is a schematic representation of the entities of layer L3 of the protocol structure of the gateway according to the invention;

fig. 3 is a schematic representation of the entities of layer L2 of the protocol structure of the gateway according to the invention;

fig. 4 is a schematic representation of the entities concurrent to the mapping operations of the QoS between layer L3 and layer L2 of a protocol stack, according to the invention;

fig. 5 is a schematic representation of the structure of a resource manager of layer

L3;

fig. 6 is a flow chart of the method for calculating the bandwidth need of the traffic flows at the protocol layer L2, according to the invention;

fig. 7 is a flow chart of the method for verifying the stabilization of the calculation of the bandwidth need, for verifying imminent congestion or bandwidth release at the protocol layer L2, according to the invention;

fig. 8 schematically shows the communication signals between entities at the protocol layer L2 according to the invention; and

fig. 9 shows the communication signals between entities of the protocol layers L2 and L3 according to the invention.

The invention concerns a method and a system that are adapted to perform the control of the bandwidth for an aggregated traffic flow at a gateway device, in which mapping operations of the quality of service (QoS mapping) are applied. More specifically, the invention concerns cases in which two or more traffic flows defined at a network layer (for example, L3) of a protocol stack are combined together in a single traffic flow at the same layer (L3) or at a lower link layer (L2), for which the exact bandwidth that must be made available to the aggregated flow is not known.

The invention firstly concerns the mapping operations of the pre-established quality of service between a first upper layer and a second lower layer of a protocol stack, or rather between different protocols operating at the same network layer of the stack. Secondly, the invention defines the physical and/or logic entities, or rather the physical hardware devices and/or software processing modules, which can be used at the layer L3 and at layer L2, respectively, of a device for accessing a network, such as a gateway device, for coordinating the actions necessary so as to optimise the controlling of the bandwidth assigned to a combined traffic flow.

The present invention shall now be described with reference to a currently preferred embodiment which considers a mapping of the quality of service from layer L3 to layer L2 of a predetermined protocol stack.

Fig. 1 schematically represents the predisposition of a network device G acting as a gateway between a first and a second communications network, for example between a portion of terrestrial network TN and a portion of wireless network WN.

Examples of interested wireless networks are satellite communication networks, WiFi communication networks, WiMAX communication networks or wireless sensor networks.

The traffic generated by the users of the networks is conveyed through the portion of terrestrial network TN in direction of the gateway G. From here, the traffic is routed outside of the portion of terrestrial network TN towards the wireless network WN. At the gateway device the lower layers of the protocol transmission stack applied by the device itself are indicated in detail.

The objective of the invention is to maintain a specific level of quality of service (QoS) along the entire communication chain. This means that the established quality of service in a first moment must be ensured both in the terrestrial network TN and in the wireless network WN as if there was no change in technology at the gateway G between the two networks. The change in technology is due to the different protocols used to offer a communication service on the terrestrial network TN and on the wireless network WN.

The quality of service is guaranteed in quantitative metric terms through the indication of an admissible data packet loss threshold, of an admissible delay (average) in the data packet transmission or of an admissible jitter (variance of the delay) in the data packet transmission. Different traffic contracts can be defined between a provider of the terrestrial communication network, a provider of the wireless communication network and one or more end users. The quality of service of a traffic contract is established through a service level agreement (Service Level Agreement, SLA) in which the aforementioned evaluation metric terms of the quality of service are declared. Each traffic class has its own service level agreement. This means that a traffic contract for each specific traffic class is declared.

In the field of the present invention it is considered that the traffic contract is of course satisfied on the portion of terrestrial network TN trough a suitable allocation of the bandwidth resources at the routing devices of the network (therefore, at the layer L3 of the protocol stack). Therefore, the metrics that define a predetermined SLA refer to the performance in terms of quality of service at the layer L3 (for example, the loss of IP packets ), since the end users must not be able to notice the technological leap between the layers L3 and L2 at the gateway.

Typically, the layer L3 is a network layer based upon IP technology (IPv4 or IPv6). Examples of layer L2 protocols are WiMAX, DVB, ATM or other dedicated encapsulations for the specific wireless channel, such as Stanag 5066 or WHDLC for IP-over-radio.

The gateway device G, or a similar network access point acts as an interface between the two layers L3 and L2, and it is responsible for mapping the traffic from layer L3 towards layer L2. Such a mapping operation essentially consists in encapsulating the packets trans- mitted at layer L3 in a data frame at layer L2 and in the selection of the specific transmission queue at layer L2 of the data frame corresponding to the packets that are defined at layer L3.

Under layer L2 there is the physical transmission channel (in this example, the wireless channel) which has its own transmission capabilities identified at layer LI : frequency spectrum, encoding methods, Bit Error Rate (BER), characteristics of fading and so on. The characteristics of layer LI however, depart from the scope of protection of the present invention and shall not be further discussed in detail in the rest of the description since they are not necessary in order to understand it. The only things that the method and system subject of the invention need to know relative to the characteristics of the physical channel concern the format of the error correction codes, for example the header of forward error correction (FEC), applied to layer L2, and shall be discussed in the rest of the description with reference to figure 4.

Figure 2 represents in detail, in a schematic form, the configuration of layer L3 with reference to the data plane UPL₃ and to the relative control plane CPu.

At the level of the data plane a set of queues Qi ^U, .... Q_N ^L3 is defined, adapted to separate different traffic classes and to ensure different levels of quality of service according to the service level agreement related to each traffic class.

Each single queue Q|^L3 (i = 1 , ... N) is made up of a respective buffer Bj^L3 in which the data packets defined at layer L3 are stored before being transmitted, and a server Sj^L3 of the buffer. The service rate defines the transmission speed of the exiting packets conveyed towards layer L2, and is a synonym of "service capability" and "bandwidth allocation".

The queues at layer L3 are obtained via hardware or software in the gateway device G. For example, in typical routing devices the queues, in which the quality of service is guaranteed through a suitable bandwidth allocation, are output queues made via software before the transmission towards the output links. In so-called Open Router architectures, based upon Open Source operating systems (typically, based upon Linux operating system) the queues at layer L3 are made by software modules included in the operating system.

A specific methodology is used at the level of the control plane CPu for classifying the traffic and the allocation of the bandwidth resources at layer L3, for example a DiffServ methodology. More specifically, at the level of the control plane the resource manager (L3RM) is responsible for allocating resources at the layer L3 of the gateway and is aware of the agreement concerning the level of service available by the network. It can also apply signalling protocols, like RSVP, to control the entire communication chain of the terrestrial portion of the network. DiffServ, IntServ, MPLS are possible examples of methodologies of engineering the traffic and quality of service used by the resource manager at layer L3.

Again with reference to the data plane UP_L3, flows of different traffic classes are identified with f,^u, and OoS, indicates the corresponding level of quality of service, for the i-th queue. Here, a flow consists of a sequence of packets, the temporal evolution of which follows a stochastic process whose statistical characteristics (average, variance) can be used by the resource manager L3RM to manage the allocation of the bandwidth resources at the layer L3.

It should be noted that the generation of packets, both at layer L3 and at layer L2, is a statistical process, and as such it shall be considered in the rest of the description. This characteristic is at the base of the potential and of the performance of the method subject of the invention.

Figure 3 represents in detail, in a schematic form, the configuration of layer L2 of the gateway with reference to the data plane UPL2 and to the relative control plane CPL2-

As a whole, the structure can be compared to the one of fig. 2 with reference to layer L3 of the protocol, and by analogy the same reference numerals have been assigned to the components. At the level of the data plane a set of queues Q|^L2, Q_N of layer L2 is defined, each single queue Q,^L2 (i = 1 , ... N) being made up of a respective buffer B,^L2 and a server Sj^L2 of the buffer.

The resource manager L2RM is aware of the entire channel capacity available on the physical channel (wireless channel, in the present example) and is responsible for the allocation of the bandwidth at layer L2.

The process of allocating the bandwidth resources at layer L2 according to the invention comprises the following operations:

i) mapping a specific traffic flow present at layer L3 towards a specific transmission queue provided at layer L2;

ii) allocation of a bandwidth to each queue of layer L2;

iii) estimating the exact bandwidth need of each queue of layer L2 sufficient for satisfying the agreement concerning the level of service.

More specifically, in the rest of the description it shall be explained how the operations at point iii) are actuated and how the results of such operations affect operations i) and ii), irrespective of the specific techniques used to actuate such operations in the first time instances in which the gateway device is running.

Typically, the queues at layer L2 are made via hardware and are available in a number that is smaller with respect to the queues at layer L3. Consequently, it is necessary to carry out some operations of aggregation of the traffic classes from layer L3 to layer L2, and the details of the mapping operations of the QoS applied between layer L3 and layer L2 are described with reference to fig.4.

Without departing form generalization, fig. 4 shows what happens for a generic queue Qj at the layer L2.

A subset of traffic flows coming from layer L3, the indexes of which are <··>'», , are con- veyed along the respective i-th queue. The flows at layer L3 are indicated fi[^'' >^■■■> /,„'^'* and the corresponding levels of quality of service are indicated Q^oS!, > -> Q^oS .

The queue at layer L2 conveys for example two types of traffic, respectively voice traffic conveyed according to the Voice over IP (VoIP) protocol, which requires a loss of packets that is not greater than 1%, and video traffic (on IP), which requires a loss of packets that is not greater than 0.1%. The flows ./,'• · _' , are aggregated together in a single flow " entering the i-th queue at an encapsulation and framing module EF. The packets at layer L3 of the flows are encapsulated on a single transmission frame to layer L2 (for example IP over ATM), possibly applying some optional encapsulation processes, like for example in the case of the CS encapsulation format in WiMAX technology.

Another process of interest which is carried out in an operation in mapping from layer L3 to layer L2 concerns the information added so as to contrast the fading of the physical channel and to limit the bit error rate (BER) at layer L2, for example codes of forward error correction, typically contained in an overhead field of the layer. The aforementioned protection codes can have variable sizes as a function of the instantaneous value of the signal/interference ratio coming from layer LI through a communication primitive typically available at the interface L2-L1 , identified in the figure by the SIR (Signal to Interference Ratio) module.

Due to the aggregation and encapsulation operations at layer L2, the corresponding statistical properties of the stochastic processes at layer L3 relative to the flows Λ' ^" are mixed together so that the estimation of the statistical properties of the stochastic processes that are entering the i-th queue, and of the relative quality of service at layer L2, is a very difficult task.

Such a task is solved thanks to the provision of a copy of the data plane at layer L2, in the context of the invention identified as a virtual copy, in comparison with the real data plane. The real data plane UPu comprises real queues Qj , QN ^" which correspond to the queues actually made in the gateway device. The respective service rate of the i-th queue, or better the respective bandwidth allocated to the i-th queue, is indicated with the symbol e ^iax . For each queue, a service rate or allocated bandwidth θ^^αχ is defined at the beginning of the service life of the gateway device G by the resource manager of the layer L2 (L2RM) on a planning basis that is dependent upon the available traffic forecast for a reasonable prolonged period of time. For example, by applying a precautionary oversizing principle, the parameter #,^Λ "^Λ for the i-th queue can be established as a function of the worst traffic condition foreseen entering into the i-th queue. In other words, this means, considering all the possible sources active at the same time and consequently establishing the #,^Mm necessary in order to simultaneously satisfy all the services requirements. For example, for ten all active VoIP sources, with respective input transmission speed equal to lOOkpbs and a header at layer L2 with overhead of 20% in the worst condition of fading of the physical channel, the principle of oversizing would mean setting #, '"^Λ = 1.2 Mbps. Other choices are of course possible by applying more sophisticated traffic forecast statistical charts, and in general different models can be used to establish the parameter #,^Λ "^Λ for every queue Q, at the beginning of the life of the gateway device.

Hereafter, we shall explain how to modify #,^Α'"^Λ over time so as to minimise the load of bandwidth resource allocation and maintain the required quality of service, exploiting the statistical characteristics of the sources of the traffic flows and the duplication of the data plane.

The virtual data plane UPL2^V comprises a plurality of virtual queues Qiv^L2, ···, QNV^L2 each of which is a software replica, or a hardware emulation of the corresponding real queue Qi^L2, QN^L2. This means that each packet from layer L3 and encapsulated at layer L2 is sent both to the real i-th queue and to the virtual i-th queue, which is a copy of the first one.

The only difference consists of the indication of the parameter of service rate, or rather of bandwidth allocated for the virtual queue : instead of #,^Wm . Θ^* is defined as the mini- mum bandwidth necessary for meeting the quality of service levels Q S^ , ..., OoS^ _{m t}i_ie i- th queue. It represents the exact bandwidth need necessary in order to satisfy the service level agreement with respect to the flows conveyed along the i-th queue after carrying out the mapping operations of the QoS.

Hereafter, we shall explain in detail the method through which the parameter θ' is calculated. It should be noted that, in general, the process of replication of the packets towards the single virtual queues (together with the calculation applied to them so as to obtain θ' ) can require the application of a dedicated microprocessor in the case in which there are computational limits in the native hardware structure of the gateway device.

In detail, fig. 5 shows the structure of the resource manager at layer L2 (L2RM) relative to the calculation of the service rate (or allocated bandwidth) of the virtual queues. It comprises a series of decision maker modules DM, each one being associated to a respective virtual transmission queue Q,y^L2- Specifically, a decision maker (DM) i-t module (DM,) is assigned to the i-th queue for the calculation of the parameter θ' . Each module DM,, based upon the calculated value of Θ, , is arranged for communicating messages of "imminent congestion" or of "bandwidth release" to a main processing module MAIN of the layer resource manager (L2RM).

Fig. 6 shows the algorithm used by each decision maker module DM, to calculate the parameter Θ ^*.

A sequence k = 1 , 2, ... of observation horizons OH,(k) is defined for each decision maker module DM,, during which the virtual i-th queue is monitored in a succession of time instances according to a pre-established periodicity. An information vector I_t(k) is formed for each observation horizon OH,(k), having as elements the entities necessary in order to carry out the calculation of estimation of the bandwidth as shall be indicated in the rest of the description, which vary from one case to another, as a function of the mathematical formula of estimation effectively used. The information vector I,(k) triggers the calculation of the service rate of the i-th queue at the time £+1 , thus generating a parameter Θ, (k + \) as specified in the rest of the description.

The calculation of Θ^* at the instant k+\ (e^*(k + \)) depends upon the levels of quality of service reached on the virtual i-th queue in the observation horizon OH,(k), and indicated

QoS!;ⁱ ' (k),...,OoS!;^>''^L-(k) _and upon the corresponding errors vector between these quantities and those indicative of the quality of service required by the service level agreement

QoS!; ...,QoS _t each element of which is indicated: «?,(·,*) = [QOS,¹ - OoS,' "^L2 (k) in which / indicates the /-th flow at layer L3 conveyed along the i-th queue at layer L2, i.e. for which

The general expression for the calculation of the parameter d'(k + 1) is the following:

e; (k + \)= F(; e,(:k)) θ' (k + 1 ) = Max (A + !),...,0, (A + 1),...,6> (k + \)

The operation ^Mx[] consists in selecting the highest bandwidth need value from the different traffic classes at the layer L3.

In the case at hand, the information vector I,(k) consists in the quantity e_l(- ) = (QoS! -QoS! "^L1(k) .

The algorithm represented in fig.6 and described above is repeated by the /-th decision maker (DM,) during each observation horizon OH,(k), k= 1,2, ...

Typically, the temporal dimensions of the observation horizon OH,( ), is in the range [1, 360]s, and is a function of the specific applications to be monitored in the i-th queue.

Different forms of the control law F(-) can be applied in the context of the present invention and hereafter we shall provide some suggestions as an example and not for limiting purposes. If the quality of service of interest is the Packet Loss Probability, PLP, or the Average Delay, AD, of the packets at layer L3, it is possible to use the model of analysis of the infinitesimal perturbations to deduct a formulation of the control law F(-) of the gradient type, as follows:

where is the gradient step size.

More specifically, for the case in which the quality of service is defined based upon the packet loss probability, the following formulas are obtained:

where:

- I ι ( · ) is the loss rate measured for the /-th traffic class in the observation horizon

OH,(k)

- is the objective loss rate determined by the agreement on the level of service defined based upon the packet loss probability with reference to the /-th traffic class, indicated PLP ^*: _{( [k )} ^{PLPj u}' ^ ^dt where a j) is the input transmission speed measured related to the /-th traffic class on the observation horizon OH,(k);

- ^Tk is the size of the observation horizon OH,(k),

and the summation considers the contribution of the /-th flow to the "dimension" of ^Nr_k periods of occupation (bp) of the /^'-th buffer.

A period of occupation (bp) is a period of time in which the buffer is not empty. The length of the period of occupation is, for the case PLP, the difference calculated between the last loss of the j-th service class during the period of occupation of the buffer and the instant in which the occupation time begins. The operator of "almost equality" (≡) indicated in the aforementioned equation is motivated by the fact that the equality is confirmed by the re- suits of the analysis of the infinitesimal perturbations only in the case in which there is a single traffic class. Recent results in literature confirm that the application of the same equation in the case of multiple traffic classes is reasonable and in any case efficient in carrying out bandwidth allocation operations.

A formulation based upon the gradient that can be compared with the previous one can be obtained with respect to the performance in terms of average delay (AD), and it can be found in scientific literature.

Further, more conventional, approaches are of course possible for the control law F(-). For example, control laws of the proportional integrative derivative type (PID) can be applied . A PID law is, for example:

(·, * )) = F_PID (; ej (; k )) = + ^■ '^ O + ^ · '*, (·)

where e_j(-) is the proportional component of the PID, ^de_j{-) is the derivative component, and 'e,( -) is the integrative component, and co_p, coj and ω, are the related tuning parameters used so as to optimize the behaviour over time of the PID law dependent upon the specific application of interest.

Actually, there is a considerable amount of literature in the scientific field on the applications of a proportional integral derivative control and the relative optimization of the parameters, even in the case of bandwidth allocation. Therefore, we shall not refer to further details concerning this particular control law in the rest of this description.

It is however interesting to point out the fact that the choice of the control law PID is actually mandatory for complicated metrics, like for example the jitter, for which no gradient formulation is possible.

It should also be noted that the application of a control law based upon the analysis of the infinitesimal perturbations ensures better performance with respect to a control law PID since the law of analysis of the infinitesimal perturbations is a tool that is capable of obtaining the exact minimization of the error for example through the sequence based upon the aforementioned gradient, whereas the law PID is simply a heuristic law applied so as to minimize the error e,(-,k).

For both cases it is well known that if the processes related to the input rate to the buffer are ergodic, the constraints {?oS, \ ...^, 0oS,^ _are predetermined (at least before the convergence) and other critical conditions are not encountered (for example, the decreasing behaviour of the gradient step size in the case of analysis of infinitesimal perturbations) the aforementioned control law converges at the exact value of θ' . This means that the required service level agreements are satisfied with the minimum amount of allocated bandwidth, corresponding to θ' .

For the sake of completeness, we specify that other control laws that are not directly dependent upon the error e_j(

can be applied, such as for example the following:

0/ (* + ! ) = F(-); e" (k + \ ) = Μαχ ϊθ, (* + 1), ..., <9, (k + \), ..., 0, (k + \)

= ^FE≠ = n>,^■ (*) + d^■ σ, (k), d = ^-2 HPLP; _iH ) - H2n)

where E^sO) is a typical equivalent bandwidth (Equivalent Bandwidth, EqB) method which can be applied in this context in the case of PLP. nij(k) and a,(k) are respectively the average and standard deviation of the input rate process of the i-th queue on the observation horizon OHj(k) and PLP*E_qB is the most stringent PLP requirement between the traffic classes of layer L3 conveyed over the i-th queue.

The EqB algorithms, like that shown above, converge at a precise value of bandwidth necessary to support the required quality of service.

Other approaches are still possible in selecting the control law F(-). For example, neural or fuzzy methods that are suitable for supporting a self-learning method for estimating Θ, could be used. In this perspective it is useful to point out that the purpose of the invention is the definition of a reliable control scheme of the bandwidth resource based upon measurements carried out at layer L2 of the protocol stack. The base logic of this choice is that analytical tools for obtaining expressions in the closed form, which are indicative of the performance in terms of quality of service L3-on-L2 are not currently available. The control must therefore be self-adapting (or based upon self-learning) with respect to traffic changes.

From this perspective all the aforementioned control laws allow for good support for self- adaptability. More in general, any algorithm based upon measurements and aimed at providing a precise estimation of the parameter Θ^* _1S useful in this context and can be applied in the scope of the present invention.

However, it is worth taking into account the computational workload required by the chosen control law F(-). Computational workloads of the aforementioned control laws are particularly small, specifically in the case of approaches by analysis of infinitesimal perturbations and PID. In the case of an EqB approach the computational workload depends upon the specific algorithm selected to estimate the average and standard deviation used by •/¾ζ?(· which, in order to keep the description brief, are not specified any further.

In a variant embodiment of the invention, the measurements of the quality of service L3- over-L2, used to calculate the error amounts in the aforementioned method, can be representative of the real levels of quality of service L3-over-L2 obtained by the traffic flows along the entire chain up to the destination, or on a sub-portion of the network path towards its destination. These levels are obtained through a measuring mechanism and are transferred to the entity for calculating the algorithm represented in fig. 6 through a suitable indication scheme that provides a periodic verification from the external network.

Fig. 7 concerns the algorithm used by each decision maker module DM, to calculate the point of stabilization in calculating the parameter Θ^* and to update the resource manager of layer L2 in the case in which there is imminent congestion of the i-th queue or the possibility of bandwidth release in the case in which there is small traffic load. The algorithm is repeated at the end of each observation horizon. It should be remembered that the actual bandwidth allocation of the i-th queue is indicated by the parameter #,^Ata , whereas Θ^* exclusively represents the current estimation of the exact bandwidth need necessary in order to satisfy the service level agreements of the traffic flows at layer L3, which are forwarded towards the i-th queue.

Concerning now the stabilization of the calculation of Θ* the steady state value of θ' is fixed when the following condition:

is met, wherein ε, is the stabilization threshold.

A reasonable value of ε, is ε, = _ο'^'" · This means that Θ* is in steady state if it has small oscillations, for example smaller than _o'^'" between two successive observation horizons. A more sophisticated stabilization condition can of course be used in the context of the present invention, if required.

As far as updating the resource manager of layer L2 is concerned, in the case in which there is imminent congestion or bandwidth release, two thresholds are defined, respectively and #,^Α'"^Λ ΑΛ»™ . Reasonable values of ^A"P and ^δ<Λ»,„ are respectively 70% and 30%.

This means that when the condition (i) ^θ,≥ ^θ, ^1αχ · Δ_ιφ j_s met, the decision maker DM, must inform the main processing module MAIN of the resource manager of layer L2 that the i-th queue is stressed by an excessive traffic load. When on the other hand the condition (ii) Θ" < 6>^{Α /}"^Λ · Δ_ίΛ,„„ is found the decision maker DM, must inform the main processing module MAIN of the resource manager of layer L2 that the current bandwidth allocation of the i-th queue is excessive, i.e. "excessively" oversized.

When both the conditions are not met, it means that the current bandwidth allocation of i-th queue is correct since it maintains the correct safety margin with respect to the calculated bandwidth need #' so as to prevent congestion in the case in which there is a rapid increase in traffic, without however incurring excessive oversizing. Finally, figs. 8 and 9 concern signalling communication between entities from the system according to the invention. Specifically, fig. 8 represents the communications between each decision maker DM, and the main processing module MAIN of the resource manager of layer L2, and fig. 9 represents the communications between the resource manager of layer L2 and the resource manager of the upper layer L3.

The communication between each decision maker DM, and the main processing module MAIN of the resource manager of layer L2 is ensured by the definition of suitable communication primitives inside layer L2, which are very simple and include the following functions:

- in condition (i):

Signal _of_L2internal _PreCongestionNotification{i , ^θ, );

- in condition (ii):

Signal _of _OverProvisioning(i , ).

The first function is used in the case in which condition (i) is met, whereas the second in the case in which condition (ii) is met, and both are triggered by a request from the relative decision maker DM,.

In the case in which the main processing module MAIN of the resource manager of layer L2 receives a message:

Signal _of_L2internal_PreCongestionNotification{i, Θ, ) from the decision maker module DM,, it can select one from three possible actions:

- increase the service rate (the allocated bandwidth) of the i-th queue (first case indicated in the figure) by a reasonable amount of bandwidth according to a specific policy. For example, a reasonable policy is an increase dictated by the formula where ^Δ/»««»ν =30%. However, it will be clear to a man skilled in the art that other policies, even more sophisticated ones, are possible and can be practically applied, and in general can be in the range of 0.1 and 0.9;

- reassign the internal addressing scheme of the traffic flows between layer L3 and the layer L2 (second case illustrated in the figure) according to a specific policy, in order to decrease the traffic load of the i-th queue. In this case the successive bandwidth need value θ, , in the successive observation horizon of the decision maker DM, must be lower with respect to the current one, or it can be unvaried as long as it is sustainable;

- alert the resource manager of layer L3 (third case illustrated in figure 8 and also in figure 9). In detail, the resource manager of layer L2 shall attribute, to the resource manager of layer L3, the responsibility of limiting the traffic load towards the i-th queue at the layer L2. In this case the communication primitive between the two layers takes up the form:

Signal _of_L2toL3 _PreCongestionNotification() where the called parameters are (TrafficClasses, θ,^Μαχ).

The first parameter of this function of communication concerns the identification of the vector of traffic classes at layer L3 that produces the potential congestion, the second parameter is the current value of the bandwidth assigned to the i-th queue, the traffic load of which is considered excessive. Consequently, the resource manager of layer L3 recognises the traffic classes that generate a state of pre-congestion at layer L2 and can act so as to limit the traffic towards layer L2 according to a predetermined intervention policy. The specification of such an intervention policy is however outside the scope of the present invention, since different choices can be made at layer L3 as a function of the specific service quality scheme applied by the control plane of layer L3. For example, in the case in which the control plane of layer L3 uses a resource reservation protocol, the resource manager of layer L3 can prevent some links of the traffic classes communicated by the resource manager of layer L2. The number of links to be prevented is derived as a function of the current value of #, ^to . The prevention policy applied by the resource manager of layer L3 also departs from the scope of the present invention.

It should be noted, however, that some actions should be carried out from layer L3, at least at the planning level if no reservation protocol is available to support actions in real time, for example when a typical DiffServ methodology is used. If no actions have been taken by the resource manager of layer L3 after receiving the communication Signal _of_L2toL3 JPreCongestionNotificationQ

there can be a possible congestion in the immediate future which could cause the agreement on the predetermined level of service to not be met.

The order of priorities at the resource manager of layer L2 in the choice between the aforementioned actions should be that shown above. The base logic is that these actions determine modifications that impact in an increasingly larger way on the current structure of the data plane of layer L2 or even at layer L3 (in the case of the last action). In the attempt of minimising the variations in operative conditions, the scale of priorities is thus that indicated, so that for example the last proposed action should be carried out only in the case in which the previous actions cannot be applied for some reason. Different policies can be applied by the resource manager of layer L2 to decide if the first actions can be applied. For example, the resource manager of la er L2 can ensure the following inequality

where N is the number of queues at layer L2 and C is the maximum channel capacity.

If the activation of the first action at a queue of layer L2 leads to the failed achievement of the previous condition, the resource manager of layer L2 cannot do anything except to apply the second action to such a queue.

Other conditions to select from the three actions mentioned above can be reasonably applied in the context of the present invention.

In the case in which the main processing module MAIN of the resource manager of layer L2 receives a message

Signal _of_Over Provisioning(i , Θ, )

from the decision maker DM, (as indicated at the bottom of fig. 8) it can apply some kind of model of bandwidth release on the i-th queue, for example of the type

with ^ ™«, =30%. Of course, as mentioned also concerning the previous cases, more sophisticated reallocation policies can be applied in the context of the present invention, and in general ^A,/«nw can be in the range of 0.1 and 0.9.

Finally, it should be noted that other primitives can be defined so as to support a recognition mechanism by the main processing module MAIN of the resource manager of layer L2 and of the decision maker module involved after notification of the messages

Signal _of_L2internal _PreCongestionNotification{) and

Signal _of Over Provisioningi)

or by the resource manager of layer L3 (L3RM) and by the resource manager of layer L2 (L2RM) after receiving a message

Signal j)f_L2toL3_PreCongestionNotification().

However, it is believed that a recognition mechanism of this type is not mandatory, even if every specific form of recognition primitives is coherent with the bandwidth allocation control scheme defined according to the present invention.

It should be noted that the proposed embodiment for the present invention in the previous discussion is purely given as an example and not for limiting purposes. A man skilled in the art can easily implement the invention with different embodiments which however do not depart from the principles outlined and that are therefore covered in the present patent.

This is particularly valid for the possibility of applying the method and the system according to the present invention in a variant embodiment in which the gateway device interconnects two portions of the same communication network where different methodologies are used to ensure the quality of service (IntServ on DiffServ, IntServ on MPLS, DiffServ on MPLS, and so on). In this case the mapping operation only consists of a modification in the aggregation balance of the traffic at layer L3 according to the interconnected QoS schemes. The gateway device acts as an edge router, for example between two portions of terrestrial network, and encapsulation or counter-fading mechanisms are not applied, but it exclusively provides a process of aggregation of traffic flows at layer L3. This essentially means that on the side of the network some traffic classes are managed separately, whereas on the opposite side of the gateway they are combined together.

The present invention thus solves in general the problem of allocating the bandwidth to a traffic flow in every embodiment in which aggregation operations of the traffic intervene at an access point to a communications network or to a portion of network based upon a technology that is different from the (portion of) the network of origin of the traffic or for which different quality of service schemes to separate or aggregate traffic classes are provided.

A system for controlling the bandwidth allocation to the traffic flows conveyed at an access point of a communications network according to the invention can be made by a combination of hardware devices or software processing modules that are adapted to execute a computer program, or be carried out in an entirely hardware form or in an entirely software form. A computer program can comprise one or more code means including instructions stored on a material support, as for example a support that can be read by a processor, possibly a removable support that can be transported (a hard disk, a CD-ROM, a ROM memory support, and the like), or distributed by a server on a communication network through any desired transmitting means and carried out by a processing system. Further, the transmitting means can be a material means like for example an optical or electric communication line or a transmitting means based upon wireless communication methods (microwaves, infrared or other transmitting methods). The code means comprising instructions for the processor can achieve all or part of the functionalities described previously. Of course, such instructions can be written in any one of the programming languages adapted to be used with any architecture of a processing system or of any operating system.

For example, when actuating the aforementioned method, various operations can be made by different entities involved in different layers of the preselected protocol stack, these operations can be actuated via software through a computer program resident on an area of memory that can be read by a processor, distributed at each of such entities.

Of course, without affecting the principle of the invention, the embodiments and the details of its actuation can be widely varied with respect to what has been described and illustrated purely as a non-limiting example without for this reason departing from the scope of protection of the invention defined by the attached claims.

Claims

1. A method for managing heterogeneous traffic flows accessing a communications network (WN), wherein an access node (G) of the network operates an aggregation of traffic flows iff) in at least a real transmission queue (Q,^L2) at a predetermined protocol layer (L2) of the network, including allocating a predetermined bandwidth ( 6^,,^Mm ) to said transmission queue (Q,^L2) in order to guarantee a related quality of service (QoS,^u) established according to a service level agreement;

characterised in that it comprises:

generating a virtual transmission queue (On -¹²) including a current replication of the aggregated traffic flow {f, ) at the real transmission queue (Q ²);

measuring at least a parameter representative of the current quality of service {OoSj ^oL2) obtained on the virtual transmission queue (Q,r^L2) in a plurality of predetermined successive observation horizons (OH,(k));

dynamically estimating the bandwidth need ( θ" ) of the virtual transmission queue ( Qi i ) complying with the quality of service as a function of the value assumed by said parameter in each observation horizon (<9H( )); and

modifying the bandwidth ( θ, ^1αχ ) allocated to the real transmission queue (Q,^L2) for the subsequent observation horizon {OH(k+ l)) and/or limiting the traffic flow {ft²) towards the queue (Q,^L2\ according to the estimated bandwidth need ( Θ^* ).

2. A method according to Claim 1 , wherein said parameter representative of the quality of service is selected from a group of parameters comprising the admissible data packet loss rate, the admissible delay in data packet transmission, the admissible jitter in data packet transmission.

3. A method according to Claim 1 or 2, wherein estimating dynamically the bandwidth need (θ') in a predetermined observation horizon {OH,{k)) includes calculating the error (£(·,£)) between the value of the parameter representing the current quality of service (QoS,^L3oL2) and the value of the parameter representing the quality of service (QoS,^u) required by the service level agreement.

4. A method according to any of the preceding claims, comprising determining the bandwidth need (d'(k + ])) estimated for a subsequent observation horizon (OHi(k+J)) according to the general expression:

e!(k + \)= F(-,e,(;k)) 0'(k + \) = Μαχϊθ, (k + \),...,9, (k + (A+l) where F(-) is a predetermined law for estimation of the bandwidth and Max[ ] operates the choice of the highest value of the bandwidth need (0^*(* + ])) among different traffic classes at a protocol network layer (L3).

5. A method according to Claim 4, wherein the estimation law F(-) is formulated as

where ηι_< is the gradient steps size, if the parameter representing the quality of service is the packet loss probability or the average delay in packet transmission.

6. A method according to Claim 4, wherein the estimation law F(-) is formulated as

F(; e, (;k)) = F_rin (·, e, (·, A')) = ω_ρ ^■ e, (-,k) + c_d ^{■ J} e , (·) + ω,^■ 'e, (^■) wherein β/(·) is the proportional component, ^de,{-) the derivative component, and 'ø,(·) the integrative component of a proportional integrative derivative estimation law (PID), and (Op, aid and co, are the related tuning parameters.

7. A method according to Claim 3. wherein the measurement of at least a parameter representing the quality of service (QoS, ^o ) mapped on a predetermined protocol layer (L2), used for calculating the error amounts (e₇(-,&)) between the value of the parameter representing the current quality of service (OoS ^3oL2) and the value of the parameter representing the quality of service (OoS, ) required by the service level agreement, is representative of the quality of service (QoS,^u) related to the traffic flows (f, ) along a complete transmission path up to the destination node of the network.

8. A method according to Claim 3, wherein the measurement of at least a parameter representing the quality of service {QoS,^L'wL2) mapped on a predetermined protocol layer (L2). used for calculating the error amounts between the value of the parameter representing the current quality of service (QoS,^Uo ) and the value of the parameter representing the quality of service (QoS,^L ) required by the service level agreement, is representative of the quality of service (QoS,¹³) related to the traffic flows {ft³) along a portion of the transmission path towards the destination node of the network.

9. A method according to Claim 1 or 2. wherein estimating dynamically the bandwidth need ( Θ* ) in a predetermined observation horizon (OH,(k)) includes calculating the equivalent bandwidth for each observation horizon (OHi(k)) according to the general expression:

0/ (A- + l ) = (-); 0' (k + \) = Max (k + ] ), ..., #, (k + \), ..., θ, (k + \) F(-) = F,._:≠(-) = m, (k ) + d^■ σ, (k), d = ^-2 \η( ΡίΡ^_Η ) - \η(2π)

where .F(\)is a predetermined law for estimation of the bandwidth, and a,(k) are the average and standard deviation, respectively, of the input rate process of the transmission queue (0,i ^L2) on the observation horizon (OHj{k)),

is the most stringent requirement of packet loss probability among different traffic classes at a protocol network layer (L3) conveyed over the transmission queue (Q,y^L2), and Μαχ[·] operates the choice of the highest value of the bandwidth need ( 0* (k + \) ) among different traffic classes at a protocol network layer (L3).

10. A method according to any of the preceding claims, wherein estimating dynamically the bandwidth need ( 6>^* )in a predetermined observation horizon (OHj(k+ ])) includes the stabilization of the estimated bandwidth need ( Θ^* ) when the condition of convergence:

is met, wherein ε, is a predetermined stabilization threshold between two successive observation horizons (OH,{k); OH,{k+l)).

1 1 . A method according to any of the preceding claims, wherein the temporal dimension of an observation horizon (OH,(k)) is in the range [ 1 , 360]s.

12. A method according to any of the preceding claims, wherein for each real transmission queue (0,^u) an initial bandwidth ( <?*'"' ) is originally allocated which is defined on a planning basis in dependence of the available traffic forecast.

13. A method according to any of the preceding claims, comprising determining a condition of imminent congestion when finding the relation θ'≥ 6>^{Λ /}"^ν■ Α,_φ , where θ"^αχ■ Α_ιφ is a predetermined comparison threshold, where A_lip is 0.7.

14. A method according to any of the preceding claims, comprising determining a condition of bandwidth release when finding the relation θ" < 6>^A'te · Δ_(/,_ηι.„ , where 6>^Wi" Δ_ι/θΜ„ is a predetermined comparison threshold, where A_{lhu i}, is 0.3.

15. A method according to Claim 13, including, when determining a condition of imminent congestion:

increasing the bandwidth ( ef^ax ) allocated to the real transmission queue (Qt²) for the subsequent observation horizon (OH,(k+ l)) of a predetermined amount, as where A„,_{i(m t}, is in the range [0.1 ; 0.9];

or, as an alternative, in a decreasing order of priority

limiting the traffic flow (ft²) towards the queue (Q ²), according to the estimated bandwidth need ( θ' ).

16. A method according to Claim 13, including, when determining a condition of bandwidth release, reducing the bandwidth ( e, ^,m ) allocated to the real transmission queue (Q^²) for the subsequent observation horizon (OH,(k+l )) of a predetermined amount, as where A_J ._t._{a i}. is in the range [0.1 ; 0.9].

17. A method according to any of the preceding claims, characterised in that it is carried out at a network layer (L3) of a communication protocol stack performing an aggregation of heterogeneous traffic flows (ft³) without changing the encapsulation format of data packets, by mapping the predetermined quality of service (OoS,^u) between different protocols operating at the same network layer (L3).

1 8. A method according to any of claims 1 to 16, characterised in that it is carried out at a data link layer (L2) of a communication protocol stack performing an aggregation of heterogeneous traffic flows (ft³) from an upper network layer (L3) by encapsulating the data packets of each flow in a data frame assigned to a transmission queue (Q, ) intended to be conveyed on a physical transmission channel, by mapping the quality of service (goS,^ij) predetermined at an upper network layer (L3) of the protocol stack.

1 9. A system for managing heterogeneous traffic flows (fj^U) accessing a communications network (WN), comprising one or more processing modules programmed for implementing a control plane (CPu) and a data plane (UP_L2) adapted to perfonn an aggregation of said traffic flows (f, ) in at least a real transmission queue (0, ) at a predetermined protocol layer of the network (L2) for conveying the traffic flows on at least a transmission channel,

characterised in that the data plane (UPu) includes a virtual transmission queue (QiY^L~) including a current replication of the aggregated traffic flow {f ²) at the real transmission queue (Q,^L2); and the control plane (CP|_₂) includes a resource manager module (L2RM) arranged for

measuring at least a parameter representative of the current quality of service (QoSj ^oL2) obtained on the virtual transmission queue {Qn-¹²) in a plurality of predetermined successive observation horizons (OH^k));

dynamically estimating the bandwidth need { Θ^* ) of the virtual transmission queue (Qiv^L~) in order to guarantee a related quality of service (QoS,^u) established according to a service level agreement, as a function of the value assumed by said parameter; and

changing the bandwidth ( 6>^Wm ) allocated to the real transmission queue (0,^u) for the subsequent observation horizon (OHj(k+ l )), or limiting the traffic flow (f,^L2) towards said queue {Q,^u), according to the estimated bandwidth need ( Θ^" ),

the control plane (CP^) and the data plane (UPL₂) being arranged for carrying out a method of managing heterogeneous traffic flows accessing the network according to any of the claims 1 to 1 8.

20. A system according to Claim 19, wherein the resource manager (L2RM) of the con- trol plane (CP_L2) of the data link layer (L2) comprises a plurality of decision maker modules (DM,), each one being associated with a respective virtual transmission queue (Qtr^U) for estimating the related bandwidth need ( θ' ) for meeting the quality of service levels (OoSj^UoL2) for the flows (f,^u) conveyed in the corresponding real transmission queue

(Q,^uy

21 . A system according to Claim 20, wherein each decision maker module (DM,), based on the estimated bandwidth need ( Θ^* ), is arranged for communicating a message of "imminent congestion" or a message of "bandwidth release" to a main processing module (MAIN) of the resource manager (L2RM).

22. A system according to Claim 21 , wherein the resource manager (L2RM) of the data link layer (LM) is arranged for communicating to at least a resource manager of a higher layer (L3RM) that no sufficient bandwidth resources are available for supporting the required quality of service {OoS, ) with respect to the current traffic flow {ft²).

23. An access node of a communications network, comprising a system for managing traffic flows according to any of the claims 19 to 22.

24. An access node of a communications network according to Claim 23, characterised in that it includes a gateway device (G), adapted to interconnect two local and/or wide area networks (TN, WN) with different architectures, and is arranged for performing a conversion of communication protocols and for carrying out a process of aggregation of traffic flows if,¹³) at a data link protocol layer (L2).

25. An access node of a communications network according to Claim 23, characterised in that it includes a gateway device (G) acting as an edge router, adapted to interconnect two portions of the same communication network (TN; WN) where two different methodologies for supporting the required quality of service are used (QoSj^U), and is arranged for carrying out a process of aggregation of traffic flows (ft³) at a network protocol layer (L3).

26. A communications network, comprising at least an access node according to Claim 23, 24 or 25.

27. A computer program or group of programs, comprising code means including instructions for carrying out a method for managing heterogeneous traffic flows accessing a communications network (WN) according to any of the claims 1 to 18.