DE60037270T2 - OPTIMIZING THROUGHPUT CURING THROUGH JOINT CONNECTION CAPACITY - Google Patents

Description

TECHNICAL FIELD OF THE INVENTION

The The present invention relates generally to the field of Communication networks and in particular to a common benefit a connection capacity and an assignment of a connection bandwidth in such networks.

BACKGROUND

Lots Today's communication networks support a so-called elastic Traffic, like the "best effort" (English: best effort) services available in Internet Protocol (IP) based networks or the available bit rate Available bit rate (ABR) traffic provided in ATM networks are. An elastic traffic is typically for transmission a digital object, such as a data file, a web page, or a video clip, for a local play set up, which at each rate up to the Limit imposed by the connection capacity sent can be. A web browsing In particular, the Internet is a good and representative example of an elastic Traffic. The "elasticity" of traffic is here as user-perceived throughput (which is normally in transmitted bits or bytes per unit of time is given) obviously, if downloading a website, for example, depends on the overall system load varies with time.

The Services that are transported through IP-based networks and the Internet, will be "best Called effort ", since the Networks generally no guarantee of quality of service of service; QoS) received by the applications deliver. The IP network is only making a best effort to get the requested one To deliver service. When an application requests to the network IP packet from one endpoint to another, can For example, the network usually does not say what the delay is through the network for this package will be. The network really does not guarantee it yet once that the package at all promoted becomes.

Terminals that are connected to an IP network, therefore, packet losses and Handling situations of excessive packet delay. These overcrowding situations have a non-zero probability with IP-based networks, since IP networks no call admission control or connection admission control (English: call admission control; CAC) exercise. In other words, restrict IP networks the number of concurrently connected users is not, and if so too many users who use the network resources it therefore an overcrowding and package losses.

With however, there is the advent of real-time traffic and QoS requirements in IP networks a need to exercise a connection admission control (CAC) to the number of connections, who are present in the network at the same time.

One important aspect of call or call admission control is that for new connections arriving on the network rejected a service can be used to protect ongoing connections. CAC algorithms like those usually in a use for a rigid traffic in conventional ATM networks are deliver a basic means to the number of To control users on the network, thereby ensuring This will allow authorized users the bandwidth that is necessary get the contracted QoS. A CAC algorithm Consequently, there is a trade-off between the blocking probability for new ones Connections and the delivered throughput for ongoing connections. In other words, the more users of the CAC algorithm in the Network allows (what reduces the blocking probability), the less it gets the delivered throughput per user, as a larger number Users will share the full bandwidth and vice versa.

A recent Research stated that it makes sense to have a connection admission control even for to exercise elastic traffic Because CAC algorithms provide a means to reduce excessive throughput to prevent TCP sessions.

On the question of applying a CAC for elastic joints and providing thereby a minimum throughput for transmission control protocol (English: Transmission Control Protocol; TCP) connections on the Internet was through Massoulie and Roberts in References [1-3]. A bandwidth will agree here according to Justice criteria assigned to different users.

It was recognized by Gibbens and Kelly in References [4-5] that there is a close relationship between a throughput probability and a blocking probability for elastic traffic and that compromise is connected with the question of billing is.

It was further described by Feng et al. in reference [6] shown that a Providing a minimum guaranteed rate for elastic services is useful, because in this case the performance of the TCP protocol is optimized can.

There The Internet is made up of a packet network that is a single best-effort class of service support developed into an integrated infrastructure for several service classes, There is also a growing interest in thinking up procedures a common benefit of one bandwidth, the different ones Need for guaranteed top rate services and elastic services Fulfills.

It is similar necessary that modern ATM networks have different classes of service, like classes of a constant bit rate (English: Constant bit rate; CBR) and one available Bitrate (Available Bit Rate, ABR), supported, and it is still an open question how the connection capacity through the is to use optimally together different service classes.

It is generally from the classic circuit-switched multi-rate frame known to be the question of a common benefit of a bandwidth in the context of dynamically incoming and outgoing traffic flows and especially if users have different throughput and blocking requirements have, an extremely complex Problem is.

The U.S. Patent 5,274,644 refers to an admission scheme that allows sharing of a common resource between N traffic classes, such that each class is guaranteed a contractually agreed minimum use of the resource, and this, whenever possible, can go beyond the contract.

SUMMARY OF THE INVENTION

The overcomes the present invention these and other disadvantages of the prior art arrangements.

A The first object of the invention is a procedure a common benefit of link capacity / bandwidth, the different needs of rigid and elastic services in an environment of mixed rigid-elastic traffic.

It is particularly desirable the issues of sharing a bandwidth and blocking probabilities for one to handle elastic traffic in a common framework. A second object of the present invention in this regard therein, means for sharing a connection capacity comprising the Throughput-blocking compromise for an elastic Traffic, to create. It would be particularly beneficial an algorithm of a shared benefit of a link capacity that the Throughput-blocking compromise optimized, developed and used.

A Another object of the present invention is to provide a suitable Link level model of a transmission link that has an elastic Carries traffic, and the link level model to measure the common To use a connection bandwidth for throughput / blocking optimality.

These and other objects are achieved by the invention as defined by the appended claims, Fulfills.

The The invention relates to an efficient procedure for common Using a connection bandwidth in a mixed environment rigid-elastic traffic as well as a common mode of operation Use of a bandwidth between elastic traffic streams.

The idea according to the invention is in short, a connection capacity in a network by dividing the connection capacity into a first common part for elastic and a rigid (non-elastic) traffic and a second part dedicated for elastic traffic, ba based on received network traffic inputs. One or more admission control parameters for the elastic traffic are subsequently determined based on the allocation of a connection capacity as well as received network traffic inputs.

The Distribution of a connection capacity is generally used to link capacity between a rigid and an elastic traffic together use and in particular a part of the connection capacity for one to reserve elastic traffic. A required minimum capacity of the common Partly in relation to a rigid traffic is preferably considered a maximum allowable Blocking probability for determines the rigid traffic. In this way, a certain degree a service (English: grade of service; GoS) in the connection level for the ensures rigid traffic on the connection.

Of the or the registration control parameters required for elastic traffic are intended to serve the number of elastic traffic streams that are present on the compound at the same time. By Formulate a link level model for elastic traffic and determining a maximum number of allowable flows of elastic traffic based on joint level constraints for elastic traffic in the In connection with a throughput and / or blocking is in particular fully considered the throughput blocking compromise. The invention is in this regard capable of one Connection width between elastic connections in the sense that that minimizes blocking probabilities under throughput conditions or vice versa in the sense that the throughput is under blocking constraints is maximized, allocate optimally. In this way delivers the Invention either in terms of minimum blocking under throughput conditions or a maximum throughput under blocking conditions maximum utilization of the connection bandwidth.

A efficient way to share bandwidth in an environment of mixed rigid-elastic traffic is accordingly provided. The algorithm ensures a shared benefit of a bandwidth in particular a maximum blocking for a rigid traffic as well as a Minimum throughput and / or maximum blocking for one elastic traffic.

One important technical advantage of the invention is its ability to the different needs of a rigid traffic and an elastic To fulfill traffic.

One Another advantage of the invention is the ability to produce a predictable quality a service for to deliver both the user and the network provider while for same time a high proceeds of one Network provider guaranteed becomes.

By Consider only the elastic traffic of the total traffic in an environment of mixed rigid-elastic traffic or alternatively by reducing the common bandwidth part to zero so that the entire connection for is reserved elastic traffic, the device is the common use of the total interconnection capacity to determine a or several admission control parameters for elastic traffic reduced. An admission control for requested new elastic Connections can then be based on such an admission control parameter (s) be exercised. By minimizing the blocking probabilities with respect to the number of allowed elastic compounds under given throughput for the elastic In particular, traffic becomes excessive blocking probabilities avoided while a given user throughput is ensured.

One Another aspect of the invention relates to the use of a link level model a compound that supports elastic traffic, for To measure the common benefit of link bandwidth for throughput / blocking optimality in one Admission-tax-enabled IP network. An elastic traffic flow is especially considered as a bandwidth that is between a minimum bandwidth and a peak bandwidth during the holding time of the traffic flow varies, modeled. The elastic Traffic is also at least a minimum accepted throughput probability or a maximum accepted blocking probability connected.

One Another aspect of the invention relates to a computational method for determining a stationary Markov chain Power Distribution, the most advantageous for size state spaces is. The Markov chain describes the dynamics of a connection, the a number of traffic classes, which is a non-adaptive elastic Include traffic, carry, and the computational process supplies a good initial approach the stationary one Power distribution based on Markov chain product shape calculations.

Other Aspects or advantages of the present invention will become apparent upon reading the following description of the embodiments of the invention deliberately.

BRIEF DESCRIPTION OF THE DRAWINGS

The Invention together with further objects and advantages thereof will be described with reference to the following description the attached Drawings are best understood in which:

1 Fig. 10 is a schematic diagram of a communication network according to a preferred embodiment of the invention;

2 Figure 3 is a schematic flow diagram of the algorithm of sharing the total interconnect capacity applied in an environment of mixed rigid-elastic IP traffic according to a preferred embodiment of the invention;

3 Figure 3 is a schematic block diagram of pertinent portions of an IP message router in accordance with a preferred embodiment of the invention;

4 is a Markov chain state space diagram for an illustrative mailing system;

5 Figure 4 is a graph illustrating the mean and variance of the throughput of adaptive elastic currents as a function of their service time for an illustrative example of a mailing system;

6 Figure 12 is a schematic diagram illustrating the elastic separation parameters that satisfy given QoS requirements for an illustrative example of a connection system; and

7 FIG. 12 is a schematic flow diagram of the algorithm of sharing the total interconnect capacity for a blended environment of CBR ABR traffic in accordance with a preferred embodiment of the invention. FIG.

DETAILED DESCRIPTION OF EXAMPLES THE INVENTION

Everywhere in In the drawings, the same reference numerals for corresponding or similar Elements used.

Everywhere in the revelation will be the terms connection and power more or less less interchangeable, what is traditionally referred to as a compound used.

System overview of an explanatory Communication network

For a better understanding, a general overview of an illustrative communication network according to a preferred embodiment of the invention will now be made with reference to FIG 1 specified.

1 Fig. 10 is a schematic diagram of a communication network according to a preferred embodiment of the invention. The communication network is shown here as an IP-based network, but may be in the form of an ATM network or other type of network or combination of networks capable of supporting elastic traffic. The communication network 10 is based on a number of interconnected IP message routers 20 (ATM switches in the case of an ATM network) that make up the core network. On the core network is through different users 30 (Computer, server, etc.) are accessed through access points, a so-called user-network interface (UNI) being defined for the interaction between the IP message routers and the user equipment. A plurality of users are typically in the form of a clustering point, such as access router (AR). 40 , which acts as an intermediary between the end users and the core network.

Sharing of link capacity, further referred to as bandwidth control in the network context, and packet scheduling are normally on the network side of the UNI within the IP message routers 20 , The bandwidth control and packet scheduling are particularly preferably on the output gate side of the message routers 20 implemented. The total bandwidth control has two main functions. The bandwidth control first serves to share the bandwidth between different traffic classes. Secondly, the bandwidth control serves to limit the number of concurrently active connections within the traffic classes. The latter function will hereinafter be referred to as call admission control (CAC) and will normally be on the input gate side of the IP message routers 20 in which connections are accepted or rejected according to a CAC algorithm. For example, the overall bandwidth control algorithm that includes the CAC algorithm may be implemented as hardware, software, firmware, or a suitable combination thereof.

Of the User-network contract (UNC) is typically defined at UNI. The UNC usually gives the QoS to which the user is entitled, and also the specification of the Traffic to the user along with supplemental data in the network can contribute. The supplementary data can for example, the time of day, during which the user has access to a particular service, etc. For example, the UNC may specify that not more than 1% the user introduced IP packets (or ATM cells) through the Network can be lost and that the user during any Second through the UNI 10 Mbits can send.

Of the CAC portion of the bandwidth control algorithm can extract the UNC information from several users to provide a basic means to the number of concurrent users in the network to control, thereby ensuring Will allow the allowed users the bandwidth that is required is received to deliver the contracted QoS. Of the CAC algorithm makes a trade-off between blocking probabilities and the delivered throughput; the more users the CAC algorithm into the network (which reduces the blocking probability), the smaller is the delivered throughput per user, as a larger number users to share network bandwidth.

One CAC can by means of classical signaling exchange, the for example, from conventional circuit-switched communication networks is known implemented become. If the majority of the elastic currents in the network are short, as with many TCP streams on the internet nowadays, however, may result from the introduction of the classic signaling exchange to an admission control perform, a large Additional expenditure. An in-run decision, to accept or reject the first packet of a stream, like is recommended in reference [3] would therefore be more favorable. For this reason is a facility based, for example, on the existing resource reservation protocol Resource Reservation Protocol (RSVP) is based to retain an overview of identities of currently active streams and for classifying packets according to these identities, as these and when they arrive, provided. To determine if a Electricity is new or not, it is sufficient to its identification with the streams on a specific list of active currents. If not Package for a particular stream within a given timeout period was received, would the Power from the list of active currents be removed. An admission control is preferably determined by determining a maximum number of allowed streams based on the prevailing traffic situation in the system and setting accordingly the size of the list active currents realized. When the list is complete, a new stream is created rejected. The stream is otherwise accepted and added to the list entered.

Algorithm of the common use of the Connection capacity - the IP network example

One Algorithm of sharing a connection capacity according to a preferred embodiment is described below with reference to the particular application of a IP-based network that has a rigid as well as an elastic Carries traffic, described.

It is first necessary that a proper link level traffic model be formulated. The application of classic multi-rate link level models for resilient traffic management, such as best effort traffic on IP networks or ABR traffic on ATM networks, is unfavorably unfavorable. For example, it is not possible to connect elastic traffic with a constant bandwidth. The bandwidth caused by elastic traffic instead, time varies depending on the current load on the link and the scheduling algorithm and rate control algorithm applied to the network nodes. The term blocking when applied to elastic traffic flows is not as simple as for rigid traffic because an incoming elastic stream could come into service even if there is no available bandwidth at the time of arrival. For many services, moreover, the actual residence time of an elastic current depends on the throughput actually received by the elastic current.

Multi-class model of a transmission connection for one mixed rigid-elastic traffic

One Example of a feasible Markov's model of a broadcast connection, which is both top-bandwidth-secured (rigid or not elastic) as well as elastic traffic classes serves, is formulated below. For a simplicity and a Clarity are just a single rigid traffic class and considered two elastic traffic classes. It should be noted that the model as well as the attached Algorithm of sharing a connection to more general Cases and Of course even simpler cases can be extended.

• Klasse 1 – Ströme einer starren Verkehrsklasse, die durch ihre Spitzenbandbreitenanforderung b₁, ihre Stromankunftsrate λ₁ und ihre Abgangsrate μ₁ gekennzeichnet sind.
• Klasse 2 – Ströme einer adaptiven elastischen Klasse, die durch ihre Spitzenbandbreitenanforderung b₂, ihre Mindestbandbreitenanforderung b₂ ^min, ihre Stromankunftsrate λ₂ und ihre Abgangsrate μ₂ gekennzeichnet sind. Obwohl die Bandbreite, die durch adaptive elastische Ströme belegt ist, als eine Funktion der Verbindungslast schwanken kann, wird ihre wirkliche Haltezeit während ihrer ganzen Residenz in dem System durch den empfangenen Durchsatz nicht beeinflusst. Dies ist der Fall zum Beispiel mit einem adaptiven Video-Codec, der bei einem Fall einer Durchsatzverschlechterung die Qualität der Videobilder reduziert und dadurch weniger Bandbreite belegt.
• Klasse 3 – Ströme einer nicht adaptiven elastischen Klasse, die durch ihre Spitzenbandbreitenanforderung b₃, ihre Mindestbandbreitenanforderung b₃ ^min, ihre Stromankunftsrate λ₃ und ihre ideale Abgangsrate μ₃ gekennzeichnet sind. Die ideale Abgangsrate wird erlebt, wenn die Spitzenbandbreite verfügbar ist. Die reale augenblickliche Abgangsrate ist proportional zu der Bandbreite der Ströme.

The system under consideration has a transmission link of capacity C, which may be considered as an example as an integer in an appropriate bandwidth unit, say Mbps. In this example, calls that arrive at the call generally belong to one of the following three traffic classes:

• Class 1 - flows of a rigid traffic class characterized by their peak bandwidth requirement b ₁ , their power arrival rate λ ₁ and their leaving rate μ ₁ .
• Class 2 - Adaptive elastic class streams characterized by their peak bandwidth requirement b ₂ , their minimum bandwidth requirement b ₂ ^min , their current arrival rate λ ₂ and their rate of departure μ ₂ . Although the bandwidth occupied by adaptive elastic currents may vary as a function of the connection load, its actual hold time throughout its residence in the system is unaffected by the received throughput. This is the case, for example, with an adaptive video codec which, in a case of throughput degradation, reduces the quality of the video images and thereby occupies less bandwidth.
• Class 3 - flows of a non-adaptive elastic class characterized by their peak bandwidth requirement b ₃ , their minimum bandwidth requirement b ₃ ^min , their power arrival rate λ ₃ and their ideal rate of departure μ ₃ . The ideal exit rate is experienced when the peak bandwidth is available. The real instantaneous rate of departure is proportional to the bandwidth of the streams.

We denote the real bandwidth allocated (reserved) to a stream of class-2 and class-3 in a given system state, with b ₂ ^r and b ₃ ^r , both of which vary in time as streams arrive and depart. We will also use the quantity r _min = b _i ^min / b _i (for i = 2 or i = 3) associated with elastic streams with minimum bandwidth requirements.

abgetastet wird, was bei dem Fall, wenn die Spitzenbandbreite b3 während der gesamten Dauer des Stroms verfügbar ist, eine exponentiell verteilte Dienstzeit mit einem Mittelwert 1/μ₃ verursacht. Da die freie Kapazität der Verbindung gemäß der augenblicklichen Anzahl von Strömen im Dienst hinsichtlich der Zeit schwankt, kann die Bandbreite, die nicht adaptiven elastischen Strömen gegeben ist, unter die Spitzenbandbreitenanforderung fallen, wobei sich bei diesem Fall die wirkliche Haltezeit des Stroms steigert.One can think of a stream of non-adaptive elastic class as a stream having on arrival an associated amount of data (W) to be transmitted, which consists of an exponentially distributed service request with a distribution

is scanned, which in the case when the peak bandwidth b3 is available throughout the duration of the current, causes an exponentially distributed service with a mean 1 / μ. ₃ Since the free capacity of the connection varies according to the instantaneous number of streams in service with respect to time, the bandwidth given to non-adaptive elastic streams may fall below the peak bandwidth requirement, in which case the actual hold time of the stream increases.

All three types of currents come in accordance with independent Poisson processes on, and the holding time for the rigid and adaptive streams is distributed exponentially. As we will see, the Moments of hold time of non-adaptive currents using theory Markov reward processes be determined. Two types of elastic traffic become short considered. An elastic traffic is with both a top- and a minimum bandwidth requirement and is in the Service only allowed as long as how the minimum bandwidth requirement is met. The two elastic traffic classes differ mainly in terms of how their residence time depends on the acquired throughput.

In order to ensure a given QoS of the different elastic classes (which generally differ in their peak and minimum bandwidth, that is, b ₂ ≠ b ₃ , b ₂ ^min ≠ b ₃ ^min ), we must have a common strategy Benefits of bandwidth between the different elastic classes generally regulates, set up. For this reason, we define the following general rules of sharing the bandwidth between the elastic classes. The following illustration concerns only two elastic classes, but of course this extends to more than two elastic classes.

If there is enough bandwidth for all streams to get their respective peak bandwidth requirements, then class 2 and class 3 streams each occupy b ₂ and b ₃ band width units.

If there is a need for bandwidth compression, that is n ₁ * b ₁ + n ₂ * b ₂ + n ₃ * b ₃ > C, then the bandwidth compression of the elastic streams is such that r ₂ = r ₃ , where r ₂ = b ₂ ^r / b ₂ and r ₃ = b ₃ ^r / b ₃ , as long as the minimum rate requirement for both elastic classes is met (that is, b ₂ ^min / b ₂ ≤ r ₂ ≤ 1 and b ₃ ^min / b ₃ ≤ r ₃ ≤ 1).

If there is still a need for further bandwidth compression, however, none of the two elastic classes tolerate further bandwidth reduction at the time of arrival of a new stream (that is, r _i is already b _i ^min / b _i for either i = 2 or i = 3), then the class of service tolerating further compression will immediately reduce the bandwidth occupied by its streams as long as the minimum bandwidth requirement is maintained for that class of traffic.

Three underlying assumptions of the previous exemplary model are remarkable. It is first of all assumed that both types of elastic currents are greedy in the sense that they always have the maximum possible bandwidth on the connection, the smaller of their peak bandwidth requirement (b ₂ or b ₃ ) and the same part (in the preceding sense) of the Bandwidth left over by the rigid currents for elastic currents (which will depend on the connection assignment strategy used) is proven. Second, it is assumed that all current elastic currents share equally the available bandwidth among them (that is, the r _i 's are equal), that is, the newly arrived elastic current and the current elastic currents are equal to the same r _i value squeezed. This assumption, as we shall see, provides a fairly "fair" sharing of resources between the elastic streams, so to have different elastic traffic classes with a significantly different QoS, it is necessary that this assumption be modified.

If a new incoming stream has reduced the elastic current bandwidth below b ₂ ^min and b ₃ ^min (that is, both elastic classes are compressed to their respective minimum), that stream will not be allowed into the system, but will be blocked and gets lost. It is permissible that incoming rigid as well as elastic currents "compress" the elastic currents in the service as long as the minimum bandwidth requirements are kept In a third point the model assumes that the rate control of the current elastic currents is ideal in the sense that an infinitesimal amount At some time after a system state change (i.e., power arrival and departure), the elastic traffic sources reset their current bandwidth on the link While this is clearly an idealizing assumption, the buffers in the IP packet layer could be made large enough to Absorbing Packets Until TCP Stops the Stations The fact that the model always adopts an immediate increase in source rate when possible makes the upcoming throughput and blocking calculations conservative and not optimistic.

definiert den Durchsatz eines adaptivenStroms mit einer Haltezeit, die gleich t ist.

(Zufallsvariable) definiert den Durchsatz eines adaptiven Stroms, wobei F(t) die exponentiell verteilte Haltezeit ist.

(Zufallsvariable) gibt die Zeit, die das System braucht, um eine Menge x and Daten durch einen elastischen nicht adaptiven Strom zu senden.It is intuitively clear that the residence time of the non-adaptive elastic currents in this system depends not only on the amount of data they want to send, but also on the bandwidth that they receive during their hold times, and vice versa, the amount depends Data sent by an adaptive elastic stream depends on the received bandwidth. To specify this relationship, define the following quantities:
θ ₂ (t) and θ ₃ (t) define the instantaneous throughput of each adaptive and non-adaptive streams at a time t. For example, if there are rigid, adaptive, and non-adaptive streams in the system at time tn ₁ , n _2, and n _3, respectively, the instantaneous throughputs for adaptive and non-adaptive streams are min (b ₂ , (C-n ₁ b ₁ - n ₃ r ₃ b ₃ ) / n ₂ ) and min (bar (C - n ₁ b ₁ - n ₂ r ₂ b ₂ ) / n ₃ ). It should be noted that θ ₂ (t) and θ ₃ (t) are discrete random variables for all t ≥ 0.

defines the throughput of an adaptive stream with a hold time equal to t.

(Random variable) defines the throughput of an adaptive stream, where F (t) is the exponentially distributed hold time.

(Random Variable) gives the time the system takes to send a set of x and data through an elastic non-adaptive stream.

(Zufallsvariable) definiertden Durchsatz eines nicht adaptiven Stroms, wobei die Menge an gesendeten Daten mit einem Parameter μ₃/b₃ exponentiell verteilt ist.θ ^ = x / T _x defines the throughput of a non-adaptive stream during the transmission of x data units. It should be noted that θ _{x is} a continuous random variable.

(Random variable) defines the throughput of a non-adaptive stream, where the amount of data sent is exponentially distributed with a parameter μ ₃ / b ₃ .

Even though a number of general rules of bandwidth sharing defined above, it is still necessary that a more specific strategy for sharing a Link capacity especially one that the different requirements of a rigid and an elastic traffic, is shown.

Algorithm of a common benefit a connection capacity

According to the invention, the partial allocation overlap policy (POL), which is known from reference [7], which describes the POL strategy for simulative analysis in the classical circuit-switched multi-rate framework, is adopted and for a traffic environment that modifies elastic traffic. According to the new so-called elastic POL strategy, the connection capacity C is divided into two parts, a common part C _COM for a rigid and elastic traffic, and a dedicated part C _ELA reserved only for the elastic currents such that C = C _COM + C _ELA .

Registration control parameters, one for each elastic traffic class present in the system, are further introduced into the new Elastic POL strategy. In this particular example, N _{EL2 designates} the admittance _{control parameter} for adaptive elastic currents, and N _EL3 designates the admittance _{control parameter} for non-adaptive elastic currents. Each admission control parameter represents the maximum number of permissible flows of the corresponding elastic traffic class. The admission control parameters are further referred to as separation parameters, as long as the maximum number of simultaneous elastic flows of a particular class are present on the connection, new elastic flows are rejected; a form of separation.

Under the considered elastic POL strategy, the number of (n ₁ , n ₂ , n ₃ ) running currents on the link is subject to the following constraints: n 1 · b 1 ≤ C COM (1) N EL2 · b 2 min + N EL3 · b 3 min ≤ C ELA (2) n 2 ≤N EL2 (3) n 3 ≤N EL3 (4)

In (1) the elastic currents are protected by rigid currents. In (2-4), the maximum number of elastic currents is limited by three constraints. The expression (2) protects rigid flows from elastic currents, while (3-4) protects the running elastic flows from incoming elastic currents. The new elastic POL strategy is completely determined by the sharing of the link _capacity specified by C _COM and the admission _{control parameters} N _EL2 , N _EL3 . On these parameters is called the off reference to the system parameters. The performance of the elastic POL strategy may be tuned by the output parameters and, in particular, it has been realized that setting the output _parameters C _COM , N _EL2 and N _{EL3 allows} tuning of the throughput blocking tradeoff for the elastic traffic classes.

In Terms of throughput blocking compromise for one elastic traffic, the invention is general to the provision a high use of the connection bandwidth out under one or multiple link level constraints that comply with at least either an elastic probability of passage or an elastic blocking probability related, directed.

According to a preferred embodiment of the invention, the algorithm of sharing the connection capacity aims at setting the output parameters of the elastic POL strategy in such a way as to allow connection jamming probabilities B ₂ and B ₃ for elastic flows while being able to provide a jam probability constraint (FIG. GoS) for the rigid flows as well as minimum throughput requirements for the elastic currents to minimize. Flow rates for the elastic flows have been introduced because it has been recognized that there is a minimum acceptable throughput below which users will not really benefit.

The rigid traffic class is therefore associated with a maximum accepted link blocking probability B ₁ ^max , and the elastic adaptive and elastic non-adaptive traffic classes are associated with respective minimum accepted throughputs θ ~ _min and θ ^ _min . The throughput pads are preferably provided as constraints on the likelihood that the throughput perceived by a user will fall below a given threshold during transmission of a particular amount of data. Such a measure of performance is easier for the user to verify than the traditional fairness criteria discussed in References [1-3].

Nevertheless, although the likelihood of likelihood of elastic traffic being minimized, it is normally advisable, though not necessary, to have a worst case guarantee of connection blocking for elastic traffic and further the two elastic traffic classes with maximum allowable blocking probabilities B ₂ ^max and B ₃ ^max connect.

In this case, the traffic input parameters of the system are the set of arrival rates (λ ₁ , λ ₂ , λ ₃ ) and outbound rates (μ ₁ , μ ₂ , μ ₃ ) obtained from the network, the bandwidths (b ₁ , b ₂ , b ₃ ), the requirements of a minimum elastic band width (b ₂ ^min , b ₃ ^min ), the blocking probability constraints (B ₁ ^max or the whole set of B ₁ ^max , B ₂ ^max and B ₃ ^max ) and the elastic flow rate constraints (θ ~ _min and θ ^ _min ). The rate of departure for a non-adaptive class may be estimated assuming that the bandwidth of the non-adaptive streams is equal to b ₃ .

The Parameters and performance measures, those with the rigid traffic class and the two elastic traffic classes are summarized in Table I below.

Table I

The problem of determining the output parameters of the elastic POL strategy under blocking and flow conditions is discussed below with reference to FIG 2 Outlining a schematic flow diagram of the overall interconnect capacity sharing algorithm according to a preferred embodiment of the invention. At the first step 101 provide the required input parameters, such as actual arrival and departure rates, bandwidth requirements and constraints imposed on resilient traffic.

At step 102 the GoS (connection blocking) request for rigid traffic is guaranteed by the proper setting of C _COM . Specifically, we determine the minimum required capacity of C _COM for rigid streams that guarantees the required blocking probability B ₁ ^max : min {C COM : B 1 ≤ B 1 Max } (5) where B _{1 is} the blocking probability of rigid streams. For example, the well-known Erlang B formula may be used to estimate such a value of C _COM based on arrival and departure rates and a peak bandwidth requirement for the rigid traffic as inputs. A maximum number of permissible N _COM rigid streams may additionally be determined based on the Erlang-B-analysis and used for admission control of the rigid traffic.

We must next determine a maximum number of elastic flows (N _EL2 , N _EL3 ) that can be present in the system at the same time as the required throughput and blocking requirements are met. It is intuitively clear that as the maximum number N _{EL2 of} adaptive elastic currents is increased, the blocking _probability B _{2 of} adaptive elastic currents is reduced and the throughput is also reduced. Changing N _EL2 adversely affects both the blocking _probability B ₃ and a throughput of non-adaptive elastic currents, and vice versa.

In this particular embodiment, the interconnection capacity sharing algorithm aims at minimizing the congestion probabilities of the elastic traffic classes below throughput thresholds. To achieve this, the invention proposes an iterative procedure that steps through 103 - 107 is generally defined, to tune the separation parameters so that the throughput thresholds are met, no more and no less. At step 103 First, initial values of the separation parameters are estimated. The performance of the system is next analyzed for elastic throughputs (step 104 ). The flow rates θ ~ and θ ^ offered in the system controlled by the initial values of the separation parameters are specifically analyzed (step 104 ) and related to the flow rate thresholds θ ~ _min and min brought (step 105 ). If the offered throughputs are too low then the separation parameters are reduced (step 106 ), whereby the blocking probabilities are increased and also the throughputs are increased. On the other hand, if the offered throughputs are greater than the throughput thresholds then the separation parameters can be increased (step 107 ), so that the blocking probabilities (as well as the throughputs) are reduced. This way you can repeat iteratively by repeating the steps 104 . 105 and 106 / 107 the blocking probabilities are minimized while the throughput requirements are still adhered to. Once the requirements have been met to a satisfactory degree, there are (step 108 ), the algorithm _evaluates the parameters C _COM , (C _ELA ), (N _COM ), N _EL2 , N _EL3 for use in controlling the true common benefit of the bandwidth of the considered transmission _link .

The steps 101 to 108 are, of course, repeated in response to changing traffic conditions in order to flexibly adapt the shared benefit of bandwidth to the prevailing traffic situation.

The Separation parameters must generally be reduced to meet the throughput requirements. Since the goal is to use the elastic blocking probabilities and, as it is advisable, a worst-case guarantee of blocking probabilities for one to have elastic traffic on the other hand, the separation parameters at the same time as large as possible and at least big enough be the worst case blocking constraints observed. Dependent from model parameters and given barriers it can be Be the case that all requirements are not satisfied at the same time can be which means that the connection is overloaded in terms of GoS requirements.

3 Figure 12 is a schematic block diagram of pertinent portions of an IP message router (or ATM switch) implementing an algorithm of sharing a link capacity in accordance with the invention. The IP message router (English: router) 20 is connected to an input connection and an output connection. The message router 20 has a control unit 21 , a CAC unit 22 , an output buffer 23 for a rigid traffic, an exit gate 24 for an elastic traffic and an exit scheduler 25 ,

The Control unit is preferably, although not necessarily, as a software that runs in a computer system realized. The software can work in almost any type of computer language, such as C, C ++, Java or even specialized proprietary languages. The connection capacity algorithm is actually mapped in a software program that, when in the computer system executed, appealing to appropriate traffic input parameters from the network and the UNCs by conventional Means are received, a set of output control parameters C_ELA, C_COM, N_ELA, N_COM generated.

The parameters N_ELA, N_COM represent the separation parameters for each one rigid traffic and one elastic traffic. In the example of 3 is considered only a single elastic traffic class, and only a single elastic parameter separation parameter N_ELA is therefore determined by the control unit 21 generated. The separation parameters become the CAC unit 22 which will accept or reject new streams based on the forwarded separation parameters. For each requested new stream, the traffic class of the stream is determined so that admission control based on the relevant separation parameter can be exercised. IP packets pertaining to accepted rigid streams (which are limited by N_COM) are used by the outbound scheduler for subsequent scheduling 25 to the output bumper 23 forwarded. In the same way, IP packets belonging to accepted elastic streams (bounded by N_ELA) are added to the output buffer 24 forwarded.

The parameters C_ELA, C_COM are provided by the control unit 21 to the starting time scheduler 25 forwarded. The initial goal scheduler 25 represents the bandwidth of the originating link, and the actual bandwidth representation used in traffic scheduling is determined by the C_ELA, C_COM parameters. At the exit time scheduler 25 For example, the bandwidth of the output link is divided into a common part C_COM and a dedicated part C_ELA which is reserved for elastic traffic only. When scheduling IP packets, the Outbound Scheduler can 25 only the common bandwidth part C_COM for IP packets from the output buffer 23 for rigid currents use. For IP packets from the output throttle 24 for elastic currents, on the other hand, the scheduler 25 both the dedicated bandwidth part C_ELA and the common bandwidth part C_COM use. In this way, the exit gate exit scheduler will decide 25 How many IP packets can be sent on the outgoing link per unit time and traffic class.

Analysis of throughput and blocking probability measures more elastic streams

The throughput requirements, in the evaluation step 105 ( 2 ) can be, for example, average throughput requirements, the separation parameters meeting throughput requirements if: E (θ ~) ≥ θ ~ min , E (θ ^) ≥ θ ^ min (6) where E is the expected value. To make a plausible interpretation of this type of constraint, suppose that the distribution of θ is fairly symmetric about E (θ). In other words, the central value of θ is close to E (θ). In this case, the probability that an elastic current receives less bandwidth than θ ^min is about 0.5.

However, users often prefer more informative throughput constraints, and an alternate constraint may require that the throughput of adaptive and non-adaptive streams be determined independently of the associated service requirements (x) or hold times (t), respectively, with predetermined probabilities (1-ε ₂ ) and (1). ε ₃ ) is greater than θ ~ _min and θ ^ _min is: Pr (θ ~ t ≥ θ ~ min ) ≥ (1 - ε 2 ), Pr (θ ^ x ≥ θ ^ min ) ≥ (1 - ε 3 ) (7)

The worst case constraints of the Elasticity Blocking probabilities may simply be considered B 2 ≤ B 2 Max , B 3 ≤ B 3 Max (8th) be expressed.

bestimmt werden.To set the elastic throughput dimensions (step 104 ) and possibly further to obtain the elastic blocking measures for given values of the separation parameters to obtain a score (step 105 ) in view of the given conditions, it is necessary that the steady-state power distribution of a Markov chain describing the dynamics of the mixed rigid-elastic traffic is determined. As implied in connection with the formulation of the preceding multi-class model, the system under test may be referred to as a Continuous Time Markov Chain (CTMC) whose state is determined by the number of streams of different traffic classes (n ₁ , n ₂ , n ₃ ) is uniquely identified. It is clear that in order to obtain the performance measures of the system we have to determine the generator matrix Q of the CTMC and its steady state power distribution P = {P _i }, where P ^T * Q = 0 and Σ _i Pi = 1 Generator matrix and a steady state power distribution of a Markov chain are considered to be well known to those skilled in the art. For a general introduction to loss networks, Markov theory, and the general stochastic backpack problem, reference is made to [8], and more particularly, pages 1-69 thereof. For given values of the parameters C _COM , N _EL2 , N _EL3 , the set of triples (n ₁ , n ₂ , n ₃ ) satisfying the constraints of the elastic POL strategy given by (1-4) satisfies , the set of feasible states of the system, denoted by S. The power of the state space can be considered

be determined.

It is easy to realize that the generator matrix Q has a nice structure since only transitions between "neighboring states" are allowed in the following sense: q _ij shall designate the transition rate from a state i to a state j. 1-4) in terms of the number of streams in the system defined by the elastic POL strategy are then the non-zero transition rates between the states q i, ik + = λ k k = 1, 2, 3 (10) q i, IK = n k · μ k k = 1, 2 (11) q i, i3 = n 3 · r 3 · μ 3 (12) where i ₁₊ = (n ₁ + 1, n ₂ , n ₃ ) when i = (n ₁ , n ₂ , n ₃ ); i _{k +} and i _k- (k = 1, 2, 3) are similarly defined. The expression (10) represents the state transitions due to a connection arrival, while (11) and (12) represented transitions due to connection drops. The set defined in (12) denotes the total bandwidth of the non-adaptive streams when the system is in state i. The generator matrix Q of the CTMC is constructed based on the transition rates defined in (10-12).

For illustrative purposes, consider a small system with a rigid class, an adaptive elastic class, and a non-adaptive elastic class in which the connection capacitance is C = 7. For simplicity, assume a division of the link capacity such that n ₁ = 1 is held fixed, that is, the available bandwidth for elastic streams is 6 bandwidth units. Furthermore, b ₁ = 1, b ₂ = 3 and b ₃ = 2. The elastic currents are characterized by their minimum accepted bandwidths, which are set here to b ₂ ^min = 1.8 and b ₃ ^min = 0.8. Setting the separation _parameters to N _EL2 = 2 and N _EL3 = 3 causes 12 _operable states, as in the Markov chain _{state space diagram of FIG} 4 is shown. There are 5 (gray) states where at least one of the elastic currents is compressed below the peak bandwidth specified by b ₂ and b ₃ . The states are identified by the number of active connections (n ₁ , n ₂ , n ₃ ). The values below the state identifiers indicate the bandwidth compression of the adaptive and non-adaptive elastic traffic (r ₂ , r ₃ ). The state (1, 2, 3) is the only one in which the bandwidth compression of the adaptive class and the non-adaptive class differs due to different minimum bandwidth requirements (r ₂ ^min = 0.6, r ₃ ^min = 0.4).

Different numerical solutions can be used to obtain the steady-state power distribution of a so-called multidimensional Markov chain. Direct methods, such as the Gaussian elimination method, calculate the solution in a fixed number of operations. However, considering the size of the state space for practically interesting cases, that is, large state spaces on the order of 10 ⁴ or larger, the computational complexity of the direct methods is usually unacceptable. An iterative procedure, such as the Biconjugated Gradient method used here, is therefore much more feasible for steady state analysis. The Biconjugated Gradient Method is further detailed in Reference [9].

The Calculation time of an iterative procedure depends on factors such as speed a convergence and the complexity of each iteration step from. The computing time hangs furthermore extremely of the initial one estimate from. A good initial estimate will significantly reduce overall computing time. For this reason is according to the invention a heuristic direct method of calculating a fairly close one initial Estimate, used in the iterative procedure. Some special multidimensional Markov chains show a so-called Product form solution which means that the steady-state probability of a state (i, j) in a product form as f (i) · g (j) h (i, j) can be efficiently determined. Due to the occasional Reduction in bandwidth (and a corresponding drop-off rate) of the non-adaptive elastic currents shows unfavorably the CTMC of the studied system have the nice properties of reversibility and a product form solution not, the proposed initial Estimate, the for the following iterative numeric procedure is used however, calculated as if the Markov chain showed a product form. In other words, the initial one Form of stationary Power distribution of a Markov chain, which is a transport system, which comprises non-adaptive elastic traffic, describes based on Markov chain product shape calculations determined and in an iterative stationary state analysis method applied.

p(n₁, n₂, i), ∀(n₁, n₂)∈ S wird aus p*(0, 0) = 1 (13)

erhalten, wobei die p*(n₁, n₂) nicht normierten Stationärer-Zustand-Wahrscheinlichkeiten Hilfsvariablen des iterativen Verfahrens sind. Aus der stationären Leistungsverteilung der starren und adaptiven Ströme (p(n₁, n₂)) wird das Gesamtstationärer-Zustand-Verhalten (p(n₁, n₂, n₃)) durch Festsetzen der Anzahl starrer Ströme (n₁ = i) und Annehmen, dass die erhaltene Markov-Kette umkehrbar ist, selbst wenn dies nicht der Fall ist, erhalten. Diese Annahme lässt zu, dass wir eine anfängliche Schätzung für das iterative Verfahren wie folgt bewerten. Für alle möglichen festen Werte von n₁ (n₁ = i): p*(i, 0, 0) = 1 (16)

The fact that only non-adaptive elastic currents interfere with reversibility is utilized, and the Markov chain describing the number of rigid and adaptive elastic currents in the system is reversible, and p (n ₁ , n ₂ ) =

p (n ₁ , n ₂ , i), ∀ (n ₁ , n ₂ ) ∈ S turns off p * (0, 0) = 1 (13)

where the p * (n ₁ , n ₂ ) non-normalized steady-state probabilities are auxiliary variables of the iterative method. From the steady state power distribution of the rigid and adaptive currents (p (n ₁ , n ₂ )), the total steady-state behavior (p (n ₁ , n ₂ , n ₃ )) is determined by setting the number of fixed currents (n ₁ = i ) and assuming that the obtained Markov chain is reversible, even if this is not the case. This assumption allows us to evaluate an initial estimate for the iterative procedure as follows. For all possible fixed values of n ₁ (n ₁ = i): p * (i, 0, 0) = 1 (16)

In other words, we group states with common parameters n _i , n ₂ , summing their probabilities to obtain a new 2-dimensional Markov chain. The obtained 2-Dimensional Markov chain shows a product form, and its steady-state power distribution is calculated using Equations (13-15). We next "share" the probability of the state groups between the individual states that define a state group using equations (16-18).

The stationary Power distribution of traffic classes other than non-adaptive Traffic class is calculated as if there were no non-adaptive Traffic in the system, and state probabilities then become assuming a balance between an incoming and one outgoing traffic of one of the other traffic classes and the non-adaptive elastic traffic class. at In the present example, the steady state power distribution of rigid traffic class and the adaptive elastic traffic class calculated as if there is no non-adaptive elastic traffic in the system, and State probabilities are assumed under the assumption that the incoming and outgoing adaptive and non-adaptive elastic Traffic is in balance, determined. It should, however, be understandable that equations (13-18) to a variety of applications, such as transport systems with several rigid traffic classes, but only a single one elastic traffic class, can be adjusted.

It should also be understood be that the previous procedure to calculate an initial approach a stationary one Power distribution generally on any multidimensional Markov chain is applicable and adapted to different applications can be.

The obtained initial approach the stationary one Power distribution is considered a good initial estimate for an iterative process, like the biconjugated gradient-based Procedure, which gradually used the initial estimate to one improved accuracy.

berechnet werden.Based on the steady state power distribution of the CTMC, the connection blocking probabilities may be considered

be calculated.

The calculation of the average throughput of the adaptive and non-adaptive elastic currents is also quite simple once the steady state power distribution of the CTMC is determined:

The Blocking probability requirements in (5) and (8) as well as the Average throughput requirements in (6) can therefore be assessed.

Unfortunately, it is much more difficult to verify the throughput thresholds in (7) because neither the distribution nor the larger moments of θ ~ _t and θ ^ _x can be analyzed based on the steady state power distribution of the previously studied Markov chain. A new analytical approach is therefore used. The adaptive elastic flow throughput threshold can be checked based on the distribution of θ ~ _t , and the non-adaptive elastic flow throughput threshold can be checked based on the distribution of T _x Pr (θ ^ x ≥ θ ^ min ) = Pr (x / T x ≥ θ ^ min ) = Pr (T. x ≤ x / θ ^ min ) (22)

Since it is computationally too difficult to evaluate the distribution of T _x and θ ~ _t for realistic models, but there are effective numerical methods to obtain their moments, we check the throughput threshold constraint by applying a moment-based distribution estimation method, as disclosed in reference [10] and summarized in the following Table II. In Table II, μ _n denotes the n-th moment of the random variable X, and the formulas represent an upper and lower bound on the distribution of X. Table II is valid for every non-negative random variable, that is, we use the It is not a fact that T _x and θ ~ _t in our system are limited upwards. Table II

The method to evaluate the moments of T _x and θ ~ _t is based on marking an elastic current arriving at the system and thoroughly studying the possible transitions from the moment this marked current enters the system until he leaves the system. The system behavior during the service of the marked stream can be described by a slightly modified Markov chain. In order to analyze θ ~ _t , a marked adaptive elastic current is considered while, to analyze T _x , a marked non-adaptive elastic current is considered. The modified system used to evaluate θ ~ _t (or T _x ) has the following properties:
Since it is assumed that at least the marked elastic current is present in the system, we exclude states where n ₂ = 0 (or n ₃ = 0).

With each state of the state space, there is an associated probability of occurrence that is the probability of the event that the modified CTMC begins from that state. When the marked elastic current finds the system in a state (n ₁ , n ₂ , n ₃ ), the system becomes a state (n ₁ , n ₂ + 1, n ₃ ) (or a state (n ₁ , n ₃ ) ₂ , n ₃ - 1)), except that it happens that the state (n ₁ , n ₂ , n ₃ ) is a blocking state of the marked current.

Let {Z (t), t ≥ 0} be the modified CTMC on the assumption that the marked elastic current never leaves the system beyond the finite state space F with a generator B. F can as 0 ≤ n 1 · b 1 ≤ C COM (23) 1 (or 0) ≤ n 2 ≤N EL2 (24) 0 (or 1) ≤ n 3 ≤N EL3 (25) be defined.

In fact, F = S \ S ₀ , where S _{0 are} the states in S, where n ₂ = 0 (or n ₃ = 0). The state transition rates in B are closely related to the appropriate rates in Q: b i, ik + = λ k k = 1, 2, 3 (26) b i, i1 = n 1 · μ 1 (27) b i, i2- = (n 2 - 1) · μ 2 (or n 2 · μ 2 ) (28) b i, i3 = n 3 · r 3 · μ 3 (or (n 3 - 1) · r 3 · μ 3 ) (29)

The initial probability of the modified Markov-chain p ⁺ (n _1, n _2, n ₃₎ is obtained immediately by viewing the system state after the labeled flow connecting the system in a stationary state. This means that the probability that the system will be in the state (n ₁ , n ₂ , n ₃ ) after the arrival of the marked current is proportional to the steady-state probability of the state (n ₁ , n ₂ -1, n ₃ ) (or (n ₁ , n ₂ , n ₃ - 1)). Consequently:

To obtain the moments of θ ~ _t , a Markov reward model according to reference [11] is defined about {Z (t), t ≥ 0}. θ ~ _t is a random variable that depends on the random arrival and random departure of the rigid, adaptive and non-adaptive elastic streams as described by B. The reward rate associated with the states of the modified Markov chain represents the bandwidth of the labeled adaptive elastic currents in that state. Let t _i be the reward rate (the bandwidth of the labeled adaptive elastic current) in state i and T the diagonal Matrix composed of t _i entries. t _i = r ₂ (i) * b ₂ , where r ₂ (i) is the bandwidth compression in state i. In this way, the dynamics of the number of streams in the system during the service of the marked stream are described by the modified Markov chain, and the instantaneous bandwidth of the marked stream is described by the instantaneous reward rate. If there are more currents in the system, the bandwidth of the marked current decreases to b ₂ ^min , and if there are fewer currents it increases to b ₂ . The generator matrix B and the reward matrix T define the Markov reward model that accumulates t · θ · _t sets of reward in the interval (0, t). This means that the reward accumulated in the interval (0, t) represents the amount of data sent by the marked stream in this interval and θ ~ _{t is} the amount of data sent / t.

T _x is the random amount of time needed to send x units of data through the marked stream. By defining a Markov reward model as above, the reward accumulated in the interval (0, t) represents the random amount of data sent by the marked stream, and T _x is therefore the time spent is to accumulate x amounts of a reward. This measure is commonly referred to as completion time.

From the initial probability distribution p ²⁺ (n ₁ , n ₂ , n ₃ ) and p ³⁺ (n ₁ , n ₂ , n ₃ ), the generator matrix B and the reward matrix T, the numerical analysis method described in reference [11 ], applied to evaluate the moments of θ ~ _t and T _x . This numerical method is applicable to Markov reward models with large state spaces (-10 ⁶ states).

Numerical examples of the application of the Algorithm of sharing the connection capacity

As an example, consider a broadcast connection of capacity C = 100 Mbps that supports three different service classes: a rigid, an adaptive elastic, and a non-adaptive elastic service class. The parameters given as network traffic inputs and determined by the shared benefit algorithm of this system are as follows:
C _COM = 20 Mbps, C _ELA = 80 Mbps;
b ₁ = 1 Mbps, b ₂ = 5 Mbps, b ₃ = 3 Mbps;
λ ₁ = λ ₃ = 12 1 / min;
μ ₁ = μ ₂ = μ ₃ = 1 1 / min;
r ₂ ^min = 0.05, r ₃ ^min = 0.001;
N _COM = 20, N _EL2 = 20, N _EL3 = 20.

The effect of the arrival rate λ _{2 of} the adaptive elastic currents on the corresponding blocking probability for a number of values of minimum accepted throughput is illustrated in the following Table III.

Table III

As the minimum accepted throughput θ ~ _min for the adaptive elastic traffic is assigned larger and larger values, the likelihood that an adaptive elastic current will receive this throughput is reduced. Increasing the arrival rate of the adaptive elastic currents results in more adaptive elastic currents in the system, and therefore the throughput is reduced along with the likelihood that the adaptive elastic currents will receive the required bandwidth.

The effect of the arrival rate λ _{3 of} the non-adaptive elastic currents on the corresponding blocking probability for a number of values of minimum accepted throughput is illustrated in the following Table IV. In this case, the system parameters are:
C = 250 Mbps;
C _COM = 50 Mbps, C _ELA = 200 Mbps;
b ₁ = 1 Mbps, b ₂ = 3 Mbps, b ₃ = 5 Mbps;
λ ₁ = 40 1 / min, λ ₂ = 25 1 / min;
μ ₁ = μ ₂ = μ ₃ = 1 1 / min;
r ₂ ^min = 0.4, r ₃ ^min = 0.05;
N _COM = 50, N _EL2 = 120, N _EL3 = 180.

It It should be noted that in this case the modified Markov chain, the system behavior during the Service of a marked non-adaptive elastic current, 1,116,951 states and has 6,627,100 transitions.

Table IV

Similar to the effects illustrated in Table III are reduced, while the minimum accepted throughput for the non-adaptive elastic Traffic bigger and larger values be assigned the probability that a non-adaptive elastic current receives this throughput. From the increase of Arrival rate, there is also a decreasing probability that the non-adaptive elastic currents have the required bandwidth receive.

To get an idea of the relationship between average throughput and throughput thresholds, see 5 representing the mean and variance of the throughput of adaptive elastic currents as a function of their service time. The graphic representation of 5 is related to the system considered for Table III, where λ ₂ = 14. The mean throughput is shown by a solid line, whereas the variance is shown by a dashed line. Thus, it can be seen that for "short" (in terms of service) connections, the variance in throughput is quite significant, and the average throughput and throughput thresholds therefore have significantly different meanings. For "long" connections, the variance of the Throughput is almost, and the mean throughput provides a meaningful description of the bandwidth available for adaptive elastic currents. It should be noted that θ ~ _t tends to approach a deterministic value, steady-state throughput, as t approaches infinity.

Finally, we examine an example of how to select N _EL2 and N _EL3 to provide the required QoS parameters. Assume that after splitting the link capacity and dimensioning the rigid class, the system parameters have the following values:
C = 100 Mbps;
C _COM = 20 Mbps, C _ELA = 80 Mbps;
b ₁ = 1 Mbps, b ₂ = 5 Mbps, b ₃ = 3 Mbps;
λ ₁ = 12 1 / min, λ ₂ = 12 1 / min, λ ₃ = 12 1 / min;
μ ₁ = μ ₂ = μ ₃ = 1 s (expressed here as average holding time);
b ₃ ^min = 0.1 Mbps;

The parameters N _EL2 and N _EL3 must be such that the elastic _{lockup probabilities are} less than 1% (B ₂ <0.01, B ₃ <0.01) and the average _{flow rate} parameters E (θ ~) ≥ 4.05 and E (θ ^) ≥ 2.35.

The set of parameters N _EL2 and N _EL3 that meet the QoS requirements is in the gray area of 6 shown. The blocking likelihood limit of the adaptive elastic class is a vertical line due to independence from the load of the non-adaptive elastic class. The blocking likelihood limit of the non-adaptive elastic class is a horizontal line. With the considered low level of total load, the average elastic _{flow rates} after a given limit are hardly sensitive to the parameters N _EL2 and N _EL3 . In this example, the tighter of the two bandwidth limits, which determines the acceptable values N _EL2 and N _EL3 , is the bound E (θ ~) ≥ 4.05.

Inversion of the optimization task

The new elastic POL strategy allows for a natural inversion of the optimization task, so that rather than minimizing elastic traffic blockage probabilities under throughput constraints, the elastic throughputs are maximized under blocking probability constraints. Similar to that in the flow chart of 2 illustrated methods of sharing the connection capacity will receive traffic input parameters (similar to step 101 ), the common connection _{capacity part} C _COM is determined (similar to step 102 ), and initial values of the separation para meters are selected (similar to step 103 ). The performance of the system will be analyzed next (similar to step 104 ), but now mainly with respect to elastic blocking probabilities. In particular, the elastic locking probabilities in the system are analyzed and related to the blocking likelihood constraints (similar to step 105 ). If the blocking probabilities are too large, then the separation parameters are increased, reducing the blocking probabilities and also reducing the throughputs. On the other hand, if the blocking probabilities are lower than the blocking constraints, then the separation parameters can be reduced, thus increasing the blocking probabilities as well as the throughputs. In this way, iteration can maximize throughput while still adhering to elastic flow blocking constraints. Since the goal now is to maximize elastic throughputs, and since it may be advisable to have a worst-case guarantee on the flow rates of elastic traffic, the separation parameters must be as low as possible and at least low enough to be the worst Compliance with flow throughput conditions while the blocking likelihood constraints imposed on elastic traffic are still met.

Of course, regardless of whether it is adapted to minimize elastic blocking or to maximize elastic throughput, the algorithm of sharing the connection capacity is also applicable to only elastic traffic. For example, in the absence of rigid traffic, C _{COM is} reduced to zero and the algorithm of sharing the total link capacity is reduced to the mathematical formulas for determining the separation parameters under throughput / stall constraints. In the case of a single elastic traffic class, it is further necessary that only a single separation parameter be determined in accordance with the previous iterative algorithm of sharing the connection.

Algorithm of the common use of the Connection Capacity - the ATM network example

Even though the algorithm of the common use of the connection capacity above Reference to an IP network that is a single rigid traffic class and two different elastic traffic classes described was, should be understood be that the invention is not limited thereto and that the Algorithm on other types of networks and other traffic classes is applicable. An example of the elastic POL algorithm used in an ATM network having a narrowband CBR (constant bit rate) Traffic and broadband CBR traffic and an ABR (Available Bit Rate = available Bitrate) carries traffic, is actually applied outlined below.

• Klasse 1 – Schmalband-CBR-Verbindungen, die durch ihre Spitzenbandbreitenanforderung b₁, ihre Verbindungsankunftsrate λ₁ und ihre Abgangsrate μ₁ gekennzeichnet sind.
• Klasse 2 – Breitband-CBR-Verbindungen, die durch ihre Spitzenbandbreitenanforderung b₂, ihre Verbindungsankunftsrate λ₂ und ihre Abgangsrate μ₂ gekennzeichnet sind.
• Klasse 3 – ABR-Verbindungen, die durch ihre Spitzenbandbreitenanforderung b₃, ihre Mindestbandbreitenanforderung b₃ ^min, ihre Verbindungsankunftsrate λ₃ und ihre ideale Abgangsrate μ₃ gekennzeichnet sind. Die ideale Abgangsrate wird erlebt, wenn die Spitzenbandbreite während der gesamten Dauer der Verbindung verfügbar ist.

In this example, connections arriving at a broadcast connection generally belong to one of the following three traffic classes:

• Class 1 - narrowband CBR links characterized by their peak bandwidth requirement b ₁ , their connection arrival rate λ ₁ and their leaving rate μ ₁ .
• Class 2 - broadband CBR connections characterized by their peak bandwidth requirement b ₂ , their connection arrival rate λ ₂ and their leaving rate μ ₂ .
• Class 3 - ABR connections characterized by their peak bandwidth requirement b ₃ , their minimum bandwidth requirement b ₃ ^min , their connection arrival rate λ ₃ and their ideal leaving rate μ ₃ . The ideal exit rate is experienced when the peak bandwidth is available throughout the duration of the connection.

It be noted that the CBR classes the rigid traffic class of the previous IP network example and that the ABR class the non-adaptive elastic traffic class, the previous one in connection with the IP network example. The assumptions in the model formulated in the IP Network example are in this regard equally applicable in the present example.

The elastic POL strategy described above is applied to the environment of mixed CBR ABR traffic in the considered ATM network. This means that the connection capacity C is divided into two parts, a common part C _COM for CBR connections and ABR connections and a dedicated part C _ABR reserved for the ABR connections, such that C = C _COM + C _ABR . An admission control _{parameter NARR} , which is further referred to as a separation parameter, is introduced for the ABR links. Under the elastic POL strategy, the number n ₁ , n _2, and n _{3 of} narrowband CBR running connections, broadband CBR connections, and ABR connections are on the Subject to the following conditions: n 1 · b 1 + n 2 · b 2 ≤ C COM (32) N ABR · b 3 min ≤ C ABR (33) n 3 ≤N ABR (34)

In (1), the ABR compounds are protected from CBR compounds. In (2-3), the maximum number of ABR connections is limited by two constraints. The expression (2) protects CBR compounds from ABR compounds, while (3) protects the current ABR compounds from new ABR compounds. In this case, the elastic POL strategy is completely determined by the division of the connection capacity specified by C _COM and the admission control parameter NARR. The performance of the Elastic POL strategy is tuned by these parameters.

According to a preferred embodiment of the invention, the algorithm of sharing the connection capacity aims at setting the output parameters C _COM and N _{ABR of} the elastic POL strategy in such a way as to allow the connection blocking probability for the ABR connections while being capable of Blocking Probability Requirements (GoS) for the different types of CBR connections and a minimum throughput requirement for the ABR connections. Each CBR class is therefore associated with a maximum accepted link blocking probability B ₁ ^max and B ₂ ^max , and the ABR class is at a minimum accepted throughput θ _min , which is similar to the minimum accepted throughput θ ^ _min for the non-adaptive elastic traffic IP network example.

Nevertheless, although the ABR blocking probability is minimized, it is normally advisable, although not necessary, to have a worst case guarantee of link blocking probability for ABR links and also connect the ABR class to a maximum allowed blocking probability B ₃ ^max .

The Parameters and performance measures, which are associated with the CBR classes and the ABR class are summarized in Table V below.

Table V

The problem of determining the output parameters of the elastic POL strategy under the previous conditions is described below with reference to FIG 7 3, which is a schematic flow diagram of the algorithm of sharing the total link capacity for a mixed CBR-ABR traffic environment in accordance with a preferred embodiment of the invention. At the first step 201 the required input parameters are delivered. At step 202 the GoS (Connection Blocking) request for CBR traffic is guaranteed by the proper setting of C _COM . Specifically, we determine the required minimum capacity of C _COM for CBR connections, which guarantees the required blocking probabilities B ₁ ^max and B ₂ ^max : min {C COM : B 1 ≤ B 1 Max , B 2 ≤ B 2 Max ); (35)

For example, the well-known Erlang B formula may be used to estimate such a value of C _COM based on arrival and departure rates and peak bandwidth requirements for the CBR classes as inputs.

We must next determine a maximum number of ABR connections (N _ABR ) that can coexist in the system at the same time that the required throughput and blocking requirements are met.

In this particular embodiment, the sharing capacity utilization algorithm aims at minimizing the blocking probability of the ABR links below a minimum throughput constraint. To achieve this, the invention proposes an iterative procedure that steps through 203 - 207 is generally defined, to tune the separation parameter so that the throughput threshold constraint is accurately met, generally not more and not less. At step 203 First, an initial value of the separation parameter is estimated. The performance of the system is next analyzed in terms of ABR throughput (step 204 ) and related to the throughput threshold pad (step 205 ). If the ABR flow rate is too low, then the separation parameter is decreased (step 206 ), whereby the blocking probability is increased and also the throughput is increased. On the other hand, if the ABR throughput is greater than the throughput threshold, then the separation parameter can be increased (step 207 ), so that the blocking probability (as well as the throughput) is reduced. This can be done by iteratively repeating the steps 204 . 205 and 206 / 207 the ABR blocking probability is reduced to a minimum while the throughput requirement is still adhered to.

The Performance measures the ABR throughput and possibly also the ABR blocking is preferably more or less the same way as above in connection with the IP network example described, analyzed. This means in short a determination the stationary one Power distribution of the Markov chain, the dynamics and the Behavior of mixed CBR-ABR environment describes and calculating Blocking and flow rates based on the determined Distribution. It should be noted, however, that here the throughput threshold pad, which is analogous to the expression (7) based on the transient Analysis of the Markov chain describing the mixed CBR-ABR environment, using the numerical one proposed in reference [11] Method using the Markov inequality.

It is self-evident possible, the optimization task in much the same way as above for explains the IP network example is, also for to invert the ATM network example.

Numerical examples of the application of the Algorithm of sharing the connection capacity

As an example, consider an ATM broadcast connection of capacity C = 155 Mbps supporting three different service classes: two CBR classes and one ABR class as described above. The input parameters of this ATM mailing system are:
b ₁ (n-CBR) = 3 Mbps, b ₂ (w-CBR) = 6 Mbps, b ₃ (ABR) = 10 Mbps;
λ ₁ = 6 1 / s, λ ₂ = 3 1 / s, λ ₃ = 12 1 / s;
μ ₁ = μ ₂ = μ ₃ = 1 1 / min;
r ₂ ^min = 0.05, r ₃ ^min = 0.001;
N _COM = 50.

It is further required that the blocking probabilities of the narrowband and wideband CBR connections be less than each of B ₁ ^max = 2% and B ₂ ^max = 4%. It follows, therefore, that the minimum bandwidth for C _COM necessary to provide these blocking probabilities is 60 Mbps, leaving C _ABR = 95 Mbps for the ABR connections.

To study the trade-off between throughput and blocking probability for ABR traffic, reference is made to the following Table VI, which illustrates the average flow rate E (θ) and the blocking probability B ₃ for the ABR traffic class for different values of NARR. Table VI N _ABR 10 20 40 60 80 100 150 B ₃ 0,310 .0811 0.0320 0.0212 0.00141 0.00112 0.000461 E (θ) 9.99 7.9 4.83 3.45 2.69 2.2 1.52

From Table VI, the trade-off between throughput and blocking is evident; large blocking = high throughput and low blocking = low throughput. In the elastic POL strategy according to the invention, this compromise is expediently controlled by means of the separation _parameter N _ABR , as can be seen from Table VI. For example, if an overlay θ ^{min is set} to 2.2 Mbps in terms of average throughput, the maximum number (N _ABR ) of simultaneously active ABR links is limited to 100.

It has been observed in simulations that the elastic POL strategy is superior to the well-known complete partitioning (CP) strategy under all loads both in terms of blocking probabilities and ABR throughput. This is partly due to the fact that the POL policy allows ABR connections to use any bandwidth of the C _COM part not currently being used by CBR connections.

Finally, to study the effect of the parameter C _COM on the blocking probabilities and the average throughput of ABR traffic, reference is made to the following Table VII. Table VII C _COM 69 66 63 60 57 54 B ₁ 0.00498 0.00770 0.0116 0.0171 0.0244 0.0342 B ₂ 0.0126 0.0192 0.0284 0.0411 0.0578 0.0794 B ₃ 0.0149 0.0141 0.0129 0.0115 0.00973 0.00773 E (θ) 2.04 2.08 2.13 2.20 2.31 2.46

The parameter C _COM provides a way of controlling the compromise between the CBR blocking probabilities on the one hand and the ABR blocking probability and the ABR throughput on the other hand. It can be seen from Table VII that both the (increasing) ABR throughput and the (reducing) ABR blocking are improved at the expense of degrading CBR blocking probabilities.

It It is important to understand that the preceding description is intended is to serve as a framework for agreement to serve the invention. The embodiments described above are just as Examples given, and it goes without saying that the present Invention is not limited thereto.

REFERENCES

• [1] L. Massoulie, J. Roberts, "Bandwidth Sharing: Objectives and Algorithms ", IEEE Infocom '99, p. 1395-1403, March 1999.
• [2] L. Massoulie, J. Roberts, "Bandwidth Sharing and Admission Control for Elastic Traffic ", ITC Specialist Seminar, Yokohama, October 1998.
• [3] L. Massoulie, J. Roberts, "Arguments in Favor of Admission Control for TCP Flows," 16 ^th International Teletraffic Congress, Edinburgh, UK, June 1999.
• [4] RJ Gibbens and FP Kelly, "Distributed Connection Acceptance Control for a Connectionless Network," 16 ^th International Teletraffic Congress, Edinburgh, UK, June 1999.
• [5] F.P. Kelly, "Charging and Rate Control for Elastic Traffic ", European Transaction to Telecommunications, p. 33-37, Volume 8, 1997.
• [6] Wu-chang Feng, Dilip D. Kandlur, Debanjan Saha and Kang. G. Shin, "Understanding and Improving TCP Performance Over Networks with Minimum Rate Guarantees ", IEEE / ACM Transactions to Networking, p. 173-187, Volume 7, No. 2, April 1999.
• [7] E.D. Sykas, K.M. Vlakos, I.S. Venieris, E.N. Protonotarios, "Simulative Analysis of Optimal Resource Allocation and Routing in IBCN's, IEEE J-SAC, Vol. 9, No. 3, 1991.
• [8] Keith W. Ross, "Multi-service Loss Models for Broadband Telecommunication Networks ", Springer-Verlag, 1995, ISBN 3-540-19918-7.
• [9] W.J. Stewart, "Introduction to the Numerical Solution of Markov Chains, p. 220-221, Princeton University Press, Princeton, New Jersey, ISBN 0-691-03699-3, 1994.
• [10] M. Frontini, A. Tagliani, "Entropy-convergence in Stieltjes and Hamburger moment problem ", Appl. Math. And Corp., 88, pp. 39-51, 1997.
• [11] M. Telek and S. Rácz, "Numerical analysis of large Markov reward models ", Performance Evaluation, 36 & 37: 95-114, August 1999.
• [12] A. Smith, J. Adams, G. Tagg, "Available Bit Rate - A New Services for ATM", Computer Networks and ISDN Systems, 28, p. 635-640, 1996th

Claims (44)

Method for sharing connection capacity in a network ( 10 ), the method comprising the step of receiving ( 101 ; 201 ) of network traffic input parameters, characterized in that the method further comprises the steps of dividing the link capacity into a first part (C _COM ) shared by the elastic traffic and the non-elastic traffic, and a second part (C _ELA ; C _ABR ) destined for elastic traffic based on at least a portion of the network traffic input parameters, and determining ( 103 - 108 ; 203 - 208 ) of at least one admission control parameter for the elastic traffic, based on the distribution of the connection capacity and at least a part of the network traffic input parameters. Method according to claim 1, characterized in that the method further comprises the step of exercising the admission control for elastic traffic based on the particular admission control parameter (s) includes. Method according to claim 1, characterized in that the splitting step comprises the step of determining ( 102 ; 202 ) of a required minimum capacity of the shared part in relation to non-elastic traffic, taking into account at least one maximum permissible blocking probability for non-elastic traffic. A method according to claim 1, characterized in that the step of determining at least one admission _{control parameter comprises} the step of determining a maximum number (N _EL2 , N _EL3 , N _ABR ) of allowable elastic traffic flows based on at least one link level imposed on the elastic traffic , Method according to claim 4, characterized in that in that the step of determining a maximum number of allowable elastic traffic based on a link level model for elastic traffic, and that the one or more interconnect level pads include at least either the throughput or blocking probability of the related to elastic traffic. A method according to claim 5, characterized in that the step of determining a maximum number of allowable elastic traffic streams comprises the steps of: determining ( 103 ; 203 ) an initial value of the number of permissible elastic traffic flows on a connection in the network, iteratively performing the following steps: i) evaluating ( 105 ; 205 ) the one or more throughput / blocking constraints imposed on the elastic traffic based on the initial current value of the number of allowed elastic traffic streams, and ii) setting the number of allowed elastic traffic flows based on the rating, and terminating the iteration process and extracting ( 108 ; 208 ) the maximum number of permissible elastic traffic streams when the overlay or pads are met. A method according to claim 6, characterized in that the step of adjusting the number of allowable elastic traffic streams based on the evaluation comprises the steps of: Reduce ( 106 ; 206 ) the number of permissible elastic traffic streams when a value related to the flow rate of the elastic traffic streams is lower than a predetermined threshold value given by the throughput / blocking constraints and increasing ( 107 ; 207 ) the number of permissible elastic traffic streams when the value associated with the throughput is greater than the threshold value. Method according to claim 4, characterized in that that the step of determining the maximum number of allowed elastic Links on minimizing the blocking probability the requested elastic traffic connections in relation to the number of allowed elastic traffic connections under at least one throughput threshold value for running elastic transport connections is based. Method according to claim 8, characterized in that that minimizing the blocking probability of the elastic Traffic below at least one throughput threshold also under at least one given run at a maximum allowable blocking probability for elastic Traffic executed becomes. Method according to claim 4, characterized in that that the step of determining the maximum number of allowed elastic Connections on maximizing elastic traffic flow in terms of the number of allowed elastic traffic connections under at least one blocking probability constraint for requested elastic transport connections is based. Method according to claim 1, characterized in that that the elastic traffic a number of elastic traffic classes and in that the step of determining at least one admission control parameter includes the step in which for every elastic traffic class a maximum number of permissible elastic traffic based on a respective throughput / blocking probability constraint, which is imposed on the elastic traffic class is determined. Method according to claim 11, characterized, that the method further comprises the following steps: Determine the associated Traffic class for each of the requested connections, and Controlling the admission each requested connection based on the corresponding one traffic class specific maximum number of allowed elastic connections. Method according to claim 1, characterized in that that the network traffic input parameters at least the connection capacity, at least a blocking probability constraint for non-elastic traffic and at least one throughput / blocking pad for elastic Traffic included. Method according to claim 1, characterized in that the network traffic input parameters further determine the arrival and departure rates and peak bandwidth requirements for non-elastic and elastic Transport as well as a minimum bandwidth requirement for elastic traffic. Method according to claim 1, characterized, that the elastic traffic of at least one of the following elements includes: Adaptive elastic traffic flows from Internet Protocol (IP) networks, non-adaptive elastic traffic flows from IP networks and Available Bit Rate (ABR) streams from ATM networks. The method of claim 1, wherein the connection bandwidth is representative of the connection bandwidth, and the elastic traffic comprises at least one elastic traffic class, and wherein the second dedicated part (C _ELA ; C _ABR ) of the connection _bandwidth comprises elastic traffic class or traffic class connections based on Utilization of the link bandwidths under at least one throughput / blocking constraint imposed on the links of the one or more elastic traffic classes. A method according to claim 16, characterized in that the method further comprises the step of determining ( 103 - 108 ; 203 - 208 ) _comprises a maximum number (N _EL2 , N _EL3 , N _ABR ) of allowable elastic traffic links on the link based on one or more throughput / blocking constraints, wherein the step of assigning bandwidth to links of the one or more elastic traffic classes to the maximum number permissible elastic traffic connections. Method according to claim 17, characterized in that in that the step of determining a maximum number of allowable elastic Traffic connections on minimizing the blocking probability elastic traffic connections in relation to the number of permissible elastic Traffic connections below at least one throughput threshold for running elastic transport connections is based. Method according to claim 17, characterized in that in that the step of determining a maximum number of allowable elastic Traffic connections on maximizing throughput running elastic Traffic connections in relation to the number of permissible elastic traffic connections under at least one blocking probability constraint for requested elastic transport connections is based. A method according to claim 17, characterized in that the step of determining a maximum number of allowed elastic traffic links comprises the steps of: determining ( 103 ; 203 ) an initial value of the number of allowed elastic traffic links on the link, iteratively performing the following steps: i) determining ( 104 ; 204 ) throughput / blocking probability measurements based on the initial / current value of the number of allowed elastic links on the link, ii) evaluate ( 105 ; 205 ) the one or more flow rate / blocking constraints imposed on elastic traffic based on the determined flow rate / blockage measurements, and iii) adjusting ( 106 / 107 ; 206 / 207 ) the number of permissible elastic traffic links based on the rating, and terminate the iteration process and extract ( 108 ; 208 ) the maximum number of permissible elastic traffic connections when they substantially satisfy the one or more conditions. The method of claim 1, further comprising the steps of: determining ( 103 - 108 ; 203 - 208 ) a maximum number (N _EL2 , N _EL3 , N _ABR ) of allowable elastic traffic links on the connection based on the network traffic _{input parameters} , the _{input parameters including} at least one throughput / _{congestion constraint} imposed on the elastic traffic, receiving requests for elastic traffic links on the connection in the network; Performing the admission control for the requested elastic traffic links based on the determined maximum number of allowed elastic traffic links. The method of claim 1, wherein the network comprises an admission control enabled IP network ( 10 ), characterized in that the method comprises the following steps: applying ( 101 - 108 ) of a link level model of a link carrying at least one nonelastic traffic class and a number n, where n is an integer equal to or greater than 1, elastic traffic classes for sizing the connection bandwidth for throughput / blocking optimality based on the sharing of link capacity in a first part (C _COM ) in common for the elastic traffic and the non-elastic traffic, and a second part (C _ELA ) assigned to the elastic traffic, the connection level model being defined by: the connection having a predetermined bandwidth capacity C and the elastic traffic for each of the n elastic traffic classes is modeled as: i) having a peak bandwidth requirement and a minimum bandwidth requirement, ii) occupying the maximum possible bandwidth within the peak and minimum bandwidth requirements, and iii) than having at least one min imalen accepted throughput probability and a maximum accepted blocking probability. Method according to claim 22, characterized, that elastic traffic for each of the n traffic classes is further modeled as: iv) dynamic arriving according to one Poisson process and with an arrival rate as well as a leaving rate connected, v) proportionately the same the available Bandwidth for the elastic traffic class between the elastic currents in common using, and vi) as having a minimum holding time. Method according to claim 22, characterized in that that the n elastic traffic classes a first traffic class for adaptive elastic currents and a second traffic class for non-adaptive elastic currents exhibit. The method of claim 1, further comprising the step of determining a Markov chain stationary power distribution that describes the dynamics of a network connection carrying traffic of a number of traffic classes, including a nonelastic traffic class and a non-adaptive elastic traffic class, by: defining a A connection sharing capacity strategy for the connection based on the division of the connection capacity into a first part (C _COM ) for sharing elastic traffic and non-elastic traffic and a second part (C _ELA ; C _ABR ) assigned to elastic traffic by determining a multidimensional Markov chain having a set of feasible states for the number of active links of the traffic classes in accordance with requirements imposed by the connection sharing strategy, calculating an initial approximation of the stationary Le distribution of the Markov chain based on Markov chain product shape calculations, and iteratively determining the steady state power distribution from the initial approximation of the steady state power distribution. Method according to claim 25, characterized, that the step of calculating an initial approximation of stationary power distribution the following steps: Determine the stationary power distribution other traffic classes than the non-adaptive traffic class, as would be there is no non-adaptive traffic in the system, and Determine the state probabilities under the assumption of equilibrium the incoming and outgoing traffic of one of the other traffic classes and the non-adaptive elastic traffic class. Method according to claim 25, characterized in that that the step of iteratively determining the steady state power distribution based on an Eikonjugierten gradient method. Method according to claim 25, characterized in that that the blocking probabilities for the traffic classes based on a stationary Power distribution can be calculated, resulting from the iterative Determination results. Device for sharing connection capacity in a network ( 10 ), the device having means for receiving network traffic input parameters, characterized in that the device comprises: means ( 21 ; 21 . 25 ) for splitting the link capacity into a first part shared for elastic traffic and non-elastic traffic and a second part destined for elastic traffic based on at least a part of the network traffic input parameters, and means ( 21 ) for determining at least one admission control parameter for the elastic traffic based on the sharing of connection capacity and at least a portion of the network traffic input parameters. Device according to claim 29, characterized in that the device further comprises means ( 22 ) for exercising the admission control for elastic traffic flows based on the particular admission control parameter (s). Device according to claim 29, characterized in that in that the dividing means comprise means for determining a required minimum capacity of the common Partly regarding non-elastic traffic considering at least a maximum allowable Blocking probability for does not include elastic traffic. Device according to claim 29, characterized in that in that the means for determining at least one admission control parameter Means for determining a maximum number of allowed elastic traffic flows based on at least one throughput / blocking pad, which is imposed on elastic traffic. Device according to claim 32, characterized, in that the means for determining a maximum number of permissible elastic traffic is configured to: initially determining a number permissible elastic traffic flows on a connection in the network, iterative evaluation of the one or more throughput / blocking requirements, which are imposed on the elastic traffic based on the current one Number of allowed elastic traffic flows and setting the number of allowed elastic traffic flows based on the rating, and Terminate the iteration process and extracting the maximum number of permissible elastic traffic streams during Fulfill one or more editions. Device according to claim 32, characterized in that in that the means for determining a maximum number of permissible elastic traffic is configured to the maximum number based on the minimize the blocking probability of requested elastic traffic connections in relation to the maximum number of permissible elastic traffic connections below at least one throughput threshold for current to determine elastic traffic connections. Device according to claim 32, characterized in that in that the means for determining a maximum number of permissible elastic traffic is configured to maximize the number based on the maximum the elastic traffic flow in relation to the maximum number permissible elastic traffic connections under at least one blocking probability constraint for elastic traffic to determine. Device according to claim 29, characterized in that the receiving means receives network traffic input parameters which the connection capacity, Arrival and departure rates, peak bandwidth requirements for not elastic as well as elastic traffic, a minimum bandwidth requirement for elastic Traffic, at least one blocking probability constraint for not elastic traffic and at least one throughput / blockage pad for elastic Include traffic. Device according to claim 29, characterized, that the elastic traffic of at least one of the following elements having: Adaptive elastic traffic flows from Internet Protocol (IP) networks, non-adaptive elastic traffic flows from IP networks and Available Bit Rate (ABR) streams from ATM networks. The apparatus of claim 29, wherein the connection bandwidth is representative of the connection bandwidth, and the elastic traffic comprises at least one elastic traffic class, and wherein the apparatus further comprises a processor (15). 21 ) configured to allocate to the second dedicated part of the connection bandwidth connections of the one or more elastic traffic classes based on the utilization of the connection bandwidth under at least one throughput / congestion constraint imposed on the connections of the one or more elastic traffic classes. Device according to claim 38, characterized in that the processor ( 21 ) is configured to allocate a required minimum bandwidth of the shared portion for non-resilient traffic class connections having at least one maximum allowed non-elastic traffic link blocking probability. Device according to claim 38, characterized in that the processor ( 21 ) is further configured to assign a maximum number of allowed elastic traffic links on the link based on the throughput / blocking constraints, and to allocate bandwidth to the elastic links based on the dedicated portion of the link bandwidth and the maximum number of allowed elastic links. Device according to claim 40, characterized in that the processor ( 21 ) is further configured to determine the maximum number of allowed elastic connections based on minimizing the blocking probability of elastic traffic links with respect to the number of allowed elastic traffic connections among at least one throughput threshold coverage for running elastic traffic links. Device according to claim 40, characterized in that the processor ( 21 ) is further configured to determine the maximum number of allowable elastic connections based on maximizing the throughput Set the elastic traffic connections in relation to the number of permissible elastic traffic connections under at least one blocking probability constraint for elastic traffic connections to determine. Device according to claim 29, characterized in that the device comprises a processor ( 21 ), responsive to the network traffic input parameters and configured to: determine a maximum number of allowed elastic traffic links based on at least one throughput / congestion constraint imposed on the elastic traffic, receive requests for elastic traffic links, and apply the admission control for the requested elastic traffic connections based on the determined maximum number of permissible elastic traffic flows. Device according to claim 43, characterized in that the processor ( 21 ) is further configured to determine the maximum number of allowable elastic connections based on the iterative enhancement of the usage of the connection bandwidth among the throughput / blocking constraints.