WO2000022785A1

WO2000022785A1 - Atm connection admission control device for dbr connections

Info

Publication number: WO2000022785A1
Application number: PCT/EP1999/007773
Authority: WO
Inventors: Pieter Hendrik Albert Venemans
Original assignee: Koninklijke Kpn N.V.
Priority date: 1998-10-12
Filing date: 1999-10-11
Publication date: 2000-04-20
Also published as: DE69933302T2; DK1121787T3; US6907467B2; US20050038903A1; EP1121787A1; US6850881B1; DE69933302D1; EP1121787B1; NL1010295C2; ES2273511T3; AU1038500A; ATE340459T1

Abstract

A device for Connection Admission Control (CAC) for an ATM switch, intended for admitting to the Deterministic Bit Rate (DBR) switch a requested ATM connection of the DBR type, such that the 'Quality of Service' of all ATM connections of the DBR type in said switch continues to satisfy certain conditions. For the composite traffic stream of DBR connections at an output port with capacity C, a single buffer with capacity B is available, while as boundary condition it holds that the total average load of the output port does not amount to more than ς x C, where ς is a constant with a value between 0 and 1. The CAC device comprises two sub-devices, which each execute a - rather complex - partial calculation, resulting in a positive or negative admission signal. The new ATM connection is only granted admission to the switch if both sub-devices emit a positive admission signal.

Description

Title: ATM Connection Admission Control device for DBR connections

BACKGROUND OF THE INVENTION

The invention is related to a device for Connection Admission Control (CAC) for an ATM switch, intended for admitting to said ATM switch an additional traffic stream, via an additional ATM connection of the DBR type, such that the "Quality of Service" of all DBR connections in said switch continues to meet certain conditions, a single buffer with capacity B being available for the composite traffic stream of all DBR connections at an output port with capacity C, while, as boundary condition, the total average load of the output port does not amount to more than p x C, where p is a constant with a value between 0 and 1.

General

Asynchronous Transfer Mode (ATM) is a network technique allowing connections to be made in a uniform manner with varying characteristics. The transport of data takes place by means of uniform cells with a length of 53 octets. With the aid of ATM, connections can be established with varying network guarantees with respect to cell loss, cell delay, cell delay variation and throughput by selecting a "Quality of Service" (QoS) class [1.356] and an "ATM Transfer Capability" (ATC) [1.371] .

Guaranteeing the values of the QoS parameters in advance is a characteristic which is focussed upon in ATM networks. In order to satisfy certain (possibly very stringent) requirements with respect to the Quality of Service in advance, including the cell loss probability, a decision must be made for each requested connection whether the latter can or cannot be admitted to an ATM switch. Connections are only admitted if the Quality of Service of the connections which make use of the ATM switch do not exceed the limits set in advance. The entirety of methods which determine admittance or non-admittance of a requested ATM connection is called a Connection Admission Control (CAC) algorithm, a device for executing a CAC algorithm being called a CAC device.

The said "ATM Transfer Capability" (ATC) describes the parameters by which an ATM connection is characterised. The ATC referred to here comprises the "Deterministic Bit Rate" (DBR) which is established in [1.371]. In this case, an ATM connection is characterised by two parameters: the Peak Cell Rate (PCR) and the Cell Delay Variation Tolerance (CDVT) . The PCR represents the maximum speed at which a connection may drive cells; the CDVT is a measure for the tolerance in the PCR. The entity CDVT denotes the degree in which the actual speed may exceed the PCR during (as a rule very short) periods of time. In order to guarantee the Quality of Service of all connections, a Usage Parameter Control (UPC) device monitors whether each of the connections indeed satisfies its characteristic during the operating period of a connection. If this is not the case, then the UPC device can resort to removing cells of the related connection from the traffic stream.

The invention relates, as indicated above, to a device which can be used for the CAC of an ATM switch, and indeed in particular for connections which satisfy the specifications of the "Deterministic Bit Rate" Transfer Capability.

THE PRIOR ART

A simple but inaccurate method for executing CAC for connections which are characterised by values for PCR and CDVT, is leaving the last parameter completely out of consideration. The method then restricts itself to determining whether sufficient capacity is available for the sum of the peak cell rates of the connections at the related output port. In this case, two matters are disregarded: • The fact that the traffic streams of the various connections consist of discrete cells, instead of continuous streams. As a result, several cells can be routed to one specific output port at (approximately) the same time when various traffic streams are mixed. In order to prevent cell loss in such situations, it is customary to use an output buffer. The said simple method for CAC, however, does not take into account the consequences of such a buffer, such as cell loss as a result of a finite buffer capacity and delay. • The fact that the traffic streams of the various connections can show fluctuations, in which case the momentary cell rate can temporarily exceed the nominal peak cell rate. Said fluctuations are possible if the Cell Delay Variation Tolerance (CDVT) of the connections is greater than zero. Through said fluctuations it is possible that, when an output buffer is used, the required amount of buffer capacity is, in reality, greater than that for "ideal" DBR connections (with CDVT being equal to 0) . This can lead to unforeseen cell loss and/or extra delay.

The literature describes a number of methods in which the above-mentioned two problems are recognised and both the discrete character of ATM traffic streams as well as the fact that the momentary cell rate of a connection can temporarily exceed its PCR are taken into account. In [E.736] it is described that this problem can be solved by either "shaping" all traffic streams (that is to say, delaying cells in such a way that the cells of the resulting traffic streams have an intermediate arrival time of exactly 1/PCR) . The resulting traffic stream can then be characterised by a CDVT which is equal to 0.

In the aforementioned reference, the "maximum burst rate" is introduced of a connection which is characterised by values for PCR and CDVT. Said maximum burst rate is the required buffer capacity to lead the related traffic stream, in an ideal situation and without cell loss, over an imaginary buffer (without mixing of other traffic streams) , assuming that the said imaginary buffer has a constant exit speed equal to PCR.

In [Gravey97] a solution with "shaping" is also proposed, albeit that connections are shaped such that their maximum burst rate is limited and small (smaller than or equal to 2) .

A disadvantage of said methods is that "shaping" is not possible with all ATM switches and that introducing extra shapers in the network can be undesirable. For these reasons, [E.736] also describes a method for the case that connections have a known maximum burst rate, bn&x. Said connections are then approximated by a b^x batch Poisson process. This method has the disadvantage of (1) the Poisson process being an over- estimation of the traffic, which leads to less connections being able to be admitted than is actually possible on the basis of the required QoS guarantees, and (2) the assumption of an unequivocally determined brøvx for all connections, which can be unnecessarily inefficient in practice. In practice, namely, said value for h_m could be quite high. In [Gravey97] it is shown that the maximally allowable load of the ATM network decreases for increasing values of D_MAX-

[COST242] describes the method essentially as in [E.736], but refines it. The maximum batches which are used for approximation by Poisson batches do not always need to be the same as in [E.736], but are dependent upon the speed of the multiplexer. For a universally valid model, the batches must indeed be just as large as those of a Poisson process by which the traffic stream is approximated.

[Mignau96] describes methods which are not based on the Poisson method but on the NxD/D/1- method, which is described below. For sources in which the maximum burst rate can assume exactly two values, a complex model is described in [Cidon95] . Said model is numerically unstable, however, and requires long calculation times. The NxD/D/1-model.

One of the models on which the present invention is based is the NxD/D/1 - model. This model provides the probability distribution of the length of the queue in a system consisting of a buffer which is controlled by one server. Said system is supplied by N ideal (i.e. CDVT equal to 0) and equal traffic streams, each having an inter-cell arrival time of magnitude D and with initial starting times uniformly distributed over an interval of magnitude D. The unit of time is the time required to control one cell. In [Cost242] it is put forward that the probability of the buffer use exceeding a certain value B is equal to:

This formula will be designated here by Q£(B)|. In [COST242] it is also put forward that this formula also provides a good upper limit for the probability of excess if the streams have different inter-cell arrival times.

SUMMARY OF THE INVENTION

The invention seeks to provide in a device of the kind indicated in the preamble, which is pre-eminently suitable as CAC for ATM connections of the DBR type, in which both the discrete character of the traffic streams as well as the fact that the cell rate of each of the individual connections can show fluctuations above its nominal peak cell rate are taken into account. It distinguishes itself favourably from CAC devices based on the methods in the literature with respect to the following aspects:

• More efficient: the device according to the invention has the property that the maximum number of connections able to be mixed is not limited much more than strictly necessary to prevent buffer overflow and thereby cell loss.

• Better suited to practical situations: in the current ATM networks, it is customary to select the CDVT such that this is a constant when it is expressed in the unit of time, which has as a result that the previously defined maximum burst rate b^x is dependent upon the peak cell rate of the connection; for this reason no unequivocal maximum burst rate can be established, which is a point of departure for the methods in the literature.

• Better practical applicability: the method used by the device according to the invention is practically applicable, numerically stable and has a faster calculation time.

Although the methods in the literature are also practically applicable, the CAC device according to the invention combines a high efficiency with practical applicability. The CAC device according to the invention also provides the possibility of optimising the maximum number of allowable connections by selecting a certain value for the freely selectable constant K, albeit that this must be a whole number greater than or equal to 0. In many cases, the optimal value for K will depend on the most frequently occurring values of the CDVT of the ATM connections. The CAC device according to the invention comprises two sub- devices, each issuing a positive or negative result (signal) .

The result of the whole device will be positive only if both sub-devices issue a positive result (signal) , and only then will a new connection to the ATM switch be grantable. The first sub-device determines whether sufficient bandwidth is available for the set of connections V, of which each connection vi is characterised by values for PCRi and CDVTi, which, after possible admission of the new connection, will make use of a certain output port; the second sub-device determines whether the available buffer capacity for the related port is sufficient.

The first sub-device comprises:

• An arithmetic unit which determines the sum (∑PCR) of the values of PCRi of each of the connections Vi of the set V.

Said sum is subsequently compared with the product of p and C. If the said sum is less than or equal to said product, then the result of the sub-device is positive; otherwise the result is negative.

The second sub-device comprises:

• An arithmetic unit which calculates a certain buffer size b_s,i for each of the connections Vi of the set V. Said buffer size is equal to either b_max,i minus the constant value K if b_max,i is greater than or equal to K, or equal to 0 if b_max,i is less than K. For buffer size b_maχ,i it holds that it is possible, in an ideal situation, to conduct the related traffic stream without cell loss over an imaginary buffer with buffer size _max,i and exit speed PCRi (without mixing with other traffic streams) . b_max,i is calculated by the device by determination of the product of PCR and CDVTi.

• An arithmetic unit which determines the sum of the calculated values b_s,i for all connections of the set V. Said sum will be designated here by B_s. • An arithmetic unit which determines a buffer size B_N, for which it holds that it is possible to multiplex N imaginary, identical traffic streams with constant cell rate, using a buffer of size B_N, and indeed such that the probability of cell loss as a result of buffer overflow will not exceed the given value of ε. N is here the number of connections in set V. It is also assumed that the average degree of occupation at the output of the buffer is equal to p. The buffer capacity B_N is determined by application of the previously introduced NxD/D/1 model, where the probability of a buffer level of B cells being exceeded is designated by Q_D(B)I. The arithmetic unit determines the lowest value of B, such that Q_D(B)|< ε, where N is equal to the number of sources, D is equal to N / p and Q_D(B)I is the formula as given above or a sufficiently accurate approximation thereof. The value of B found in this way forms the value of B_N.

• An arithmetic unit which multiplies the calculated value B_N by the constant K. For this result, to be referred to here as B_NK, it holds that it is possible to multiplex N imaginary, identical traffic streams, each with a burst rate of K, using a buffer of size B_NK, such that the probability of cell loss as a result of buffer overflow will not exceed the given value of ε. In this regard it will be assumed that the average degree of occupation at the output of the buffer is equal to p.

• An arithmetic unit which determines the sum of the calculated values for B_s and Bκ- Said value is designated here by B_R.

• A device which compares the calculated value of B_R with the given capacity B of the output buffer. If the value of B_R is less than or equal to B, then the result of the sub-device is positive; if the value of B_R is greater than B, then the result of the sub-device is negative.

The operation of the device will be further explained hereinafter with reference to an implementation example.

DESCRIPTION OF THE FIGURE

Fig. 1 shows a CAC device in general. The traffic parameters (PCR and CDVT) of an arbitrary number of already existing ATM connections Vι...v_N-ι are supplied to the CAC device, as well as the traffic parameters of a newly requested ATM connection v .

A maximally allowable cell loss probability parameter ε is furthermore supplied, as well as the applicable buffer capacity B and link capacity C. A maximally allowable degree of occupation p is also supplied. The result of the device is an

"admit/reject" signal, on the basis of which the requested ATM connection v_N is or is not admitted to the ATM switch.

Fig. 2 shows a diagrammatic overview of a CAC device according to the invention. The CAC device is composed of a sub-device 1 and a sub-device 2. Each sub-device can emit a positive ("admit") or a negative ("reject") signal. Both signals are supplied to a logical AND port 3, of which the output signal is only positive if both input signals are positive. Fig. 3 shows sub-device 1 in more detail. The PCR values of the N ATM connections are summed (∑PCR) in a device 4. In a device 5, the total capacity C of the output port is multiplied by a previously established coefficient p (0<p<l) , corresponding to the desired maximum load. The result of the arithmetic unit 6 is positive if the value of ∑PCR calculated in device 4 is less than or equal to the product of p x C calculated by device 5; if that is not the case, the result is negative.

Fig. 4 diagrammatically shows the sub-device 2. The traffic parameters of ATM connection i are formed by the entities Peak Cell Rate(PCRi) and Cell Delay Variation (CDVTj_.) . For each connection the sub-device 2 comprises a processor 7, which is shown in more detail in Fig. 5. In a device 13 (see Fig. 5), each of the processors 7 calculates by multiplication the maximum burst rate b_maX/i of the two said traffic parameters. The difference between b_max,i and K (b_max,i - K) is calculated by a device 14. If this difference is 0 or positive, which is verified by a device 15, then b_s,i is equal to the calculated difference of b_max,i - K; if the difference is less than 0, b_s,i is equal to 0. Sub-device 2 (see Fig. 4 again) consists of a number of processors 7, one for each ATM connection. The b_s,i results of each processor 7 are summed in a device 8 and result in a total value of B_s. The total number of ATM connections N is entered in arithmetic module 9. Said module determines the lowest value of B, such that QS(B)|< ε. This results in a value B_N. Said value B_N is multiplied by the constant value K in a device 10 and forms the value B_NK. Finally, a device 11 sums the values of B_s and B_NK and a device 12 compares the summed value with the given available buffer size B. If B_S+B_NK is less than or equal to B, then a positive "admit" signal, or otherwise a negative "reject" signal, is emitted.

REFERENCES

[1.371] ITU-T 1.371: Traffic Control and Congestion Control in B-ISDN; ITU-T recommendation 1.371 (08/96); Geneva, August 1996.

[E.736] ITU Telecommunication Standardization Sector, Draft Recommendation E.736, "Methods for Cell Level Traffic Control in B-ISDN", 10 January 1997.

[Mign96] J. Mignault, A. Gravey, C. Rosenberg, "A survey of straightforward statistical multiplexing models for ATM networks", Telecommunication Systems 5 (1996) 177-208.

[Grav97] A. Gravey, J. Boyer, K. Sevilla, J. Mignault, Resource Allocation for Worst Case Traffic in ATM networks, Performance Evaluation 30 (1997), 19-43. [COST242] J. Roberts, U. Mocci, J. Virtamo (eds.), Broadband Network TeleTraffic -performance evaluation and design of broadband multiservice networks- Final Report of Action, COST 242, Lecture Notes in Computer Science Vol 1155, Springer- Verlag, Berlin; Heidelberg, 1996, ISBN 3-540-61815-5 [Cidon95] I. Cidon, R. Guerin, I. Kessler and A. Khamisy,

Analysis of a statistical multiplexer with generalized periodic sources, Queuing Systems 20 (1995) 139-169.

[1.356] ITU-T 1.356: B-ISDN ATM layer cell transfer performance; ITU-T recommendation 1.356 (10/96); Geneva, October 1996.

Claims

1. A device for Connection Admission Control for an ATM switch, intended for admitting to the switch a requested ATM connection of the DBR type, such that the "Quality of

Service" of all ATM connections of the DBR type in said switch continues to satisfy certain conditions, in which a single buffer with capacity B is available for the composite traffic stream of DBR connections at an output port with capacity C, while as a boundary condition it holds that the total average load of the output port does not amount to more than p x C, where p is a constant with a value between 0 and 1, characterised in that the device comprises two sub- devices which each emit a positive or negative admission signal, admission to the switch only being granted to the new ATM connection if both sub-devices issue a positive admission signal, said first sub-device (1) comprising:

• a first arithmetic unit (4) which calculates the sum (Σ PCR) of the nominal traffic parameters Peak Cell Rate

(PCRi) of each of the ATM connections of the DBR type at the related output port, including the newly requested connection;

• a second arithmetic unit (6) which compares the calculated sum (∑PCR) with the value of p x C, the result of the first sub-device being negative if ∑PCR is greater than p x C, and the result of the sub-device being positive if ∑PCR is less than or equal to p x C; and said second sub-device (2) comprising:

• for each of the ATM connections of the DBR type at the related output port, including the newly requested connection, a third arithmetic unit (7), which calculates a buffer capacity b_S/i, the value of b_s,i being equal to zero if the product of the nominal Peak Cell

Rate (PCRi) and Cell Delay Variation Tolerance (CDVT_±) of the related connection is less than or equal to the constant K, and the value of b_s,i being equal to said product minus the value of K if said product is greater than K;

• a fourth arithmetic unit (8) which calculates the sum (B_s) of the calculated values b_s,i for all ATM connections at the related output port; • a fifth arithmetic unit (9) which calculates a buffer capacity B_N, such that upon multiplexing of N independent, identical and ideal (CDVT = 0) traffic streams, using a single buffer with a buffer capacity of B_N, and assuming a maximum link load having a value of p , the average probability of cell loss as a result of buffer overflow will not exceed the given value of ε;

• a sixth arithmetic unit (10) which calculates the product (B_NK) of the value of B_N and the constant value K; • a seventh arithmetic unit (11) which determines the sum (B_R) of the calculated values for B_s and B_NK;

• a comparison device (12) which compares the calculated sum B_R with the given capacity B of the output buffer, a positive admission signal being emitted if the value of B_R is less than or equal to B, and a negative admission signal being emitted if the value of B_R is greater than B.