GB2253970A - Data network management - Google Patents

Abstract

In an iterative method of analysing a circuit-switched trunk network in which the network is considered to contain reliable nodes connected by unreliable links, and in which a path-loss sequence is generated for each pair of nodes, a calculation is carried out for path usage probabilities and path/loss traffic, a further calculation is carried out for link carried traffic and link blocking probabilities, and a convergence condition is carried out, whereby it the calculated line blocking probabilities have changed, relative to the last calculation by less than a critical predetermined error, then a calculation is made as to the quality of the network in terms of service available at that moment. <IMAGE>

Description

DATA NETWORK MANAGEMENT The present invention concerns the management of data networks, and in particular circuit-switched (C-S) communications networks. An example-of such a network is the Integrated Services Digital Network (ISDN).

An ISDN network consists of a number of interconnected nodes, for example exchanges or concentrators, between which data is transferred with the possibility that data originating from one node and intended for another node can follow a number of alternative routes under the control of network management.

The question of selecting the most appropriate route with regard to a network and its traffic conditions and reliability is becoming one of increasing importance. In the United Kindom one carrier has had a virtual monopoly only fairly recently changed to a still one-sided duopoly. However this situation is now changing. Given the difficulties and capital costs of establishing data paths between nodes it is probable that in the coming more competitive environment the telephony network will have to be responsive to many different operators, thus increasing the importance of network management. Furthermore the expected increase in traffic will require that network management decisions be made as quickly as possible.

Accordingly the present invention is concerned with a method and system of analysing circuit-switched networks which can provide reduced execution times with regard to already known procedures.

An aspect of the method according to the present invention is that it utilises a single moment method because of the consequent reduction of both complexity and execution time.

Accordingly the present invention comprises an iterative method of analysing a circuit-switched trunk network in which the network is considered to contain reliable nodes connected by unreliable links, and in which a path-loss sequence is generated for each pair of nodes, a calculation is carried out for path usage probabilities and path/loss traffic, a further calculation is carried out for link carried traffic and link blocking probabilities, carrying out a convergence condition whereby if the calculated line blocking probabilities have changed, relative to the last calculation by less than a critical predetermined error, then a calculation is made as to the quality of the network in terms of service available at that moment.

In order that the invention may be more readily understood an embodiment thereof will now be described by way of example and with reference to the accompanying drawings, in which: Figure 1 shows a small communication network, Figure 2 is a flow diagram showing steps taken in the implementation of the invention, Figure 3 is a routing table with respect to the network shown in Figure 1, Figure 4 is a network configuration table with respect to the network of Figure 1, Figure 5 is a Link Capacity and Availability table with respect to the network of Figure 1, and Figure 6 is a matrix showing busy-hour traffic over the network of Figure 1, Figure 7 shows two models of unreliable links and associated equations, and Figure 8 shows further equations associated with blocking probabilities.

Referring now to Figure 1 of the drawings, this shows a small communication network having interconnected nodes A, B, C, D, and E, where integers indicate link numbers.

It will be appreciated that in practical terms there will be many more nodes and that in an ISDN network these nodes will be individual telephony exchanges or concentrators. An example routing table for the network of Figure 1 is shown at Figure 3. In general the routing table will consist of an n x n matrix of entries, where n is the number of nodes in the network, except that the forward diagonal is empty. Entries in the table are ordered set of nodes (N1, N2 .... Nn) such that the entry at position (S, T) represents links (S-N1, S-N2, .... S-Nn). These are searched in a given order to find a free circuit for a call arriving at S destined for T. In this simple routing table shown in Figure 3, a call arriving at C and destined for A could pass via the links CA or CB.

What is known as a Call Control Rule is used to determine how the routing table is used to select paths through the network.

In the present embodiment four routing disciplines can be used to find paths through the network. These routines are known as Successive Office Control (SOC), Originating Office Control (OOC), Originating Office Control with Spill-Forward (OOC-SF) and Tandem Node Matric routing (TNM). In SOC, when a call reaches a given node, a free circuit will be searched for trying the first choice, and then the other links set out in the routing table. If a free link is found then control is passed to the next node over the link. Thus if in the table of Figure 3 a call originating at A destined for D would first try the direct link to D, next the link via C and if the link to C was blocked, the link via B. If the link to B was free then control would be passed to Node B. If no links are free then the call is lost.

In OOC, if a call is at its originating mode it will be routed as described for SOC. At an intermediate or tandem node however only the first choice link will be tried. If this first choice link is not available then control is returned to the originating node where the next choice link available to the originating node is tried. The call is only lost if the last choice link, and by definition all previous choices, available to the originating node is unavailable.

OOC-SF is intermediate between the two previous call control rules. It is identical to OOC unless one of the nodes in an entry in the routing table is marked as a spill-forward node.

In that case, if a call reaches a spill-forward node then call control is passed to the spill-forward node, and as far as routing is concerned, that node acts as the originating node. In Figure 3 spill forward nodes are marked by a lower-case letter in the appropriate entry in the routing table. TNM is a special case of OOC-SF which is only used in purely hierarchial networks. A description of TNM can be found in Chan, W.S. "Recursive Algorithms for Computing End-to End Blocking in a Network with Arbitrary Routing Plan" IEEE Trans Commun vCOM-28 n2 Feb 80 p153-164.

Referring now to Figure 4 of the drawings the Network Configuration Table shown in the figure is an array of link numbers, with an empty forward diagonal. In this table the (S,T) entry is either the link number of a link directly interconnecting the two nodes, e.g. 7 for C-D, or - 1 if there is no direct link.

The Link Capacity and Availability Table shown in Figure 5 contains the capacities (i.e. the number of circuits) and the availabilities of the links interconnecting the nodes of the network.

The final table, Figure 6, is the Busy-Hour Traffic Matrix which shows the busy-hour traffic in Erlang's between each node pair.

The Call Control Rules and the four tables are used to provide a model of an unreliable trunk network which can be analysed in accordance with the present invention.

The algorithm for the analysis is shown in Figure 2. As can be seen from this figure the first concrete step to be carried out is the generation of path-loss sequences and initialisation of link blocking probabilities.

Thus from the routing table of Figure 3, and the call control rule, which is effectively incorporated into the routing table by the identification of the spill-forward nodes, the path-loss sequence is obtained for each node pair. A path-loss sequence consists of the ordered sequence of paths through the network, between two nodes, that would be chosen using a given routing table and call control rule. The letter 'L' is used to represent a fictitious loss-node. For example, the path-loss sequence for node pair (D,A) in the Network of Figure 1 is [P1 (DA), P2 (DBA), P3 (DCA), P4 (DCBA), L1 (DCL), L2 (DL)). The successful paths are represented by a Pn where n is the path number, and the unsuccessful paths, the loss paths, by Ln.In the present embodiment this step of the flow diagram is achieved by the use of the Pathf Algorith, a recursive algorithm based on the work of Lin et al, which is given in a paper by Lin, P.M., Leon, B.J & BR< Stewart, C.R. entitled "Analysis of Circuit-Switched Networks Employing Originating-Office Control with Spill-Forward", IEEE Trans Commun vCOM-26 n6 Jun 78 p754-765.

From the link blocking probabilities and the path-loss sequences for all node pairs, the path usage probability is obtained for every network path (both success and loss paths). The path usage probability is the probability of a particular success path (loss path) in a node pair path-oss sequence being the one used to complete (lose) a call. These probabilities are calculated in accordance with a general purpose recursive algorithm, for networks of arbitrary size, which has been given by Chan, W.S.

"Recursive Algorithms for Computing End-to-End Blocking in a Network with Arbitary Routing Plan" IEEE Trans Commun vCOM-28 n2 Feb 80 p153-164.

To make use of Chan's work, the path-loss sequence must be converted into a link-set sequence. A link-set is simply a set of links (in link number order), and a link-set sequence is an ordered set of link-sets. Thus, for example, the (D, A) path-loss sequence.for the Network of Figure 1 corresponds to a link-set sequence Eul, U2, U3, U4, U5, U6), where: U1 = E2], U2 = EO,5), U3 = El,7] U4 = EOA,7], U5 = E7], U6 = > - the empty set. The probability of a path Pn or Ln, which corresponds to a link set Un in a link-set sequence Eul,...Um,...Ui], being used is equal to Q ([U1,...,Um]), where Q () is Chan's Algorithm 1 as set out in the paper by Chan which has already been referred to.

In the next step of the flow diagram of Figure 2 the traffic carried by each link is calculated by summing the traffic carried by all the success paths that make use of it. The link offered traffic is then obtained from the carried traffic and the current estimate of the link blocking probability. Finally, the link blocking probability is itself recalculated from the link offered traffic and the appropriate entries in the link capacity and availability table, using the Markov model of an unreliable trunk.

The Markov Model will now be discussed in relation to the modules and equations shown in Figure 7. Consider a trunk of N circuits with a mean call arrival rate of, a mean holding time of 1/, a mean trunk failure rate of fs and a mean repair time of i/JAR. All processes are assumed Poisson, and when the trunk fails, all calls are cleared. System states are represented by integer pairs, the first representing the failure condition of the trunk (0 for good, 1 for bad), and the second the number of calls in progress, up to the maximum of N.

Consider the model as N increases, concentrating on the calculation of the probability of blocking, B (N), which is equal to P (1,0) + P (O, N), where P(a,b) represents the probability of state (a,b) in statistical equilibrium. Figure 7 shows the Models for N = 1 and 2 and two equations 1 and 2 respectively associated with these Models.

Close examination of the equations B(1), 8(2), etc.

reveals a recursive formula for B(N). This formula in Figure 8 (1, 2 and 3) is given using the C tertiary operator: (boolean)? exprl: expr2, where the value returned by the operator is exprl if the boolean is true, or expr2 if the boolean is false.

In addition to the exact equations given at 1, 2 and 3 in Figure 8, an approximation to the trunk blocking probability can be made by taking the sum of the trunk unavailability and the product of the trunk availability and Erlang's Loss Formula for the corresponding reliable link as is shown in Figure 8, equation 4 in which E () is Erlang's Loss Formula.

Referring now to Figure 2, having dealt with the basic procedures shown in the steps in greater detail, this figure shows how estimated link-blocking probabilities are used with the stored path-loss sequences to provide the initial set of data for a subsequent iterative procedure.

This iterative procedure starts by the calculation, as already described, of path usage probabilities and Path/Loss traffic. From these the Link Carried Traffic and finally the Link Blocking probabilities themselves are calculated. If the Link Blocking probabilities have not converged another iteration is carried out.

The convergence condition for the overall method as set out in Figure 2 is that if all the link blocking probabilities have changed, relative to the last iteration, by less than a critical error (e.g. 0.005X) then the method is complete. On convergence the node-to-node grade-of-service (NNGOS) for each node pair is obtained from the path usage probabilities. It is equal to either the sum of the probabilities that the loss paths in the node pair path-loss sequence are used, or 1 - the sum of the probabilities that the successful paths are used.

The network grade-of-service (NGOS) is calculated from the NNGOss using the formula based on that given by Bonaventura and shown in Figure 8 (5). Bonaventure et al in Bonaventura, V., Cacopardi, S., Decina, M & Roveri, A. "Service Availability of Communication Networks" Proc 1980 Nat Telecommun Conf, Houston, Dec 80 p15.2.1-15.2.6, where G is the NGOS, Gij is the (i,j) NNGOS, and rjj is the busy-hour traffic matrix entry for node pair (i,j).

One difficulty with the model given above is that it requires five parameters: the mean call arrival rate the mean holding time l/ss,the mean trunk failure rate)iF, the mean repair time 1/ R, and the trunk capacity (number of circuits) N. In contrast, at the point in the method where the Markov model is used, only three parameters are available: the link offered traffic A, the link availability p, and the link capacity N. This is resolved by assuming a value for the mean call holding time and the mean repair time.

Claims (3)

1. An iterative method of analysing a circuit-switched trunk network in which the network is considered to contain reliable nodes connected by unreliable links, and in which a path-loss sequence is generated for each pair of nodes, wherein a calculation is carried out for path usage probabilities and path/loss traffic, a further calculation is carried out for link carried traffic and link blocking probabilities, and further carrying out a convergence condition whereby if the calculated line blocking probabilities have changed, relative to the last calculation by less than a critical predetermined error, then a calculation is made as to the quality of the network in terms of service available at that moment.

2. A method as claimed in Claim 1, wherein the quality of the network, 6, is given by:

where G is the Network Grade Of Service, Gij is the (i,j) Node-to-Node Grade of Service and rij is the busy-hour traffic matrix entry for node pair (i,j).

3. A method as claimed in Claim 1 and sybstantially as hereinbefore described.