JPH1174934A - Routing method, network node and communication system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a routing method capable of simplifying a network configuration, efficient in performance without limiting objective network topology and capable of easily coping with failure and congestion by constituting a network so that it has no database to be accessed in common by respective nodes. SOLUTION: The node 20 consists of a basic layer processing part 21 and a knowledge layer processing part 22 and outputs an input packet from an external environment 23 to the environment 23 in accordance with its destination. Each network node 20 manages probability information that is submitted to decision to select each output link in each address which can be a destination address and an evaluation value that is submitted to decision for a method that updates the probability information. Each node 20 fetches an evaluation value that corresponds to a sending source address of an input packet, compares it with a cumulative cost value in the packet, updates probability information that is related to the sending source address in accordance with a comparison result and also updates the evaluation value in accordance with the cumulative cost value in the input packet.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a routing method, a network node, and a communication system.

[0002]

2. Description of the Related Art In a communication system for transmitting a packet from a source terminal to a destination terminal via a network, a routing method for allowing the packet to pass through the network without waste is important. When such routing methods are classified, they can be classified as follows (see Document 1).

[0003] Reference 1 "Cesur Baransel et al.," Routing in Multihop Pa
ticket Switching Networks: G
b / s Challenge ", IEEE Network
rk May / June 1995; 38-41 "(a) Table-based method When a packet arrives, there is a routing method in which a node examines a routing table owned by itself or managed centrally by a network management unit, and selects an output line. This routing method includes (a1) a shortest path routing method and (a2) an optimal routing method.

[0004] The show test path routing method refers to the characteristics of the destination address, the transmission distance, the number of hops (the number of relay switches), the link capacity, and the link congestion status at each node. This is a method of selecting an output line by deciding which output line is to be optimized by a distance evaluation value in consideration of the above.

[0005] On the other hand, the optimal routing method is a method in which, at each node, the communication cost based on the link usage rate or the transmission delay is used as an evaluation value, and an output line that minimizes the cost is selected.

(B) Self-routing method The self-routing method is a method in which each node routes a packet without accessing a network database.

[0007] For example, (b1) In the case of a network having a regular topology, each node can store data without describing the position of another node in the network.
It is possible to know the network configuration and perform autonomous routing. In the case of “(b2) flooding network”, the arriving packet at each node is output to all lines other than the arriving line,
This ensures that the packet is transmitted to the destination terminal.

[0008]

By the way, in realizing the routing method, there are the following various requirements.

(1) It is preferable not to have a database that is commonly accessed between the nodes because the network configuration can be simplified.

(2) The target network topology is not limited, and it is preferable to have a degree of freedom because an arbitrary network can be formed.

(3) It is more efficient and preferable that information exchange for routing be small between the nodes. For example, it is preferable to be able to autonomously know that a node other than the added node has been autonomously added, rather than sending it to each node from an additional node or a management device, such as additional information of a node in the network.

(4) It is desirable that the routing method be capable of easily coping with a failure or congestion occurring in the network.

However, it is difficult for the conventional routing method to satisfy many of these requirements. Therefore, there is a need for a routing method that can meet many of the technical requirements for the routing method, and a network node and a communication system that realize the routing method are desired.

[0014]

According to a first aspect of the present invention, there is provided a routing method in which each network node determines an output link from its own node according to a destination address inserted in an input packet. In the method, each network node (1) manages, for each address that can be a destination address, probability information used for determining each output link and an evaluation value used for determining a method of updating the probability information. (2) When a packet having a single destination address is input from another network node, (2-1) the evaluation value corresponding to the source address of the input packet is extracted, and When comparing with the inserted cumulative cost value and obtaining a comparison result that the current route of the input packet is suitable, Among the elements of the probability information related to the source address, the value of the element related to the link to which the input packet was input was increased, the values of the other elements were reduced, and a comparison result indicating that the current route of the input packet was not suitable was obtained. When, among the elements of the probability information related to the source address, the value of the element related to the link to which the input packet was input is reduced, the values of the other elements are increased, and the evaluation value corresponding to the source address is Is updated according to the accumulated cost value inserted in the input packet, and (2-2) the link related to the element having the largest value in the probability information corresponding to the destination address of the input packet is set as the link for outputting the input packet. Decide,
The accumulated cost value of the input packet is updated by the cost corresponding to the determined link, and the updated packet is transmitted to the determined link.

According to a second aspect of the present invention, there is provided a network node for determining an output link from the own node in accordance with a destination address inserted in an input packet.
Probability information for use in determining each output link for each address that can be a destination address, evaluation value for use in determining how to update these probability information for each address that can be a destination address, and an output value corresponding to each output link (2) when a packet is input from another network node, retrieves the evaluation value corresponding to the source address of the input packet, and stores the evaluation value in the packet. When the comparison with the inserted accumulated cost value is performed and a comparison result indicating that the current route of the input packet is suitable is obtained, the input packet is input among the elements of the probability information related to the source address. Increase the value of the element related to the link, lower the value of the other element, and compare the result that the current route of the input packet is not suitable. When, among the elements of the probability information related to the source address, the value of the element related to the link to which the input packet was input is reduced, the values of the other elements are increased, and the evaluation value corresponding to the source address is Information updating means for updating the value according to the accumulated cost value inserted in the input packet, and (3) outputting the input packet to the link related to the element having the largest value in the probability information corresponding to the destination address of the input packet. Packet output means for determining the link, updating the accumulated cost value of the input packet by the cost corresponding to the determined link, and transmitting the updated packet to the determined link.

Furthermore, a third invention is characterized in that it has a plurality of network nodes according to the second invention, and these network nodes are connected according to an arbitrary network topology.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a routing method, a network node and a communication system according to the present invention will be described below in detail with reference to the drawings.

Although not shown, the communication system (network) of this embodiment also includes a plurality of network nodes (hereinafter simply referred to as nodes) such as exchanges and terminals, and links that appropriately connect the nodes. The packet from the source terminal is appropriately routed and transferred to the destination terminal. In the case of this embodiment, the topology of the communication system (network) may be arbitrary.

Each node has substantially the same configuration as the conventional hardware. That is, each node
As shown in FIG. 1, an input unit 10 for receiving and processing a packet from another node or a transmission terminal, a switch unit (built-in buffer memory) 11 for exchanging a packet received by the input unit 10, and a switching unit 11 The output unit 12 that sends the received packet to another node and the header of the packet received by the input unit 10 (the entire control packet if the packet is not an information packet but a control packet) are provided. The control unit 13 includes a control unit 13 that performs other control operations, and a database unit 14 that stores various information used by the control unit 13 in various control operations.

However, each node of this embodiment has a detailed configuration (functional model) shown in FIG. 1 from the viewpoint of routing processing.

In FIG. 1, the node 20 is roughly
It comprises a basic layer processing unit 21 and a knowledge layer processing unit 22, and outputs an input packet from the external environment 23 to the external environment 23 according to its destination. Here, the external environment 23 generically refers to nodes other than the own node 20.

The base layer processing unit 21 includes a stochastic learning automaton (Stochastic Learning A).
The output link selection probability is learned by an autonomous learning mechanism based on automata (hereinafter abbreviated as SLA; see document 2), and based on the output link selection probability that is appropriately updated through the learning operation, the output for the current input packet is output. Determine the link. That is,
It performs routing processing in unicast communication.

Reference 2 "Masao Kubo et al.," Evaluation of cooperative operation of multi-agent system by ant bait game ",
IPSJ Transactions, Vol. 85, no. 8,199
4 / August ”On the other hand, the knowledge layer processing unit 22
The present invention realizes a routing function and the like in advanced communication other than unicast communication while utilizing the basic routing function. For example, the base layer processing unit 2 executes a routing function in multicast communication or closed-area communication, or according to the situation of the external environment 23 or the like at that time.
1 controls transmission of the packet to a link other than the output link selected. Also, the knowledge layer processing unit 2
2 indicates the received control information (for example, failure of an adjacent node,
Based on the addition of a new node and the deletion of an existing node), information on routing managed by the own node 20 is appropriately updated and controlled.

The basic layer processing section 21 is composed of the following functional elements in detail. That is, the header extraction unit 3
0, reverse link evaluation unit 31 and output link determination unit 32
Consists of

The header extracting unit 30 extracts information of a header part in an input packet x according to a format as illustrated in FIG. 3, and gives each information in the header part to each unit corresponding to the information. The contents of the data part of the packet are also given to each part determined by the packet type.

In FIG. 3, the packet of this embodiment includes, for example, a packet type indicating whether the packet is an information packet or a control packet, a source address defining a source terminal, a destination address defining a destination terminal, and a It is composed of a header section including a cumulative cost value up to the node (cumulative cost value until reaching the own node 20) and the like, and a data section in which the transfer data body is inserted. here,
The cost refers to, for example, a sum of a transmission cost and a switching cost according to a packet transmission delay time, a distance between nodes, and the like.

Here, the header extracting unit 30 stores a set φ = {φ0, φ1,..., ΦN−1} of values serving as addresses of an internal table corresponding to a source address and a destination address. When a source address i and a destination address d are given, address values φi and φd are output. Here, N represents the number of nodes in the network. In addition, as the address values φi and φd,
The source address i and the destination address d may be used as they are.

The reverse link evaluation section 31 comprises an evaluation algorithm processing section (A) 31a and an address-based evaluation value table (C) 31b.

The address-based evaluation value table (C) 31b stores, in association with each address value φ0, φ1,..., ΦN−1, the accumulated cost value {C0, C1 in the previous packet from the current transmission source terminal. ,..., CN-1} are stored as judgment reference evaluation values (see FIG. 6 described later).

The evaluation algorithm processing unit (A) 31a
By comparing the accumulated cost value c inserted in the input packet x with the criterion evaluation value Ci stored in the address-based evaluation value table 31b corresponding to the source address i (ψi) of the packet x, An output link selection probability vector Pi corresponding to a source address i (ψi) described later.
To form an enhancement signal β that defines the updating method of Note that the enhancement signal β takes 0 or 1 in this embodiment.

The output link determination unit 32 includes a probability vector set management unit 32a and a link output processing unit 32b.

The probability vector set management unit 32a has a number of output link selection probability vectors P0, P1,..., PN-1 equal to the number of nodes corresponding to each node (each address $ i).
(See FIG. 6 described later). Each output link selection probability vector Pn (n is 0 to N-1) is the number of links l + 1 connected to the node 20.
, Pn1,..., Pnl.

When the enhancement signal β is given, the probability vector set manager 32a updates the output link selection probability vector Pi associated with the source address #i in the current input packet x as follows.

When the enhancement signal β is 1, the probability vector set management unit 32a outputs the output link selection probability vector P
In i, the value of the element PiL related to the link L (L is 0 to 1) to which the current input packet x is input is increased, and the other elements Pij (j is any of 0 to L-1, L + 1 to 1) Update to lower the value of. On the other hand, when the enhancement signal β is 0, the probability vector set management unit 32a lowers the value of the element PiL related to the link L to which the current input packet x has been input in the output link selection probability vector Pi,
Update to increase other elements Pij.

As long as the above-described updating can be performed, any updating method may be used. Some examples are as follows.

When the reinforcement signal β is 1, the output link selection probabilities PiL and Pij are updated according to the equation (3) using the reinforcement learning function Fij shown in the equation (1), and the reinforcement signal β is updated.
Is 0, the reinforcement learning function Gij shown in equation (2)
The output link selection probability Pi according to equation (4) using
Update L and Pij. The sum 総 in the expressions (3) and (4) is for j (j is 0 to L−1, L + 1 to 1). Also, r in the equations (2) and (3)
eward and penalty are constants in the range of 0 to 1, respectively. Further, t is a time parameter.

Fij = reward * Pij (t) (1) Gij = penalty * Pij (t) (2) When β = 1 PiL (t + 1) = PiL (t) + ΣFij Pij (t + 1) = Pij (t ) −Fij (3) When β = 0 PiL (t + 1) = PiL (t) −ΣGij Pij (t + 1) = Pij (t) + Gij (4) As another method example, the range is 0 to 1. Re is a constant of
Instead of using ward and penalty, (5)
Dynamically changing reward expressed by the equation (6)
And a method of updating by using the formulas (1) to (4) by using the formula and the penalty. Here, a and b are constants in the range of 0 to 1. Further, c is the accumulated cost value inserted in the input packet x, and Ci is the accumulated cost value inserted in the packet that previously arrived from the source.

Reward = a * | c-Ci | / | c + Ci | (5) penalty = b * | c-Ci | / | c + Ci | (6) The link output processing unit 32b performs the normal operation except for the advanced communication. In the communication (unicast communication), among the elements of the output link selection probability vector Pd relating to the destination address d (ベ ク ト ル d) of the input packet x, the element having the maximum probability value is found, and the link relating to the maximum probability element is transmitted to the packet. Is determined as the output link k.

When outputting a packet to each link, the link output processing unit 32b internally stores a set α of costs α0 to αl for each link which is added to the accumulated cost value of the input packet x. When the validity of the output link k is confirmed by the knowledge layer processing unit 22,
The link output processing unit 32b adds the accumulated cost value c in the input packet x to the additional cost α for the output link k.
k, and link the processed packet to link k
Output to

When the output processing for a plurality of links is started from the knowledge layer processing unit 22 (for example, by multicast communication or closed area communication), the link output processing unit 32b performs the above-described accumulation for each output link. Update the cost value.

The link output processing unit 32b executes output processing to the output link when the knowledge layer processing unit 22 instructs output to another link instead of the output link k determined by itself. .

Hereinafter, the routing process at the time of unicast communication in the base layer processing unit 21 will be described with reference to the flowcharts of FIGS. 4 and 5 and the processing conceptual diagram of FIG. The routing process at the time of unicast communication in the base layer processing unit 21 can be roughly divided into an output link selection probability vector update process and a packet output process. An example is
Although the processing is performed in the order of the update processing of the output link selection probability vector and the output processing of the packet, the processing order is not limited to this.

When the packet x is input, the source address i, the destination address d, the accumulated cost value c added to the input packet x, and the link number L to which the packet x is input are recognized (step S1). 100). Thereafter, an address set φ is searched using the source address i as a key, and an output link selection probability vector Pi corresponding to the address φi obtained by the search is specified from the probability vector set P (step 101).

Then, it is determined whether or not the corresponding address φi and the output link selection probability vector Pi exist (step 102).

If it does not exist, the address set {
, A new element ψi related to the current source address i is added, and a new element Pi related to the current source address i is also added to the probability vector set P, and the process proceeds to step 110 described below (step 110). 103). Here, as the output link selection probability vector Pi to be added, an initial link selection probability vector having the largest value of the probability of the link L to which the packet x has been input is applied.

On the other hand, when the address φi and the output link selection probability vector Pi exist, the previous accumulated cost value Ci is taken out from the evaluation value table C by address using the address φi as a key (step 104). The previous accumulated cost value Ci is compared with the accumulated cost value c inserted in the input packet x (step 10).
5). Then, if c = <Ci, an enhanced signal β having a value of 1 is output (step 106), while if c> Ci, an enhanced signal β having a value of 0 is output (step 107).

Here, in order to loosen the criterion,
Value C obtained by adding offset value y to previous accumulated cost value Ci
i + x may be compared with the accumulated cost value c inserted in the input packet x to form the enhancement signal β.

When the comparison process as described above is completed, the value Ci corresponding to the address φi in the address-based evaluation value table C is updated to the accumulated cost value c inserted in the input packet x (step 108).

Thereafter, according to the value of the enhancement signal β, the value of the output link selection probability vector Pi related to the current transmission source address i is updated by the above-described method (step 10).
9). As described above, the update is mainly performed on the output link selection probability PiL of the link of the link number L to which the packet x has been input. That is, the link between the transmission source (i) and the link L is changed by updating the corresponding part of the probability vector Pi for the link L. Such updating of the output link selection probability vector Pi is reflected in the determination of the output link of a future arriving packet having the current source address i as the destination address d.

FIG. 7 is an explanatory diagram showing a specific example of the update processing of the output link selection probability vector. In the case of this example, a packet having a source address i of 25, a destination address d of 10, and a cumulative cost value c of 30 is input from a link having a link number L of 1. If the evaluation value (previous accumulated cost value) Ci (C25) corresponding to the source address 25 is 40, the accumulated cost value c = 30 in the current input packet is smaller, so the value of the enhancement signal β is 1 Becomes As a result, the link number 1 in the output link selection probability vector Pi (P25) corresponding to the source address 25
Is increased (from 0.25 to 0.55), and the probability of elements of other link numbers is decreased (from 0.25 to 0.15). Thereafter, the evaluation value Ci (C25) corresponding to the source address 25 is updated to the accumulated cost value 30 in the current input packet.

When the output link selection probability vector Pi is updated as described above, the process proceeds to a packet output process. In the packet output process, first, an address set φ is searched using the destination address d as a key, and an output link selection probability vector Pd corresponding to the address φd obtained by the search is specified from the probability vector set P (step 11).
0). Note that if the packet is a target packet for closed communication, an address set for closed communication may be searched as the address set す る to be searched.

Then, it is determined whether or not the corresponding address φd and the output link selection probability vector Pd exist (step 111).

If not, the input packet is discarded (step 112). It should be noted that all links other than the input link may be determined as the output links when they do not exist, except in the case of closed communication.
In this case, for example, each node detects a stray packet that does not reach the destination node for a long time by comparing the time stamp inserted in the packet with the current time, and discards the packet at this stage. To

On the other hand, when the address φd and the output link selection probability vector Pd exist, the link relating to the element having the highest probability value in the output link selection probability vector Pd is determined as the output link, and the knowledge layer is determined. The processing unit 22 is notified (step 113).

At this time, the knowledge layer processing unit 22 confirms the validity of setting the determined link as the output link based on the failure information of the built-in link (step 114). If not, change to the link with the next probability value.

Thereafter, the link-specific cost value αk for the output link k, for which the validity determination processing has been performed by the knowledge layer processing unit 22, is calculated as the cumulative cost value c of the input packet x.
And the accumulated cost value c of the packet is added to the added value c +
αk (step 115), and the updated packet x
Is transmitted to the link k (step 116).

Next, a process (routing process) of determining an output link for each destination at the time of multicast communication performed by the knowledge layer processing unit 22 will be described with reference to the flowchart of FIG.

The process shown in FIG. 8 is performed when the own node 20 is a multicast transmission source and transmits to all other nodes, or when a packet given from another node is transmitted to a plurality of destinations for multicast communication. When an address is inserted, or when a packet does not have a specific destination address inserted, information for instructing multicast communication is inserted, and multiple destination addresses to be used for multicast communication are recognized based on stored information contained in the packet. It is executed when it is done. Further, when the packet to be provided to the multicast has come from another node, the above-described update processing of the output link selection probability vector Pi is executed before executing the processing illustrated in FIG.

First, the knowledge layer processing unit 22 outputs output link selection probability vectors Pd1 to Pd1 corresponding to all destination addresses d1 to dm (m is the number of destinations) to be provided for multicasting.
dm is extracted (step 200).

Thereafter, all the elements Pd11 to Pd1 corresponding to the respective output links of the output link selection probability vectors Pd1,..., Pdm for each destination address
,..., Pdm1 to Pdm1 (represented by the symbol pi) are determined as thresholds T for determining whether to adopt the output link.
H (TH is a fixed value of 0 or more), and according to the comparison result, the transmission flags fd11 to fd11,.
The value of fdml (represented by the symbol fi) is set (step 201). That is, if pi> TH, fi
= 1 and fi = 0 if pi = TH.

Next, for each of the output links 0,..., L, the transmission flags of all the destination addresses for which no packet transmission has been executed at that time are totaled, and the output link Lmax having the maximum total value is set to 1 (Step 202).

Then, all the destination addresses whose transmission flags to the determined output link Lmax are 1 are
The packet is inserted into the destination address field of the output packet and transmitted to the output link Lmax (step 203).

Thereafter, it is determined whether or not there is a destination address for multicast communication for which packet transmission has not been completed (step 204). If not, a series of processing ends. On the other hand, when the output link Lmax remains, the process returns to the above-described step 202 to search for an output link Lmax (different from the above-described one) to be used for output.

According to the above embodiment, the following effects can be obtained. (1) Since routing processing is autonomously performed at each node, there is no need to provide a common database for access between the nodes, and the network configuration can be simplified.

(2) Since routing processing is performed in each node without considering the network topology, the network topology to be processed is not limited, and there is a degree of freedom. That is, any network
This embodiment can be applied to topologies.

(3) There is no need to exchange information such as the status of line use between nodes for routing. That is, information exchange for routing is small, and information transfer is efficient. If the transmission source is an additional node, information necessary for routing is automatically set by recognizing a new transmission source address, so that it is unnecessary to notify that an additional node or the like has been added. .

(4) By registering in the knowledge layer processing unit an output link which is disabled due to a failure or congestion occurring in the network due to transmission / reception of a control packet or the like, a failure or congestion occurring in the network is registered. Can be easily dealt with. Even when such a control packet is not exchanged, if the link related to the failure or congestion is not used due to the failure or congestion, the value of the link element in the output link selection probability vector approaches 0, and Will not be used more and more.

(5) It is possible to easily cope with changes in multicast communication, closed area communication, changes in output links, and the like.

In the description of the above embodiment, various embodiments obtained by modifying the above embodiment have been described.
The following modified embodiments can be given.

In the above embodiment, the evaluation value Ci is
Cumulative cost value c inserted in current input packet x
However, the evaluation value Ci may be updated by weighted addition of the evaluation value Ci so far and the accumulated cost value c inserted in the current input packet x.

In the above embodiment, the value of each element of the output link selection probability vector takes an arbitrary value within the range of 0 to 1. However, in order to reduce the memory capacity, the following is used. Such a template vector may be used.

That is, the possible values of each element of the output link selection probability vector are limited, a plurality of patterned combinations are defined, template vectors are assigned, numbers are assigned to each of them, and the input link The probability value may be updated to a template vector that is increased or decreased by one step, and storage may be performed using the template vector number to reduce the memory capacity. Note that even when the template vector itself is stored, the possible values of each element value are limited (because the number of bits is small), so that the memory capacity can be reduced.

When the template vector is used,
Since the update of the output link selection probability vector is not an update using the learning reinforcement function, the selection accuracy when selecting the output link may be reduced.

In order to correct this, a node to be corrected is determined, the number of input packets for each link from the node is monitored, and an actual output link selection probability vector is calculated.
Compared with the currently used template vector, if the Manhattan distance or the Euclidean distance between the two is larger than a certain value, the currently used template vector may be changed. As a changing method, a template vector having a shorter Manhattan distance or a Euclidean distance is selected, or the method is converted to a method using actual observation values without using a template vector.

Further, in the above embodiment, the destination address in the output packet is the same as the input packet. However, the output packet after the address conversion such as the conversion from the special address to the general address is transmitted. May be. Such a conversion process is performed by the knowledge layer processing unit.

Further, the method of determining the output link at the time of the multicast communication is not limited to the above embodiment. For example, an output link is determined for each destination address in the same manner as in the unicast communication, and at the time of output to the output link, one or more destination addresses determined for the same output link are inserted into the packet. Is also good.

[0077]

As described above, according to the routing method, the network node, and the communication system of the present invention, since there is no database that is commonly accessed between the nodes, the network configuration can be simplified, and The topology is not limited, and a routing method, a network node, and a communication system that can efficiently cope with a failure or congestion occurring in the network can be realized with less information exchange for routing between each node and more efficiently. .

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a functional configuration of an embodiment.

FIG. 2 is a block diagram illustrating a hardware configuration of the embodiment.

FIG. 3 is an explanatory diagram illustrating an example of a packet format according to the embodiment;

FIG. 4 is a flowchart (1) showing a routing process at the time of unityast communication of the embodiment.

FIG. 5 is a flowchart (2) showing a routing process at the time of unityast communication of the embodiment.

FIG. 6 is an explanatory auxiliary diagram of a routing process at the time of unityast communication of the embodiment.

FIG. 7 is an explanatory diagram illustrating an example of updating an output link selection probability vector according to the embodiment;

FIG. 8 is a flowchart illustrating output link determination processing during multicast communication according to the embodiment.

[Explanation of symbols]

Reference numeral 20: node, 21: basic layer processing unit, 22: knowledge layer processing unit, 30: header extraction unit, 31: reverse link evaluation unit, 31a: evaluation algorithm processing unit, 31b: address-based evaluation value table, 32: output Link determination unit, 3
2a: probability vector set management unit; 32b: link output processing unit.

Claims (5)

[Claims] 1. A routing method in which each network node determines an output link from its own node in accordance with a destination address inserted in an input packet, wherein each network node outputs an output link for each address that can be a destination address. It manages the probability information used to determine the link selection and the evaluation value used to determine how to update the probability information, and when a packet having a single destination address is input from another network node Then, the above evaluation value corresponding to the source address of the input packet is taken out, compared with the accumulated cost value inserted in the packet, and a comparison result that the current route of the input packet is suitable is obtained. Of the probability information elements related to the source address, When the value of the element is increased, the values of the other elements are decreased, and a comparison result indicating that the current route of the input packet is not suitable is obtained, the input packet is input among the elements of the probability information related to the source address. Lowering the value of the element relating to the link, increasing the value of the other elements, updating the evaluation value corresponding to the source address according to the accumulated cost value inserted in the input packet, and The link in the probability information corresponding to the destination address having the largest value is determined as the link that outputs the input packet, and the accumulated cost value of the input packet is updated by the cost corresponding to the determined link,
A routing method, comprising transmitting an updated packet to a determined link. 2. When transmitting the same packet to a plurality of destinations, the link to output the packet is determined in the order of the link in which the value of the element of the probability information for each destination address is equal to or greater than a threshold value, and the determined link is determined. 2. The routing method according to claim 1, further comprising outputting a packet in which all destination addresses in which the value of the element of the probability information relating to the link is equal to or greater than a threshold value are inserted. 3. Probability information to be used in a network node that determines an output link from its own node according to a destination address inserted in an input packet in order to select each output link for each address that can be a destination address Information storage means for storing at least an evaluation value for determining a method of updating the probability information for each address that can be a destination address, and additional cost information corresponding to each output link; and When a packet is input, the evaluation value corresponding to the source address of the input packet is extracted and compared with the accumulated cost value inserted in the packet, and the current route of the input packet is suitable. Is obtained, among the elements of the probability information related to the source address, When the value of the element relating to the link to which the input packet is input is increased, the values of the other elements are decreased, and the comparison result indicating that the current route of the input packet is not suitable is obtained, the element of the probability information relating to the source address is obtained. Among the above, the value of the element related to the link to which the input packet was input is reduced, the value of the other element is increased, and the evaluation value corresponding to the source address is determined according to the accumulated cost value inserted in the input packet. Means for updating the probability information corresponding to the destination address of the input packet, the link relating to the element having the largest value in the probability information corresponding to the destination address of the input packet is determined as the link for outputting the input packet, and only the cost corresponding to the determined link is determined. Update the cumulative cost value of the input packet,
A packet output unit for transmitting the updated packet to the determined link. 4. When transmitting the same packet to a plurality of destinations, the link to output the packet is determined in the order of the link in which the value of the element of the probability information for each destination address is equal to or greater than the threshold, and 4. The network node according to claim 3, further comprising: a multicast processing unit that outputs a packet in which all destination addresses in which the value of the element of the probability information related to the link is equal to or larger than a threshold value are inserted. 5. 5. A communication system comprising a plurality of network nodes according to claim 3, wherein the network nodes are connected according to an arbitrary network topology.