JP2010011285A - Network topology candidate listing method and apparatus, and network topology designing method, system, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To remarkably reduce calculating time required for generation of network topology candidate set at the time of selecting adequate network topology, considering simultaneously a plurality of evaluation standards. <P>SOLUTION: A node data/restriction upper limit value inputting apparatus 101 stores the pre-input restriction condition data (when positions of all nodes forming a network are given and a link can be set among any nodes of all nodes, possibility in connection among all nodes can be maintained even under the normal operating condition and the condition where a fault occurs in any single link) to a memory. A candidate topology generating apparatus 102 reads the restriction condition data stored in the memory and generates all topologies which can be structured under the restriction condition as a candidate topology set. An optimum topology selecting apparatus 104 selects the adequate network topology only for such candidate topology set, considering simultaneously the plurality of evaluation standards. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a network topology design technique, and in particular, efficiently specifies a topology candidate to be obtained when an optimum network topology is determined by simultaneously considering a plurality of evaluation measures having different units using AHP or the like. The present invention relates to a suitable technique.

  Physical topology design of how to configure physical network topology for telecommunications carriers and some ISPs (Internet Service Providers) that manage and operate physical resources of networks such as optical fibers and routers Is an important issue.

  The physical topology is a factor that determines the facility costs of nodes and links, network management / maintenance costs, reliability against failures, and the like. In addition, if a logical path such as LSP (Label Switched Path) or wavelength path cannot be installed and the route to which the packet is transferred cannot be set explicitly, the physical topology directly affects the link load or the packet route. This is a factor that determines user quality such as transmission delay quality and throughput.

  On the other hand, when a logical path is installed on a physical network such as an MPLS (Multi Protocol Label Switching) network or a wavelength path network, and a path path between grounds can be explicitly designed, the physical topology is designed by logical topology design. The influence on the user quality can be eliminated to some extent. In this case, in addition to the physical topology design, it is necessary to appropriately design the logical topology. In this way, network topology design problems can be broadly divided into [physical topology design problems] and [logical topology design problems]. The outline of each will be described below.

  [Physical topology design problem] Consider three types of edge nodes, core nodes, and links as generalized components of a network. The edge node is a node for switching traffic and accommodates end hosts either directly or via a lower network. On the other hand, the core node is a node having only a relay function and does not accommodate users. The link plays a role of connecting edge nodes and core nodes. It is assumed that an AC traffic matrix between arbitrary edge nodes is given. Telecom operators already have civil infrastructure such as pipelines and stations that are arranged in consideration of geographical characteristics, and it is not realistic to redesign them. Therefore, the position of the edge node and the position where the core node or link can be installed are given.

  It is a practical problem to select these existing pipelines and stations and examine where to place equipment such as optical fibers and routers in accordance with changes in the requirements of the network. Physical topology design Can be defined as “an act of selecting a candidate position to be actually installed for a given core node and link installation candidate position”. However, edge nodes are installed at all edge node installation positions.

  [Logical topology design problem] When an MPLS router or OXC (Optical Cross Connect) is used, a virtual path (LSP or wavelength path) can be installed on a physical network, and a route through which traffic between each ground flows can be established. Can be explicitly designed. The physical topology is difficult to change once constructed, and the design cycle ranges from one year to several years, but the logical topology can be easily changed, and several years according to changes in AC traffic It is possible to perform flexible operations such as dynamically changing paths at intervals of several hours.

  The logical topology design is what kind of route a virtual path is set up for a given physical topology and an AC traffic matrix between edge nodes. Since the design is based on physical network resources, it is usually necessary to design in consideration of various constraints such as link capacity.

  When the logical path can be installed on the physical topology, the logical topology can be designed at the same time when the physical topology is designed. Since the design period of the logical topology is much shorter than that of the physical topology, only the logical topology is designed after the operation is started.

  It is desirable for the user to avoid congestion at the transit node and to shorten the transmission delay so that the total distance of the route is short. From the viewpoint of utilization efficiency and reliability of transmission resources, it is desirable to avoid traffic concentration on a specific link and to distribute the traffic load to all links as much as possible.

  However, reducing the number of installed nodes and links to reduce equipment costs and management / maintenance costs reduces the flexibility of path setting, making it difficult to properly distribute traffic and reduce path length. The possibility that the load is concentrated on a specific link and the possibility of losing connectivity between nodes in the event of a failure of the link or node is increased, resulting in a decrease in reliability.

  When designing a network topology in this way, it is necessary to simultaneously consider a plurality of evaluation measures that are different in unit, such as the degree of traffic distribution, the average path length, the network cost, and the reliability, and that are mutually contradictory. .

  In such a technology for making a rational decision making considering a plurality of evaluation measures at the same time, it is used in the field of microeconomics described in Non-Patent Document 1 as a technology used for network topology design. The concept of Pareto Optimization, DEA (Data Environment Analysis), and the like are known.

Pareto optimal is a state in which the utility of a particular person cannot be increased any more if the utility of someone else is not sacrificed when making a certain choice. That is, in the Pareto optimal concept, there are M evaluation measures V _{I to} V _M , and if the m-th evaluation measure value of the candidate x is V _{m, x} , “V _{m, x} ≦ V _{m, y} ”and if there is _{m such} that“ V _{m, x} <V _{m, y} ”, the candidate x is said to dominate the candidate y in a Pareto manner (however, all evaluations The smaller the scale, the better). Then, it is assumed that all candidates for which there are no other controlled candidates are Pareto optimal.

  However, since there are a large number of candidates determined to be optimal by Pareto optimization, it is difficult to effectively narrow down the candidates.

  DEA is an evaluation method that has been developed mainly as a method for evaluating investee entities, and it is possible to obtain efficiency from the attitude of evaluating each business entity's strength (evaluation scale). Is also being applied to network topology design.

  However, DEA has an intuitively difficult concept of efficiency, and it is difficult to determine the optimality of the obtained solution. In addition, the efficiency calculation needs to solve a linear programming problem, and there is a problem in scalability.

  In addition to these Pareto optimal techniques and DEA, as a technique for utilizing rational decision making considering a plurality of evaluation measures at the same time for network topology design, AHP (Analytical Hierarchy), which will be described later, was proposed by the present inventor. There is a technique (Japanese Patent Application No. 2006-353926) using a process (hierarchical analysis method).

  According to this network topology design technology using AHP, a plurality of evaluations with different units that can be intuitively understood by adding a difference to the degree of importance in the evaluation scale and reflecting the difference in the distribution form of the evaluation scale in the result. It is possible to obtain an optimum network topology while taking the scale into consideration simultaneously.

However, even in this network topology design technique using AHP, when evaluating the network topology using AHP, it is necessary to generate a candidate topology set in advance. Since the number of configurable topology candidates increases on a scale of “2 ^{N × N} ” with respect to the number of nodes N, the time required to generate a candidate set explosively increases as the network scale increases. There is.

“Network Topology Design Considering Multiple Evaluation Scales” IEICE Technical Report, vol. 106, no. 118, IN2006-31, pp. 85-90, June 2006.

The problem to be solved is that, in the conventional technology, for example, when evaluating the network topology using AHP, it is necessary to generate a candidate topology set in advance. On the other hand, since it increases on a scale of “2 ^{N × N} ”, the time required to generate a candidate set increases explosively with an increase in the network scale.

  An object of the present invention is to solve these problems of the prior art and provide an efficient network topology candidate generation technique capable of greatly reducing the calculation time required for generating a network topology candidate set.

  In order to achieve the above object, in the present invention, (1) given the positions of all nodes, and it is possible to install links between arbitrary nodes, all normal and any single link failure All configurable candidate topology sets are generated under the condition that connectivity between nodes is maintained. Specifically, (2) iteratively dividing the original problem that enumerates all the topologies that satisfy the constraint conditions into child problems that determine the presence or absence of link installation for some link installation possible positions. When the solution candidates become one, the constraint condition is determined, and when the constraint condition is satisfied, the solution is output as a solution candidate, thereby using the division method. Note that (3) the total link length, the total hop length weighted by the traffic volume ratio, and the maximum traffic ratio applied to the link are used as evaluation measures. Also, (4) by setting a constraint upper limit on the total link length, it is possible to determine that the condition is not satisfied for the entire child problem including a plurality of solution candidates, thereby reducing the time required for enumerating the solutions. Shorten. Also, (5) the time required for enumerating solutions is reduced by performing link installation determination in order from the longest distance link.

  According to the present invention, when the positions of all nodes are given and it is possible to install a link between arbitrary nodes, connectivity between all nodes is maintained during normal operation and any single link failure. It is possible to generate all configurable candidate topology sets at high speed under the condition that

  The best mode for carrying out the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration example of a network topology design system according to the present invention, FIG. 2 is an explanatory diagram showing an example of a network topology to be processed by the network topology design system in FIG. 1, and FIG. FIG. 4 is an explanatory diagram showing association information between each node shown in FIG. 2 and the accommodated population of the node, and FIG. 4 is an explanatory diagram showing basic characteristics of the network topology to be processed by the network topology design system in FIG. .

  5 is an explanatory diagram showing the relationship between the distribution of the total link length of the network topology to be processed by the network topology design system in FIG. 1 and the calculation time, and FIG. 6 is the network topology design system in FIG. FIG. 7 shows an example of a relationship between the total link length of each candidate topology to be processed by the network topology design system in FIG. 1 and the evaluation score. It is explanatory drawing.

  8 is an explanatory diagram showing a first evaluation score example of each candidate topology to be processed by the network topology design system in FIG. 1, and FIG. 9 shows an evaluation score in the process by the network topology design system in FIG. FIG. 10 is an explanatory diagram showing a second example of each candidate topology whose evaluation score is higher in the processing by the network topology design system in FIG. 11 is an explanatory diagram showing a second evaluation score example of each candidate topology to be processed by the network topology design system in FIG. 1, and FIG. 12 is an example of a hierarchical structure of AHP used in the network topology design system in FIG. FIG. 13 is an explanatory diagram showing an example of a coincidence comparison value applied in the AHP used in the network topology design system in FIG. 1, and FIG. 14 is related to the present invention. It is a flowchart illustrating a processing operation example of topology candidates listed methods and the network topology design method.

  The network topology design system in FIG. 1 has a computer configuration including a CPU (Central Processing Unit), a main memory, a display device, an input device, and an external storage device, and a storage medium such as a CD-ROM via an optical disk drive device or the like. After the program and data recorded in the external storage device are installed in the external storage device, the information is read from the external storage device into the main memory and processed by the CPU, whereby the node information / constraint upper limit value input device 101, candidate topology generation device 102, candidate Functions in each of the topology output device 103, the optimum topology selection device 104, the evaluation scale input device 105, and the optimum topology output device 106 are executed.

  The node information / constraint upper limit information input device 101 inputs the positions of all nodes, the population accommodated by each node, and the constraint condition information, and stores them in a storage device (not shown). All candidate topologies that satisfy the above conditions are generated, and the generated candidate topologies are output from the candidate topology output device 103 to the optimum topology selection device 104. The optimum topology selection device 104 is stored in advance in the storage device in the evaluation scale input device 105. Among the candidate topologies output from the candidate topology output device 103 using the constraint condition information, those satisfying the constraint conditions are selected and output via the optimum topology output device 106.

  The node information / constraint upper limit value input device 101, the candidate topology generation device 102, and the candidate topology output device 103 constitute a device that executes the network topology candidate enumeration method according to the present invention, and further includes an optimum topology selection device 104, By adding the evaluation scale input device 105 and the optimum topology output device 106, a system for executing the network topology design method according to the present invention is configured.

  In other words, the node information / constraint upper limit value input device 101, the candidate topology generation device 102, and the candidate topology output device 103 are devices that execute programmed computer processing, and have appropriate network topologies while simultaneously considering a plurality of evaluation measures. The node information / constraint upper limit value input device 101 performs a process of narrowing down the number of topology candidates generated in advance at the time of selection, and the node information / constraint upper limit value input device 101 is given preliminarily input constraint condition information (positions of all nodes constituting the network). In the case where it is possible to install a link between any of the previous nodes, the connectivity between all the nodes is maintained during normal operation and during any single link failure). The candidate topology generation device 102 reads the constraint condition information stored in the storage device, and all of the configurable conditions can be configured under the constraint condition. Generates a topology as a candidate topology set, candidate topology output device 103 outputs the candidate topology set that is generated and stored into the storage device, or directly to the best topology selection device 104.

  Note that the candidate topology generation apparatus 102 of this example uses a partitioning method as an enumeration algorithm used to generate a candidate topology set, and can install some links for the original problem of enumerating all topologies that satisfy the constraint conditions. It is repeated that the link problem is divided into sub-problems that determine the presence or absence of links, and when the candidate solution becomes one, the constraint condition is determined. To generate a candidate topology set.

  Also, in this example, as an evaluation measure that is input by the evaluation measure input device 105 and used by the optimum topology selection device 104, the total link length, the sum of the hop lengths weighted by the traffic amount ratio, and the maximum traffic ratio applied to the link Is used.

  Furthermore, in this example, a restriction upper limit is set for the total link length, and the optimum topology selection apparatus 104 determines whether or not the total link length of the place where the link is installed exceeds the restriction upper limit. The subsequent division processing is stopped. Thereby, the time required for enumerating solutions can be shortened.

  Further, in this example, the optimum topology selection device 104 performs link installation determination in order from the longest distance link. Thereby, the time required for enumerating solutions can be further shortened.

  The optimum topology selection device 104, the evaluation scale input device 105, and the optimum topology output device 106 of this example perform the topology design of the network by executing programmed computer processing. 105 stores, in the storage device, a plurality of pieces of evaluation scale information used for specifying the network performance inputted in advance, and the optimum topology selection device 104 reads out the plurality of pieces of evaluation scale information stored in the storage device. AHP described below for only the topologies included in the candidate topology set generated by the candidate topology generation device 102 and output from the candidate topology output device 103 when selecting an appropriate network topology while simultaneously considering Is used to select the optimum network topology.

  Hereinafter, as a technique related to the network topology design system of this example having such a configuration, the technique “network topology design method and design system using AHP” proposed by the inventor of the present application described above (Japanese Patent Application No. 2006-353926). No.) will be explained.

  This technology (network topology design method and design system using AHP) is based on the above-mentioned conventional Pareto optimal concept, DEA, method using average rank, etc. It solves the problems in doing.

  In AHP, a plurality of candidates (“alternatives”) for achieving a certain purpose (solving a “problem”) and evaluation criteria (“evaluation scale”) for selecting candidates (alternatives) Decide and represent them in a hierarchical diagram, and identify candidates (alternatives) suitable for the purpose (problem).

  Thus, in AHP, related elements are grasped by a hierarchical structure, qualitative measures such as importance of evaluation measures are quantified by pair comparison, and rational decision making is possible. By applying to network topology evaluation and using AHP, it is possible to design an appropriate network topology according to the importance of the designer's subjective evaluation scale.

  As described above, in AHP, there are usually three types of elements in decision making: “Problem P”, “Evaluation Scale V”, and “Alternative A”. In AHP, these elements are grasped in a hierarchical structure as shown in FIG.

However, the elements related to each other are connected by a straight line (only a part is shown in FIG. 12 for the sake of convenience), and the evaluation scale V has a plurality of hierarchical structures V ₁ , V ₂ ,... (2 hierarchies in FIG. 12). It is also possible to take

The Strength of association between elements a (weight) calculated, and finally, to derive the score S _i of each alternative A _i.

  When calculating the weight for the problem P of the evaluation scale V, there is no quantitative evaluation value from the beginning, and it is necessary to quantify it by some method.

  In AHP, quantification is achieved by comparing a pair of importance levels between all elements for an element set belonging to the same hierarchy.

First, the element _{X i} and _{X j} belonging to the hierarchy c, is set to indicate 13 the relative importance _{a ij} for _{X j} of _{X i.}

If the original weight of the element X _i is w _i , the relationship “a _ij = w _i ÷ w _j ” is ideally satisfied.

When a paired comparison matrix having importance a _ij as an element is A and a vector w having an element weight w _i as an element is multiplied from the right side of the matrix A, “Aw = nw” is established. However, n is the number of elements. Therefore, the weight vector is an eigenvector of the matrix A, and n is an eigenvalue (maximum eigenvalue).

Actually, since it is difficult to perform a pairwise comparison with perfect consistency, it is necessary to determine the consistency of the matrix A. When the maximum eigenvalue of the matrix A is λ _max , λ _max ≧ n.

Therefore, consistency is determined using the degree of consistency “CI” defined by the formula “CI = (λ _max −n) ÷ (n−1)”.

As the consistency increases, λ _max decreases, and when perfect consistency is achieved, “λ _max = n”. Therefore, the smaller the “CI” obtained by the above formula, the higher the consistency, Generally, if “C.I.” is 0.1 or less, it is judged as acceptable.

The weight of the element i in the hierarchy c with respect to the element j in the hierarchy c-1 is w _ij ^c, and among the elements in the hierarchy c-1, the set of elements related to the element V _i ^{c in the} hierarchy c is Φ _i ^c. , the score _S ^{i c} to the problem P of _V ^{i c} can be calculated from equation 1 below.

In the hierarchy 1, the weight of each element with respect to P becomes the score S _i ¹ as it is. Hereinafter, the hierarchical c = 2,3, it is possible to calculate the score _S ^{i c} in the order of ..., finally, to obtain a score _{S i} of each alternative _{A i.} The alternative with the highest score is the best option.

When such AHP is applied to network topology design, the problem P (in this case, “optimal network selection”) to be solved is shown in the top hierarchy (hierarchy 0), and the evaluation metric is shown in the central hierarchy (hierarchy 1). V _i is and candidate topology lined bottom (Tier 2).

  The score (weight) for the problem P of the hierarchy 1 (evaluation scale) can be obtained by paired comparison. The score (weight) for each element (evaluation scale) in hierarchy 2 (candidate topology) does not need to be evaluated by pairwise comparison when the evaluation scale has a quantitative evaluation value.

AHP highly evaluates elements with high scores. Therefore, when each evaluation scale is so good that it takes a small value, the reciprocal of the value V _ij of the evaluation scale j of the candidate topology i is set to X _ij (X _ij = 1 ÷ V _ij ), and based on X _ij Calculate the weight.

Weight _{w ij} ² of the hierarchy of elements 2, normalization condition ^{_{"Σ i = 1 N2 w ij 2}} = 1 " _{(N 2} is the number of layers of the element 2) it is necessary to satisfy the usually the _{X ij} In AHP Since values normalized among all candidates are regarded as weights, “w _ij ² = X _ij ÷ Σ _{k = 1} ^N2 X _kj ”.

However, the network topology the number of candidates, since a very large number, the denominator becomes large, the difference between 111 different values of X _{i j} between candidates, a much smaller value compared to the score difference between the rating scale. As a result, there is a strong tendency to select a candidate whose evaluation scale that is simply regarded as important is good, and values of other evaluation scales are hardly considered.

In order to solve such a problem, the inventor's proposal of the present application does not use the evaluation scale value X _ij itself, but weights a value obtained by normalizing a value Y _ij obtained by linearly converting X _ij .

That is, a and b are arbitrary real numbers, and “Y _ij = a (X _ij + b)”. At this time, the weight w _ij ² is represented by the following equation.

w _ij ² = a (X _ij + b) ÷ Σ _{k = 1} ^N2 a (X _kj + b) = (X _ij + b) ÷ (Σ _{k = 1} ^N2 X _kj + bN ₂ )

For normalization, w _ij ² is independent of a. Also, here, a value obtained by multiplying the minimum value among all candidates of X _ij by “−1” is set to b so that all weights take values of “0” or more (b = −min _i {X _ij }). ).

In this way, by performing linear transformation, it is possible to widen the difference in weight between candidates, and in particular, it is possible to greatly reduce the weight for candidates with large evaluation values (small X _ij ). The problem of falsely evaluating a topology that includes a very bad rating scale can be avoided.

However, when evaluating a network topology using such AHP, it is necessary to generate a candidate topology set in advance, and the number of configurable topology candidates is “2 ^{N × N} ” with respect to the number N of nodes. Since it increases in scale, there is a problem that the time required to generate a candidate set increases explosively as the network scale increases.

  In the processing of the network topology candidate enumeration system of this example, such a problem can be solved, and the required calculation time required for generating the network topology candidate set can be greatly shortened.

  Hereinafter, details of the processing (network topology candidate enumeration method) of the network topology candidate enumeration system of the invention of this example will be described.

  First, the [candidate topology generation method] will be described.

  Consider nodes and links as generalized components of a network. Assume that the node positions and the AC traffic matrix between the nodes are given, and it is possible to establish links between all the nodes.

  That is, assuming that the number of nodes is N, the number L of places where links can be installed is “L = N (N−1) / 2”. However, it is assumed that the installation link is bidirectional, and traffic flows in both directions between nodes at both ends.

With respect to these L link installable positions, a candidate topology is generated by selecting a location where a link is to be installed, so the total number of configurable candidate topologies is “2 ^L ”.

  However, the network topology requires at least connectivity between all nodes.

  Also, failures occur on various links on a daily basis, and when a network is operated as a commercial service, high reliability is required to maintain connectivity between all nodes even in the event of a failure. Further, in an IP (Internet Protocol) backbone network, about 70% of unintended failures except for those caused by maintenance are single link failures.

  Therefore, here, it is assumed that the connectivity between all the nodes is maintained even in the case of an arbitrary single link failure in addition to the normal time, and only the topologies satisfying this constraint condition are listed as candidates.

  Next, [Evaluation Scale] will be described.

  As a packet transfer route, a minimum cost route by OSPF (Open Shortest Path First) is set, and when there are a plurality of minimum cost routes, it is assumed that traffic flows evenly distributed to all.

Various evaluation measures can be considered, but in this example, one measure related to cost (total link length V ₁ ) and two measures related to quality (hop length V ₂ weighted by the traffic amount ratio). Consider three measures of the maximum traffic ratio V ₃ ) applied to the link.

(1) Total link length (V ₁ ): Total length of the link to be installed, where d _e is the distance of the link e and E is a set of installed links, the total link length V ₁ is expressed by the following equation (2) It is defined by an expression.

(2) Sum of hop lengths weighted by traffic volume ratio (V ₂ ): When the amount of AC traffic originating from node i and arriving at node j is T _ij , and the amount of AC traffic of the entire NW (network) is T Assume that a ratio r _ij (r _ij = T _ij ÷ T) of the traffic amount T _{ij to} the total traffic amount T is given. V ₂ is defined as the sum of the hop lengths of the flows weighted by the traffic ratio.

That, V ₂ is the average over the hop length per unit traffic, represented by the numerical formula 3 below. In Equation 3, V is a node set, σ _ij is the minimum number of cost paths between nodes i and j, P _ij is the set, and h _{ij and p} are hop lengths of the path p.

When links are installed at all L link installable positions, the hop length of the path between all nodes is “1”, and V ₂ takes the minimum value “1”.

(3) Maximum traffic ratio (V ₃ ) applied to the link: For each link, the sum of the traffic ratios of the flows that pass through is calculated, and the maximum value is defined as V ₃ . That is, it is defined as shown in the following equation (4). In Equation 4, σ _ij (e) is the number of paths that pass through the link e in the path between the node i and the node j.

V ₃ takes the maximum value “1” when the AC traffic between all nodes passes through a single link, and V ₃ decreases as the degree of load distribution between the links increases.

It is desirable that all of the evaluation scales V ₁ , V ₂ , and V ₃ described above are smaller. However, as a general trend, the more suppressed the number of installed links to a small number, while the total link length V ₁ is the reduction net cost decreases, the flexibility of the path set by increased over the number of hops of the flow reduction However, since the total hop length V ₂ and the maximum traffic ratio V ₃ increase, there is no network topology in which all evaluation measures are optimal.

  Next, [Calculation amount and enumeration problem] will be described.

  When generating a candidate topology set, it is necessary to determine whether or not each candidate topology satisfies the constraint condition of connectivity between all nodes at the time of normality and an arbitrary single link failure.

Moreover, since network topology evaluation using AHP is performed based on the evaluation scale value of each candidate topology, it is necessary to calculate _three evaluation scales V ₁ , V ₂ , and V ₃ for each candidate topology.

  Here, the amount of calculation required for each candidate topology is considered.

  In a topology where connectivity between all nodes is satisfied in the event of a single link failure, connectivity between all nodes is satisfied even during normal operation. Only connectivity determination needs to be performed. Hereinafter, this constraint condition is referred to as condition C.

  In order to determine connectivity at the time of a single link failure of an arbitrary link e included in the link set E, Dijkstra's minimum cost path calculation algorithm is applied to the topology excluding the link e from the target topology, What is necessary is just to determine whether the cost of the minimum cost path | route from an arbitrary node to all the other nodes becomes a finite value.

The calculation time required for this processing is O (N ² ) for the number N of nodes. Since it is necessary to repeat this process for all links (L), when the determination of condition C is performed for one topology candidate, the calculation time of O (LN ² ) = O (N ⁴ ) as a whole Is required.

Next, the time required to calculate the above-described three evaluation measures V ₁ , V ₂ , and V ₃ will be considered.

Calculation of the evaluation scale V ₁ (total link length) requires time of O (L) = O (N ² ).

Since the amount of calculation for calculating the evaluation scale V ₂ (the total hop length) is proportional to the number of node pairs, the calculation time for O (N ² ) is still required.

If the average hop length of the path between all nodes is h, the calculation of V ₃ (maximum traffic ratio) takes O (N ² h), but the average hop length h is compared with the number N of nodes. Since it is much smaller, it can be regarded as a calculation amount of O (N ² ).

Therefore, the determination of the constraint condition C is a factor that determines the overall calculation time, and the calculation time of O (N ⁴ ) is required for each candidate topology.

An intuitive method for generating a set of candidate topologies that satisfy the constraint condition C is to determine the constraint condition C for all 2 ^L possible solutions, and three for all solutions that satisfy the condition C. One evaluation measure is V ₁ , V ₂ , V ₃ . Hereinafter, this method will be referred to as an all checking method.

  However, in all investigation methods, the calculation time increases exponentially with the increase of link installable position L (the number of nodes N squared). For example, two 3 GHz CPUs and 8 GB memory are installed. When calculating with a personal computer (PC), the calculation time of about 25 years is required when the number of nodes N = 6, about 30 minutes when the number of nodes N = 7, and about 1000 years when the number of nodes N = 10.

  For this reason, all such survey methods cannot be applied to large-scale networks. Therefore, the enumeration problem described below is used.

  In general, the problem of outputting all candidates that satisfy a certain condition without duplication is called an enumeration problem. In the enumeration problem, the problem is how to reduce the time required for enumeration while satisfying both the completeness of enumerating all candidates that satisfy the conditions and the uniqueness of enumerating without duplication.

  Regarding the construction of the enumeration algorithm, three techniques of a backtrack method, a division method, and an inverse search method are mainly known.

  The backtrack method searches for a solution set without duplication and returns to the case where the solution set protrudes from the solution set, thereby eliminating the search for the solution set that does not satisfy the condition and reducing the calculation time.

  This technique requires that a set of solutions satisfying a condition exists in a connected region, and is applicable only when the solution set has monotonicity. However, this technique cannot be used because a set of topologies that satisfy the condition C does not have monotonicity.

  The partition method is an enumeration method that divides a solution set of a problem and recursively solves each as a child problem. As a result of repeating the division, the number of solutions included in the child problem is only one. The solution is output in stages.

  When it can be determined that the child problem generated by the division does not have a solution that satisfies the condition, the division of the child problem is interrupted and returned to reduce the calculation time.

  In order to apply this division method, it is necessary to solve the problem recursively, and in order to reduce the calculation time, there is no solution that satisfies the condition for a child problem including a plurality of candidates. It is necessary to be able to judge.

  In the reverse search method, a search route is first constructed, and a solution set is searched along the constructed route. This method has few restrictions on the enumeration problem that can be applied, and is often applicable to a problem to which the other two methods cannot be applied. However, it is not easy to construct an efficient search route, and there is a problem that it takes time to construct the search route itself.

  In this example, as described below, the calculation time is shortened by using the above-described “division method”.

  Hereinafter, [Enumeration of candidate topologies using the division method] will be described.

  Since the problem of enumerating network topologies that satisfy the condition C can be recursively divided by the method described below, in this example, application of the division method is attempted.

Numbers ₁ to L are arbitrarily assigned to each of the L link installable positions, and each link installable position is denoted as l ₁ , l ₂ ,..., L _L.

If a binary variable that takes “1” when a link is set at l _e and takes “0” when it is not defined is defined as x _e, it represents an arbitrary topology using an L-dimensional vector X having x _e as an element. be able to.

Let P _{0 be the} original problem of listing all topologies that satisfy condition C for L link installable positions. At this time, P ₀ can be divided into two child problems depending on whether or not a link is set at l ₁ .

Therefore, under the condition that no link is set at l ₁ , the child problem of P ₀ that lists all the topologies that satisfy condition C is P ₁ (0), and under the condition that a link is set at l ₁ Let P ₁ (0) denote another child problem of P ₀ that lists all topologies that satisfy C.

Similarly, in the range of 1 ≦ e ≦ k, when the presence / absence of a link for l _e is expressed by a k-dimensional vector X _k having x _e as an element, 1 p _e ( _k ) is expressed as P _k (X _k ). It can be divided into two child problems depending on whether or not a link is set at l _{k + 1} .

Under the terms of X _k, representing the two subproblems enumerating all topology that meets the condition C _{P k (X} k) by _{P k + 1 (X k,} 0) and _{P k + 1 (X k,} 1).

In this way, when it is determined whether or not links are partially installed for each of l _e , the combination of the presence or absence of links for the remaining l _e is recursively divided into child problems having a solution set. It is possible to apply the division method to the network topology enumeration.

The division is repeated, and the candidate topology is determined at the stage where the link installation status is assigned to all l _e of 1 ≦ e ≦ L. Therefore, the constraint C is determined. Evaluation scales V ₁ , V ₂ , V ₃ are calculated and output as one of the candidate topologies.

In order to shorten the calculation time, in a child problem including a plurality of solutions, it is determined that all candidates do not satisfy the condition C without determining whether each candidate topology set satisfies the condition C. Need to do. However, since the solution set that satisfies the condition C does not have monotonicity, the calculation time cannot be reduced only by the condition C. In this paper, we consider the newly provided that the constraint limit D on the total link length V ₁ which is one of the evaluation measure.

That is, in the child problem P _k (X _k ), the total link length (Σ _{e = 1} ^k d _e x _e ) of the locations where the links are installed in the range of 1 ≦ e ≦ k where the presence / absence of the link is already determined. ) Exceeds D, all candidate topologies included in this child problem always satisfy V ₁ > D, so that it is not necessary to perform subsequent division processing.

In addition, since the calculation of the total link length of the link installation location requires only O (k), the newly required calculation amount is negligible compared to the calculation amount of condition C. By providing a constraint upper limit on the length V ₁ , a reduction in calculation time can be expected.

As the number of nodes N increases, it is actually impractical to construct a network topology close to a complete graph in which links are installed between all nodes because the network construction cost becomes enormous. From this point of view, the provision of the constraint limit to the total link length V ₁ was considered to be a rational design Porishii.

However, by setting a constraint upper limit on the total link length V ₁ , it is expected that a generated network topology set does not include those having a large V _1, and a difference occurs in the result when AHP is applied. This will be described later in [AHP Application Result].

Also, to reduce the time required to enumerate the solution set, small range determined the establishment of the link, it is desirable that constraint limit exceeded total link length V ₁ in whatever child issues a larger number of candidates topology can be determined.

Therefore, since the label pickled can be done quite arbitrarily with respect to link the installation locations of L present, it is assumed that link distance d _e performs label pickled in descending order. As a result, d _e order is gradually being finalized presence of links installed on a large order, it can be determined constraints exceeded child problems early as possible.

  [Performance Evaluation]: Next, performance evaluation of candidate topology listing processing according to this example will be described.

  In order to confirm the effectiveness of the candidate topology enumeration method of this example, Nap. An evaluation result when enumerating candidate topologies for the node arrangement of Net and LLC will be described.

Nap. The topology of Net, LLC is shown in FIG. FIG. 3 summarizes the population of each city where six nodes exist (the population of node k is U _k ).

In order to increase the redundancy and improve the reliability and quality for the star type topology with “Node 0” as a hub with a large population, the bias link is installed between some nodes. It can be confirmed that Links can be installed position for a number N = 6 L = 15 pieces of node exists, but in FIG. 2 (b), illustrates these fifteen _{l e.}

The distance d _e of each link, here set to the linear distance between nodes. Further, the alternating traffic ratio r _ij between the nodes is proportionally distributed by the product of the populations of the nodes at both ends. That is, “r _ij = U _i U _j ÷ Σ _{sd, s ≠ d} U _s U _d ” is set.

If the total link length V ₁ without the constraint limit, the candidate topology that meets the constraints C is present 11,968 pieces. FIGS. 4A to 4C show the cumulative complementary distribution (CCD) of each evaluation scale for all candidate topologies. FIG. 4D summarizes the minimum and maximum values of each evaluation scale among all candidates.

Among the three evaluation measures, the evaluation measure (total hop length weighted by the traffic amount ratio) V ₂ has a fast CCD reduction speed, and most candidate topologies have a small value of V ₂ . This is because the number of nodes N of the network used for evaluation is as small as “6”, and in a network with a large number of nodes N, it is expected that the variation in V ₂ among candidates will be large.

  [Generation result of candidate topology set]: Next, the generation result of the candidate topology set according to this example will be described.

FIGS. 5 (a) shows a result value of the total link length V ₁ for the 11,968 pieces of all candidate topology were plotted in ascending order. However, the abscissa indicates the rank that is normalized by dividing the rank by the total number of candidates 11,968.

From FIG. 5A, only a small number of candidate topologies have a total link length V ₁ value of less than about 13 × 10 ³ km or more than about 25 × 10 ³ km, and most candidate topologies have a total link book. It can be confirmed that V ₁ is between about 13 × 10 ³ km and about 25 × 10 ³ km.

When the number of candidates topology of interest has a constraint limit D of the total link length V ₁ so as to be 10% increments from all 11,968 pieces of 10%, the time required to generate the candidate topology set 5 As shown in (b) of FIG. This figure also shows the results of all survey methods.

  From (b) of FIG. 5, in all the investigation methods, since determination is performed for all candidates regardless of the number of candidates that satisfy the conditions, the required calculation time is irrelevant to the constraint upper limit D of the total link length. When the division method is used, it can be confirmed that the required calculation time is reduced as the upper limit D of the total link length is reduced and the number of candidate topologies satisfying the condition is reduced to a small number.

  [AHP Application Result]: Next, a case where the candidate topology set generation technique of this example is applied to a network topology design technique using AHP will be described.

As described above, the time required for the configuration of the candidate topology set is reduced by using the division method, but the topology in which the total link length V ₁ exceeds the upper limit constraint D is excluded from the generated candidates. There is a concern that the evaluation results will be affected.

  Therefore, the results of applying AHP to the candidate topology set generated using the division method will be considered below.

  As examples of the relative importance of each evaluation scale, two cases are considered, where quality is important (scenario 1) and cost is important (scenario 2). FIG. 6 shows a paired comparison table between evaluation scales in each of these two scenarios (Scenario 1, 2).

The score S _i ¹ of the evaluation scale is S ₁ ¹ = 0.2, S ₂ ¹ = 0.4, S ₃ ¹ = 0.4 in the case of “Scenario 1”, and S _{1 in} the case of “Scenario 2”. _{It was 1} ¹ = 0.5, S ₂ ¹ = 0.25, S ₃ ¹ = 0.25. Consistency CI was “0” in any scenario, and the paired comparison values were determined to be consistent.

In Figure 7, when the 11,968 pieces of all candidate topology generated without providing the constraint limit D on the total link length V ₁ applying the AHP to the subject, of the total link length V ₁ of the respective candidate topology AHP The results plotted in descending order of evaluation score are shown.

When evaluation is performed with emphasis on cost (scenario 2), it can be confirmed that a topology with a short total link length is highly evaluated. Therefore, in “Scenario 2”, it is expected that even if a candidate topology is generated by providing a constraint upper limit D for the total link length V ₁ , no significant change will be seen in the topology evaluated higher by AHP.

On the other hand, if (Scenario 1) was evaluated with an emphasis on quality, clear correlation total link length V ₁ and evaluation rank is not observed. The topology is highly regarded AHP, long topology of the total link length V ₁ have also been also include shorter topology, even if the generated candidate topology imposes constraints limit D on the total link length V _1, these upper Of the topologies to be evaluated, some topologies with a short total link length V ₁ can still be expected to be highly evaluated.

FIG. 8 shows the top six topologies having the highest AHP evaluation scores in “Scenario 1”, three evaluation scale values V _ij and their relative values (NV: Normalized Value) among all 11,968 candidates. ).

However, if the maximum value and the minimum value among all candidates of each evaluation scale are max {V} and min {V}, the relative value NV is “(V _ij −min {V}) ÷ (max {V}”. −min {V}) ”. That is, the relative value NV is a value obtained by adjusting the scale so that the three topologies V _ij have the smallest topology “0 (zero)” and the largest topology “1”.

  Since all evaluation scales are desirable as the value is smaller, the smaller the relative value NV is, the closer to zero (0), the better the evaluation scale is.

A restriction upper limit D is set for the total link length V ₁ , and the generation candidate topologies are reduced to 60% (11 (968) in FIG. 8) or 20% (FIG. 8 (c)). )) shows a hand, when the FIG. 8 (b) and with respect to the (c) in FIG. 8, directed to 11,968 pieces of all candidates without providing a constraint limit D on the total link length V ₁ The evaluation order (“Rank w / o limit”) of AHP in ((a) of FIG. 8) is also shown.

As shown in FIG. 8A, when the upper limit D is not provided for the total link length V ₁ and AHP is applied to all candidate topologies that see the constraint C, the quality of “scenario 1” is Since the emphasis is on, topologies having good evaluation scale V ₂ and evaluation scale V ₃ are highly evaluated.

In general, as the number of installed links is increased, the hop length of the path between opposing nodes and the load on each link are reduced. Therefore, it is considered that the evaluation measure V ₂ and the evaluation measure V ₃ are reduced. It can be confirmed that the total link length V ₁ of the top rated topology is not a value close to “1” but a medium value.

Even if a moderate number of links are installed, there are sufficiently good topologies for the evaluation scale V ₂ and the evaluation scale V ₃ , and such an excellent topology can be selected by performing evaluation using AHP. .

  FIG. 9 shows the top four topologies. As shown in FIG. 3, the “node 0” that protrudes and has a large population is used as a hub, and the topology with a relatively high degree of other nodes is highly evaluated.

The constraint limit D on the total link length V ₁ provided, if the candidate topology produced was all 11,968 pieces of 60% to ((b) in FIG. 9), which is targeted for all candidates (in FIG. 9 Some topologies evaluated within 6th place in (a)) are also rated within 6th place.

When the constraint upper limit D of the total link length V ₁ is made stricter and the total number of candidate topologies to be generated is 11,968 (20% in FIG. 9C) ((c) in FIG. 9), the top rated top 6 topologies Does not include those rated within 6th place for all candidates.

  However, in this case as well, the top rated topologies are those with a ranking of about 200 or less when all are targeted, and among the 11,968 total topologies, the top ranking is a high relative ranking. Can be confirmed.

For this reason, it is possible to find such an excellent topology even when a candidate topology set is generated by providing a constraint upper limit D for the total link length V ₁ and evaluation is performed by AHP with an emphasis on quality.

  10 and 11 similarly show the results in “Scenario 2”. However, in “Scenario 2”, when the candidate topology is reduced to 60% and to 20%, the topologies evaluated within the top six are completely the same.

When cost is important, topologies with a small total link length V ₁ tend to be highly evaluated. Therefore, even when a restriction upper limit D is provided for the total link length V ₁ and candidate topologies are narrowed down, the evaluation result of AHP It can be confirmed that it has almost no effect.

  Next, processing operations of the network topology candidate enumeration method and the network topology design method according to the present invention by the network topology design system in FIG. 1 will be described with reference to FIG.

  First, the node information / constraint upper limit value input device 101 is given preliminarily input constraint condition information (positions of all nodes constituting the network are provided, and a link can be set between arbitrary nodes of the previous node. In some cases, connectivity between all nodes is maintained during normal operation and during any single link failure) in the storage device (step S101) and stored in the storage device by the candidate topology generation device 102. The constraint condition information is read, all the topologies that can be configured under the constraint condition are generated as candidate topology sets (step S102), and the generated candidate topology set is stored in the storage device by the candidate topology output device 103 Or directly to the optimum topology selection device 104 (step S103).

  Then, in the optimum topology selection device 104, an optimum network topology selection process is executed (step S104). At that time, the evaluation scale input device 105 inputs in advance and stores it in the storage device (step S105), reads a plurality of pieces of evaluation scale information used for specifying the network performance, and sets an appropriate network topology while simultaneously considering each evaluation scale. At the same time, an optimal network topology selection process is executed using AHP for only the topologies included in the candidate topology set generated by the candidate topology generation apparatus 102 and output from the candidate topology output apparatus 103. The optimum network topology selected by the optimum topology selection device 104 is output by the optimum topology output device 106 to an internal storage device or display device, an external communication terminal device connected to the network, or the like.

  As described above with reference to FIGS. 1 to 14, in this example, in the process of narrowing down the number of topology candidates generated in advance when selecting an appropriate network topology while simultaneously considering a plurality of evaluation scales, Configuration, provided that the connectivity between all nodes is maintained during normal operation and during any single link failure. Generate all possible candidate topology sets. At this time, using the division method, the original problem that enumerates all the topologies that satisfy the constraint conditions is repeatedly divided into child problems in which the existence of link installation is determined for some link installation possible positions. When the number of solution candidates becomes one, the constraint condition is determined, and when the constraint condition is satisfied, it is output as a solution candidate. Further, as a plurality of evaluation measures, three of the total link length, the total hop length weighted by the traffic volume ratio, and the maximum traffic ratio applied to the link are used. Furthermore, by providing a constraint upper limit on the total link length, it is possible to determine that the condition is not satisfied for the entire child problem including a plurality of solution candidates, and the time required for enumerating solutions is shortened. Moreover, the time required for enumerating solutions is shortened by performing link installation determination in order from a link with a long distance.

  As a result, according to this example, the number of topology candidates generated in advance when selecting an appropriate network topology while simultaneously considering a plurality of evaluation measures can be narrowed down at high speed, and network topology design can be performed efficiently. It becomes possible.

  In addition, this invention is not limited to the example demonstrated using FIGS. 1-14, In the range which does not deviate from the summary, various changes are possible. For example, in this example, the present invention can also be applied to a case where another technique such as EAD described using AHP is used as a technique for designing a network topology while simultaneously considering a plurality of evaluation measures.

  In addition, this example can be applied to both physical topology design and logical topology design.

  Also, regarding the computer configuration example in this example, a computer configuration without a keyboard or optical disk drive may be used. In this example, an optical disk is used as a recording medium. However, an FD (Flexible Disk) or the like may be used as a recording medium. As for the program installation, the program may be downloaded and installed via a network via a communication device.

It is a block diagram which shows the structural example of the network topology design system which concerns on this invention. It is explanatory drawing which shows the example of a network topology used as the process target of the network topology design system in FIG. It is explanatory drawing which shows the matching information of each node shown in FIG. 2, and the accommodation population of the said node. It is explanatory drawing which shows the basic characteristic of the network topology used as the process target of the network topology design system in FIG. FIG. 2 is an explanatory diagram showing a relationship between a distribution of total link lengths of a network topology to be processed by the network topology design system in FIG. 1 and calculation time. It is explanatory drawing which shows the example of a pair comparison table between the evaluation scales used as the process target by the network topology design system in FIG. It is explanatory drawing which shows the related example of the total link length of each candidate topology used as the process target by the network topology design system in FIG. 1, and an evaluation score. It is explanatory drawing which shows the 1st evaluation score example of each candidate topology used as the process target by the network topology design system in FIG. It is explanatory drawing which shows the 1st example of each candidate topology to which the evaluation score became high by the process by the network topology design system in FIG. It is explanatory drawing which shows the 2nd example of each candidate topology to which the evaluation score became high by the process by the network topology design system in FIG. It is explanatory drawing which shows the 2nd evaluation score example of each candidate topology used as the process target by the network topology design system in FIG. It is explanatory drawing which shows the hierarchical structure example of AHP used with the network topology design system in FIG. It is explanatory drawing which shows the example of the coincidence comparison value applied with AHP used with the network topology design system in FIG. It is a flowchart which shows the processing operation example of the network topology candidate enumeration method and network topology design method which concern on this invention.

Explanation of symbols

  101: Node information / constraint upper limit input device, 102: Candidate topology generation device, 103: Individual topology output device, 104: Optimal topology selection device, 105: Evaluation scale input device, 106: Optimal topology output device

Claims (10)

A method for narrowing down the number of topology candidates generated in advance when selecting an appropriate network topology while simultaneously considering a plurality of evaluation scales by an apparatus that executes programmed computer processing,
As a process execution procedure of the programmed computer,
Pre-entered constraint condition information (when the position of all nodes constituting the network is given and it is possible to install a link between any of the previous nodes, normal and any single link failure A first procedure of storing in a storage device)
A network topology candidate enumeration method comprising: a second procedure of reading the constraint information stored in the storage device and generating all the topologies that can be configured under the constraint as a candidate topology set. The network topology candidate enumeration method according to claim 1,
In the second procedure, a division method is used as an enumeration algorithm used for generating the candidate topology set,
Candidate solutions by repeatedly dividing the original problem enumerating all the topologies that satisfy the above constraints into child problems that determine the presence or absence of link installation at positions where some links can be installed A network topology candidate enumeration method characterized by generating the candidate topology set by determining the constraint condition when the number becomes 1, and outputting as a solution candidate when the constraint condition is satisfied. A network topology candidate enumeration method according to claim 2,
A network topology candidate enumeration method characterized by using a total link length, a total hop length weighted by a traffic amount ratio, and a maximum traffic ratio applied to a link as the evaluation measure. A network topology candidate enumeration method according to claim 3,
Set a limit on the total link length above,
In the second procedure, it is determined whether or not the total link length of the place where the link is installed exceeds the upper limit of restriction, and if it exceeds, the subsequent division processing is stopped. Network topology candidate enumeration method. The network topology candidate enumeration method according to claim 4,
In the second procedure, a network topology candidate enumeration method characterized by performing link installation determination in order from a link with a long distance. A network topology design method for designing a topology of a network by an apparatus that executes programmed computer processing,
As a process execution procedure of the programmed computer,
A procedure for storing a plurality of pieces of evaluation scale information used for specifying network performance, which is input in advance, in a storage device;
A procedure for reading out a plurality of evaluation scale information stored in the storage device and selecting an appropriate network topology while simultaneously considering each evaluation scale;
Each procedure in the network topology candidate enumeration method according to any one of claims 1 to 5,
In the procedure of selecting an appropriate network topology, the network topology selection processing is executed for the topologies included in the candidate topology set generated by executing each procedure in the network topology candidate enumeration method. Network topology design method. An apparatus for narrowing down the number of topology candidates generated in advance when selecting an appropriate network topology while simultaneously considering a plurality of evaluation measures by programmed computer processing,
As a means of performing programmed computer processing,
A network topology candidate enumeration apparatus comprising means for executing each procedure in the network topology candidate enumeration method according to any one of claims 1 to 5. A network topology design system for designing a topology of a network by programmed computer processing,
As a means of performing programmed computer processing,
First means for storing, in a storage device, a plurality of pieces of evaluation scale information used for specifying network performance, which are input in advance;
Second means for reading out a plurality of evaluation scale information stored in the storage device and selecting an appropriate network topology while simultaneously considering each evaluation scale;
A third means for generating the candidate topology set by executing each procedure in the network topology candidate enumeration method according to any one of claims 1 to 5;
The network topology design system, wherein the second means executes the network topology selection processing for the topologies included in the candidate topology set generated by the third means.   The program for making a computer perform each procedure in the network topology candidate enumeration method in any one of Claims 1-5.   The program for making a computer perform each procedure in the network topology design method of Claim 6.

Images