JP2016096432A - Multicast route search device and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable the search of a low-cost multicast route in comparison with a simple application of BBMC algorithm.SOLUTION: Whenever a determination branch is created by a reference agent (0), a multicast route search device determines whether or not a derivation agent (k) is created from the agent (0), where k=1, 2,..., and creates the derivation agent to perform route search if a condition is satisfied. In addition, whenever a determination branch is created by the derivation agent k, the device determines whether or not a derivation agent is created in the same manner, and creates the derivation agent to perform route search if a condition is satisfied. When determination branches are obtained by all the agents eventually, the device selects a multicast route of which the multicast route cost is the minimum as a final tree from the determination branches.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a multicast route searching apparatus and method for generating a multicast tree for use in multicast communication.

  In the IP (Internet Protocol) network, there is an increasing demand for broadcast-type services represented by multicast video streaming. To implement this type of service, the total cost of all the links that make up the tree is minimized. A multicast tree needs to be created.

  As an algorithm for generating a multicast tree, for example, a BBMC (Branch-Based Multi-Cast) algorithm is known. The BBMC algorithm is a multicast that starts with only a multicast start node on a network composed of a plurality of nodes and a plurality of directional links (links having different link costs depending on the link direction). Define a tree, add the shortest path that reaches one terminal node of the set of multicast terminal nodes whose paths have not yet been determined from the multicast tree to the multicast tree as a new branch, and finally perform this process for all multicast tree branches (For example, see Non-Patent Document 1).

  FIG. 6 is a flowchart showing the processing procedure and processing contents of the BBMC algorithm. First, in step S100, as an initial condition of the algorithm, a multicast end node set E, a start node set V used in step S101 as a link start node, a BCL (Branch Candidate List) representing a set of branch candidates, A set R of branches constituting the final tree is stored in the storage unit. Of these, BCL and R are set to an empty set as initial values. Further, only the multicast start point node is stored in the start point node set V.

  The attribute information of each node of the network that creates the multicast tree is stored in the node information storage unit 202 as shown in the BBMC routing module shown in FIG. The “reached attribute” indicates whether the corresponding node has become a component of the multicast tree. At the beginning of the algorithm, only the start node is included in the multicast tree, so it is set to “yes” and all other nodes are set to “no”. Is set. “The shortest branch candidate to the node and its cost” are set to “shortest branch candidate” = φ and “shortest branch candidate cost” = ∞ as initial values at the start of the algorithm.

  In step S101, a link originating from each node in the starting point node set V is selected, and a branch candidate is created from the link. At that time, if the branch end point is the multicast end point stored in the multicast end point node set E, that is, if “yes” is determined in the immediately preceding step S103, the link becomes a branch candidate as it is. On the other hand, when the end point is not the multicast end point, that is, when step S103 = "no", a route connecting the link after the shortest branch candidate selected in the immediately preceding step S102 becomes a branch candidate.

  In the first routine of the algorithm, the link becomes a new branch candidate as it is. This new branch candidate creation process is performed for all links originating from the node stored in the start node set V. If the “reached attribute” of the link end node is “yes”, the node has already been created. Since it will be included in the multicast tree, it will not be a branch candidate.

  On the other hand, when the “reached attribute” is “no” and the branch cost stored in the “shortest branch candidate cost” of the end node attribute is larger than the branch cost of the new branch candidate, the new branch candidate is Registered in the branch candidate set BCL. When a new branch candidate is registered in the branch candidate set BCL, if the “shortest branch candidate” of the end node of the new branch candidate is an empty set, there is no branch candidate deleted from the branch candidate set BCL. However, when the “shortest branch candidate” exists, the branch candidate is deleted from the branch candidate set BCL. When the new branch candidate is registered, the values of “shortest branch candidate” and “shortest branch candidate cost” are replaced with the new branch candidate and its path cost value.

  If the branch cost stored in the “shortest branch candidate cost” is less than or equal to the branch cost of the new branch candidate, the new branch candidate is not stored in the branch candidate set BCL, and there is no branch candidate to be deleted from the BCL. . Further, the “shortest branch candidate” and “shortest branch candidate cost” values of the node do not change.

  In step S102, the shortest branch candidate in the branch candidate set BCL, that is, the branch candidate with the lowest branch cost in the BCL is extracted. At that time, “shortest branch candidate to the node” in the attribute of the end point node of the shortest branch candidate is set to an empty set. Further, the end node of the shortest branch candidate is replaced with the existing node set in the start node set V and stored in V.

  In step S103, it is determined whether the end node of the shortest branch candidate selected in step S102 exists as a multicast end node in the multicast end node set E. If it exists, the process proceeds to step S104. Return to.

  In step S104, the shortest branch candidate selected in step S102 is stored in the set R as a new branch, and the end node of the new branch is deleted from the end node set E.

  In step S105, it is determined whether or not the end node set E is an empty set. If the end node set E is an empty set, it is the same value that branches to all multicast end points have been obtained, so the algorithm ends. On the other hand, if it is not an empty set, the process proceeds to step S106.

  In step S106, the node set in the start point node set V in which the end node of the new branch is currently stored is replaced with all the node sets except the start point on the new branch stored in the branch set R in step S104. In addition, the “reached attribute” of all nodes included in the node set in the start node set V is set to “yes”.

H. Matsuura, “Improvement of Steiner Tree Algorithm: Branch-Based Multi-Cast”, IEICE Transaction on Information and Systems, vol. E96-D, no. 12, pp. 2743-2752, Dec. 2013

  However, the above BBMC algorithm sometimes creates a tree far from the minimum tree cost. An example is shown in FIG. In the figure, a network that is a target for creating a multicast tree with BBMC is indicated by s for the multicast tree start node, e1 to e3 for the end node, and n1 to n4 for the other nodes. There is a link between each node, the arrow on the link indicates the direction of the link, and the number next to the link indicates the link cost.

  FIG. 7A shows the result of the BBMC algorithm registering the link originating from the node s set as the start node set V in step S101 as a branch candidate in the branch candidate set BCL. In this case, since the branch candidate s-n2 has the lowest branch cost, s-n2 is extracted from the branch candidate set BCL in step S102 shown in FIG. 6, and the start node set V is replaced with n2. Since n2 does not belong to the end node E in step S103, the process moves to step S101.

  FIG. 7B shows a process of registering a branch candidate in which BBMC links a link originating from n2 to sn2 in the second step S101 in the branch candidate set BCL. Since s-n2-e2 is the first branch candidate up to e2, it is registered in the branch candidate set BCL. However, since s-n2-n1 has a branch cost “6” and is larger than the branch cost “5” of the shortest branch candidate s-n1 already stored in the node information storage unit 202 of n1, Not registered in the aggregate BCL. Since s-n2-e1 is the first branch candidate up to e1, it is registered in the branch candidate set BCL. As a result, since s-n2-e1 has the lowest branch cost among the set of branch candidates BCL, it is extracted from the set of branch candidates BCL, and e1 is a multicast end node in step S103. n2-e1 is stored in the branch set R. At this time, e1 is deleted from the end node set E, so E = {e2, e3}. However, since it is not yet an empty set, the process proceeds to step S106, and n2 and e1 are stored in the nodes in the start point node set V. As a result, V = {n2, e1}, and the process returns to step S101.

  FIG. 7 (3) shows a process in which the BBMC registers the link originating from the nodes n2 and e1 in the third step S101 as a new branch candidate in the branch candidate set BCL. Since “yes” is selected in step S103 of the previous routine, each link becomes a branch candidate independently. Since n2-e2 has a branch cost lower than that of the branch candidate s-n2-e2 registered in e2, s-n2-e2 is replaced and registered in the branch candidate set BCL. Since n2-n1 has a lower branch cost than the branch candidate s-n1 registered in n1, s-n1 is replaced and registered in the branch candidate set BCL. As a result, since n2-e2 has the smallest branch cost as n2-n1, it is deleted from the branch candidate set BCL, and is stored in the branch set R because e2 is an end point node.

  FIG. 7 (4) shows processing for registering a link from e2 in which BBMC is stored in the start node set V as a new branch candidate in the BCL. As a result, e2-n4 and e2-e3 are registered in the branch candidate set BCL, and since n2-n1 is the branch candidate having the smallest branch cost, it is deleted from the branch candidate set BCL.

  If the search process is continued in this way, the branches that are finally stored in the branch set R become three branches s-n2-e1, n2-e2, and e2-e3. The total tree cost is “12”, resulting in a costly tree.

  The present invention has been made paying attention to the above circumstances, and an object of the present invention is to provide a multicast route search apparatus and method capable of searching for a multicast route with a lower cost compared to a case where the BBMC algorithm is simply applied. It is to provide.

In order to achieve the above object, the present invention comprises the following various aspects.
(1) Used in a network in which a plurality of nodes can be connected to each other via a communication link, and searches for an optimum route from an arbitrary start node that performs multicast communication to a plurality of multicast end nodes among the plurality of nodes. A multicast route search device, wherein a reference agent searches a plurality of branch candidates on a route from the start point node to the plurality of multicast end point nodes and stores the plurality of branch candidates in a reference branch candidate storage unit, and the reference branch candidate storage unit A path search process in which a branch candidate whose end point reaches one of the plurality of multicast end point nodes and has a minimum branch cost is stored in the reference branch storage unit as a determined branch from the plurality of branch candidates stored in For all multiple multicast endpoint nodes A means to be executed until a decision branch is obtained and each time a decision branch is obtained by the reference agent, a first derivative agent derived from the reference agent is created and stored in the reference branch candidate storage unit at this time All the branch candidates that have been determined are copied to the first branch candidate storage unit of the first derivation agent, and the branches other than the determined branch among the branches stored in the reference branch storage unit are First derivative agent creating means for copying to the first branch storage unit of the derived agent, and the plurality of the branch agents stored in the first branch candidate storage unit by the first derivative agent as a reference. Search for branch candidates on the route to the multicast end node A branch candidate that is stored in the first branch candidate storage unit, and whose end point reaches one of the multicast end point nodes among the branch candidates stored in the first branch candidate storage unit and has the smallest branch cost The decision branch is obtained by the means for executing the route search processing for saving the decision branch in the first branch storage unit until the decision branch is obtained for all of the plurality of multicast end nodes, and the first derived agent. Each time, a second derivation agent that is further derived from the first derivation agent is created, and all the branch candidates stored in the first branch candidate storage unit at this time are stored in the second derivation agent. Are copied to the second branch candidate storage unit and stored in the first branch storage unit. The second derived agent creating means for copying the branch excluding the determined branch among the existing branches to the second branch storage unit of the second derived agent; and The branch candidates stored in the branch candidate storage unit are searched for branch candidates on the route to the plurality of multicast end-point nodes, stored in the second branch candidate storage unit, and the second branch candidates A path search process for storing a branch candidate whose end point reaches one of the plurality of multicast end point nodes from among the branch candidates stored in the storage unit and having the smallest branch cost as a determined branch in the second branch storage unit; , And execute until a decision branch is obtained for all of the multiple multicast end nodes. Means, and means for managing attribute information including agent costs of the first and second derived agents created by the first and second derived agent creating means using a derived agent list, and finally all When the decision branch is obtained for all of the plurality of multicast end nodes by the agent, the multicast path costs based on the decision branch respectively obtained by the reference agent and the first and second derived agents are compared, Means for selecting the multicast route having the lowest multicast route cost as the optimum route.

  (2) In (1), the first and second derived agent creation means are configured to determine the multicast route cost based on the decision branch obtained by the reference agent and the multicast route predicted to be obtained by the new derived agent. A first means for calculating a cost difference from the cost and determining whether or not the new derivative agent can be created based on the calculated cost difference is provided.

  (3) In (2), when it is determined that the first and second derived agent creation means can create the new derived agent, the derived agent list is based on the derived agent list. A second means for determining whether or not the number of already created derived agents has reached a preset upper limit, and creating the new derived agent when the upper limit is not reached, and the derived agent list When it is determined that the number of created derivative agents registered in the list has reached the upper limit, the maximum of the agent costs of the derived agents registered in the derived agent list and the new Compared with the estimated agent cost of the derived agent, and if the predicted value is smaller than the maximum value, Create a Do-derived agent, a third means for replacing the derived agents of the agent it costs up to a new derivation agents created above, further comprising.

(4) In (2), the reference agent and the first and second derived agents determine the quasi-shortest branch candidate according to the shortest branch candidate to the node and the branch cost for each node in the course of the route search process. The derivation agent stores the quasi-shortest branch candidate held by the ending node of the decision branch selected immediately before by the derivation source agent in the branch candidate storage unit as a new branch candidate. The first and second derived agent creating means sets the predicted value of the multicast route cost by the new derived agent to | dev_R (k) | and the determination obtained by the reference agent immediately before creating the new derived agent. Multicast route cost based on branch | R (0) | When the number of cast end points is ENDS, the coefficient is W (W ≧ 0), and the evaluation value based on the cost difference is EVA_VALUE 1, the predicted value | dev_R (k) | The branch cost of the quasi-shortest branch candidate held by the terminal node of the decision branch obtained by the derivation source agent immediately before the execution and the minimum branch in the branch candidate stored in the branch candidate storage unit of the derivation source agent The smaller of the costs is calculated by adding to the multicast path cost based on the decision branch excluding the decision branch obtained by the agent of the derivation immediately before creating the new derivation agent, and the evaluation The value EVA_VALUE 1 is
EVA_VALUE 1 = | dev_R (k) | − | R (0) | −W × ENDS
When the calculated evaluation value EVA_VALUE 1 is 0 or less, it is determined that the new derived agent can be created.

(5) In (3), when the third means of the first and second derived agent creation means obtains the estimated value of the agent cost of the new derived agent as AGENT_COST 1 and the new derived agent The expected multicast route cost value is | dev_R (k) |, the multicast route cost based on the decision branch obtained by the reference agent immediately before the creation of the new derived agent is | R (0) | The number of derivation stages of the agent from the above reference agent is DEVS, the coefficient is W1 (W1 ≧ 0), the number of unreachable multicast endpoints at the time when the derived agent is derived is ENDS, and the coefficient is W2 (W2 ≧ 0) The decision value obtained by the derivation source agent immediately before the creation of the derivation agent, using the predicted value | dev_R (k) | Immediately before creating the derivation agent, the smaller of the branch cost of the quasi-shortest branch candidate held by the terminal node and the minimum branch cost of the branch candidate stored in the branch candidate storage unit of the derivation source agent Is added to the multicast path cost based on the decision branch excluding the decision branch obtained by the agent of the derivation, and the agent cost estimate AGENT_COST 1 of the new derivation agent is
AGENT_COST 1 = | dev_R (k) |-| R (0) | + W1 × DEVS−W2 × ENDS
Calculated by

  (6) In any one of (1) to (3), each time a decision branch is obtained by the first or second derived agent, the decision branch obtained by the first or second derivative agent is used. A cost difference between the multicast path cost and the multicast path cost based on the decision branch obtained by the reference agent is calculated, and based on the calculated cost difference, the first or second derived agent is determined as the derived agent. Derived agent deletion means for deleting from the list is further provided.

(7) In (6), the derived agent deleting means sets the multicast route cost by the first or second derived agent to | R (k) |, and the first or second derived agent is reached by this time. The multicast route cost based on the decision branch obtained by the reference agent immediately before creation is | R (0) |, the number of unreachable multicast end points is ENDS, the coefficient is W (W ≧ 0), and the cost difference is When the evaluation value based on EVA_VALUE 2 is the above evaluation value EVA_VALUE 2,
EVA_VALUE 2 = | R (k) |-| R (0) | -W × ENDS
When the calculated evaluation value EVA_VALUE 2 is larger than 0, the first or second derived agent is deleted from the derived agent list.

(8) In any one of (1) to (3), an agent that updates an agent cost of a derived agent registered in the derived agent list every time a decision branch is obtained by the first or second derived agent Cost update means, wherein the agent cost update means determines the multicast path cost of the registered derivative agent at this time as | R (k) | and the determination obtained by the reference agent up to this time The multicast route cost based on the branch is | R (0) |, the number of derivation stages from the reference agent of the relevant agent is DEVS, the coefficient is W1 (W1 ≧ 0), and the unreachable multicast end point when the relevant agent is derived The agent code to be updated is assumed when the number of is ENDS and the coefficient is W2 (W2 ≧ 0). The door AGENT_COST 2
AGENT_COST 2 = | R (k) | − | R (0) | + W1 × DEVS−W2 × ENDS
Calculated by

(1) Each time a decision branch is obtained by the reference agent, a first derivation agent is created, and each time a decision branch is obtained by the first derivation agent, a second derivation agent is created. Then, by these derived agents, branch search processing is performed based on branch candidates and branches inherited from the derivation source agent. That is, each time a decision branch is obtained, a new derived agent is created step by step, and a branch search process is performed by each of the plurality of agents. When a decision branch is finally obtained for all of a plurality of multicast end nodes by all agents, the multicast route having the lowest cost among the multicast route costs based on the decision branch is selected as the optimum route. .
For this reason, it is possible to find the multicast route missed by the route search by the reference agent by the first and second derived agents, thereby reducing the cost of the multicast route compared with the case of using only the route search processing by the reference agent. It becomes possible to search with high probability.

  (2) When creating a new derived agent, considering the cost difference between the multicast route cost predicted to be obtained by the new derived agent and the multicast route cost by the reference agent, Whether or not creation is possible is determined. For this reason, a derived agent is created only when there is a high probability that a multicast route with a lower cost than the reference agent can be searched, so that a route with the minimum tree cost can be searched with a higher probability.

(3) When it is determined that a new derived agent can be created, a new derived agent is created when the number of created derived agents in the derived agent list has not reached the upper limit. On the other hand, when the number of created derivative agents has reached the upper limit, the maximum value of the agent costs of the derived agents registered in the derived agent list and the predicted value of the agent cost of the new derived agent are described above. When the predicted value is smaller than the maximum value, the new derived agent is created, and the entry of the derived agent list is replaced with the new derived agent from the derived agent with the largest agent cost.
For this reason, when a new derived agent is created, a derived agent having a smaller agent cost within the range of the upper limit value is created based on the derived agent list. For this reason, the probability that a multicast route with a lower cost can be searched increases. In addition, since the upper limit is set for the number of derived agents, the processing load required for the route search can be suppressed as compared with the case where no upper limit is set for the number of derived agents.

  (4) When determining whether or not to create a new derived agent, the multicast route cost | R (0) | by the reference agent and the predicted multicast route cost by the new derived agent | dev_R (k) | In addition to the cost difference with the number, the number of multicast termination points ENDS that the branch has not yet reached is also considered, and it is determined whether or not a new derivative agent can be created. In addition, when calculating the predicted value | dev_R (k) | of the multicast route cost by the new derived agent, the branch cost of the quasi-shortest branch candidate held by the corresponding node is also taken into consideration, so the predicted value | dev_R (k ) | Can be calculated more accurately. As a result, whether or not a new derived agent can be created can be determined with higher accuracy.

  (5) When determining whether a new second agent can be created based on the agent cost, the multicast route cost | R (0) | by the reference agent and the predicted value of the multicast route cost by the new derivative agent The determination is performed in consideration of not only the cost difference from | dev_R (k) | but also the number of derivation stages DEVS from the reference agent of the derivation agent and the number of unreachable multicast end points ENDS at the time of derivation. In addition, when calculating the predicted value | dev_R (k) |, the branch cost of the quasi-shortest branch candidate held by the corresponding node is also taken into consideration, so that a more accurate predicted value can be calculated. As a result, when determining whether or not a new derived agent can be created, it is possible to determine with higher accuracy.

  (6) Every time a decision branch is obtained by a derived agent, if the evaluation value based on the cost difference between the multicast route cost by the derived agent and the multicast route cost by the reference agent is greater than 0, the derived agent is deleted Is done. Therefore, an effective multicast route search can be performed using only derived agents that are highly likely to obtain a multicast route cost smaller than the multicast route cost obtained by the reference agent.

  (7) When calculating the evaluation value, the number of multicast end points ENDS that have not yet reached the branch is also considered. For this reason, agent deletion can be performed compared with the case where the agent deletion permission / inhibition is determined only by the cost difference between the multicast route cost | R (k) | by the derived agent and the multicast route cost | R (0) | It becomes possible to determine the propriety more accurately.

  (8) Every time a decision branch is obtained by a derived agent, the agent cost of the derived agent registered in the derived agent list is updated. Therefore, the agent cost of the derived agent registered in the derived agent list can always be kept up-to-date. In addition, when calculating the agent cost of the derived agent, the multicast route cost | R (k) | by the derived agent and the multicast route cost obtained by the reference agent immediately before the creation of the derived agent | R (0) In addition to the cost difference with |, the number of derivation stages DEVS from the reference agent of the derivation agent and the number of unreachable multicast end points ENDS at the time of derivation are considered. For this reason, the agent cost of the derived agent can be calculated more accurately.

  That is, according to the present invention, it is possible to provide a multicast route searching apparatus and method that can search a multicast route with a lower cost as compared with the case where the BBMC algorithm is simply applied.

The block diagram which shows the function structure of the multicast route search apparatus which concerns on one Embodiment of this invention. The figure which shows the routing module structure of the multicast route search apparatus shown in FIG. The flowchart which shows the procedure and content of the multicast route search operation | movement performed by the multicast route search apparatus shown in FIG. The figure which shows an example of the multicast route search operation | movement shown in FIG. The figure which shows the routing module structure of the conventional multicast route search apparatus which uses a BBMC algorithm. The flowchart which shows the procedure and content of the multicast route search operation | movement performed by the conventional multicast route search apparatus. The figure which shows an example of the multicast route search operation | movement shown in FIG.

Embodiments according to the present invention will be described below with reference to the drawings.
[One Embodiment]
(Constitution)
FIG. 1 is a block diagram showing a functional configuration of a multicast route search apparatus according to an embodiment of the present invention, and RSV indicates a routing server as a multicast route search apparatus. FIG. 2 shows a module configuration of the routing server RSV.

  The routing server RSV executes a route search process using the Multi agent BBMC algorithm, and includes a control unit 1, a database 2, agent memories 3a0 to 3an, a shared memory 4, and a communication interface unit 5. . The communication interface unit 5 communicates with the network NW under the control of the control unit 1. The network NW is composed of an IP network and includes a plurality of nodes connected to each other by communication links.

  The database 2 includes a link information storage unit 21 and a node information storage unit 22. The link information storage unit 21 stores the link cost in association with information representing each communication link connecting the plurality of nodes. In the node information storage unit 22, for each of the plurality of nodes, information representing a communication link connected to the node, the shortest branch candidate and its cost from the start node to the node, and the quasi-shortest branch candidate and its cost , “Reached attribute” (yes / no) is stored. The “reached attribute” indicates whether the node has become a component of the multicast tree. In the initial state before the route search is started, only the start node is included in the multicast tree, so it is set to “yes”. All other nodes are set to “no”.

  The database 2 is not provided in the routing server RSV, but is provided in a database server or a cloud system provided independently for the routing server RSV, and these are required to be accessed from the routing server RSV via the network. Such information may be downloaded.

  In the agent memories 3a0 to 3an, a branch candidate storage unit 31 and a branch storage unit 32 are provided. Each of the branch candidate storage unit 31 and the branch storage unit 32 is divided into areas for each agent. Each area of the branch candidate storage unit 31 stores a branch candidate generated by the route search process for each agent. Each area of the branch storage unit 32 stores a branch determined by the route search process for each agent.

  The shared memory 4 is shared by each agent and includes an agent list storage unit 41. The agent list storage unit 41 stores pointers to a plurality of agents sorted by agent cost. An upper limit is set for the number of agent pointers stored in the agent list storage unit 41.

  The control unit 1 includes a central processing unit (CPU) as a main control unit. As a control function necessary for executing the Multi agent BBMC algorithm according to this embodiment, a network topology management processing unit 11, It has a branch candidate processing unit 12, a minimum tree creation processing unit 13, an agent processing unit 14, and a final tree determination unit 15. All of these control functions are realized by causing the CPU to execute a program stored in a program memory (not shown).

  The network topology management processing unit 11 manages a plurality of nodes in the network NW and communication links connecting these nodes. The link information storage unit 21 manages the link information for each link, and the node Each node information is managed by the node information storage unit 22. The link information manages the cost for each direction (AZ / ZA) for all the links connecting the nodes. The node information manages, for each of the plurality of nodes, the shortest branch candidate and its cost, the semi-shortest branch candidate and its cost, and “reached attribute” (yes / no) to the node.

  The minimum tree creation processing unit 13 uses the BBMC algorithm to search for the shortest branch candidate between nodes using the link information and node information managed by the network topology management processing unit 11 for each agent. Further, based on the branch candidates stored in the branch candidate storage unit 31, the branch candidate having the lowest branch cost reaching the multicast end node is selected, and the branch candidate is stored in the branch storage unit 32 as a determined branch. Process.

  The branch candidate processing unit 12 stores the branch candidate searched by the minimum tree creation processing unit 13 in the branch candidate storage unit 31 in accordance with the instruction from the minimum tree creation processing unit 13, and from the branch candidate storage unit 31. Is read out and output to the minimum tree creation processing unit 13.

  The agent processing unit 14 shares and manages each agent using the agent list storage unit 41, and each time the minimum tree creation processing unit 13 of each agent creates a new branch, the agent processing unit 14 stores the agent in the agent list storage unit 41. Decide whether to keep it in or delete the agent. Further, it is determined whether or not a new agent is derived from the agent.

  The final tree determination unit 15 accesses the agent list storage unit 41 to acquire the final tree cost of each agent, and performs processing for selecting a tree with the lowest tree cost.

(Operation)
Next, a multicast route search operation using the Multi agent BBMC algorithm executed by the routing server RSV configured as described above will be described. FIG. 3 is a flowchart showing the processing procedure and processing contents.

  In step S0, various data used when the agent (0) as the reference agent executes the BBMC algorithm is initialized. That is, since the multicast end point node that has not yet reached the reference agent (0) is stored in the multicast end point set E (0), all multicast end point nodes are stored as initial values. V (0) is a set of start nodes handled by the reference agent (0) in step S3, and a multicast start node is stored as an initial value. The branch candidate storage unit 31 of the reference agent is represented by a branch candidate list BCL (0), and the branch candidates up to each node handled by the reference agent (0) are stored in a state sorted by the branch cost. In the reference agent (0), branch candidates are not yet obtained, so the empty set is set. The branch set R (0) stored in the branch storage unit 32 stores the branch determined by the reference agent (0), but is initially set as an empty set.

  Further, the “reached attribute” of each node is stored in the node information storage unit 22 as shown in FIG. 2, but since no branch has yet arrived at a node other than the multicast start node, only the multicast start node “ “yes” is set, and “no” is set for other nodes. Further, the “shortest branch candidate” and “quasi-shortest branch candidate” of each node are set to empty sets, and their branch costs are set to “∞”. Initialization of the data of derived agents other than the reference agent (0) is not performed in step S0.

  When the initial data setting is completed as described above, k = 0 is set in step S1. Where k is the agent number. Accordingly, here, the reference agent (0) is selected as an operation target. Next, in step S2, it is determined whether or not the number of end points | E (0) | in the multicast end point set E (0) is greater than “0”. Here, when | E (0) | = 0, it means that the branch to all multicast end nodes has been determined by the reference agent (0), so multicast route search using the Multi agent BBMC algorithm The process ends.

  On the other hand, when the end point node remains in the multicast end point set E (0), the process proceeds to step S3. In the following description, agent (k) indicates a reference agent when k = 0, and indicates a derived agent when k ≧ 1. In step S3, a branch candidate is created with a link originating from a node registered in the multicast start point set V (k) of the derived agent (k) as the last hop. Further, a process for updating the branch candidate list BCL (k) is performed as necessary. This process corresponds to the process of step S101 using the BBMC algorithm shown in FIG. That is, when k = 0, the same processing as that described in step S101 in FIG. 6 is performed in the memory allocated to the reference agent (0) by the reference agent (0). When the value of k is 1 or more, the same processing is performed by the derived agent (k) in the memory allocated to the derived agent (k).

  In step S3, a branch candidate having a link originating from a node registered in the multicast start point set V (k) of the derived agent (k) as a final hop is created, and the branch candidate list BCL (k) is updated. Is done. This process corresponds to the process of step S101 in the BBMC algorithm of FIG. That is, when k = 0, the reference agent (0) executes the branch candidate search process described in step S101 shown in FIG. 6 using the memory area 3a0 assigned to the agent (0). When k ≧ 1, the derived agent (k) executes the same processing as the branch candidate search processing described in step S101 of FIG. 6 using the memory areas 3a1 to 3an allocated to the agent (k).

  For the end node of the target link, the shortest branch candidate and the quasi-shortest branch candidate terminating at the node are held for each agent. The shortest branch candidate is replaced with a new branch candidate if the new branch candidate is smaller than the branch cost of the existing shortest branch candidate. At the same time, the branch candidates in the branch candidate list BCL (k) are also replaced with new branch candidates. At this time, the quasi-shortest branch candidate is not registered in the branch candidate list BCL (k) and is held only at the multicast end node. This semi-shortest branch candidate is replaced with a new branch candidate when the branch cost of the new branch candidate is equal to or higher than the branch cost of the existing branch candidate and smaller than the branch cost of the existing semi-shortest branch candidate.

  In step S4, the agent (k) selects the shortest branch candidate in the branch candidate list BCL (k) and replaces the node in the starting point set V (k) with the end point of the selected shortest branch candidate. This process corresponds to executing the process using the BBMC algorithm in step S102 shown in FIG. 6 in the memory environment assigned to the agent (k).

  In step S5, it is determined whether the end point of the shortest branch candidate selected in step S4 is a multicast end node. If the result of this determination is that it is a multicast end node, processing proceeds to step S6. On the other hand, if it is not a multicast end node, the process returns to step S3. In other words, until YES is determined in step S5, the process of searching for a branch reaching the multicast end node is repeatedly executed by the same agent (k).

  In step S6, the branch candidate selected immediately before in step S4 is stored in R (k) of the branch storage unit 32 as a new determined branch up to the multicast end node, and the multicast end node is set as the multicast end point set E. Processing to delete from (k) is performed.

  In step S7, after the start point node set V (k) is emptied, a process of storing all the nodes on the branch newly created in step S6 in the start point node set V (k) is performed. Since the multicast tree has already arrived at these nodes, the “reached attribute” of the nodes stored in the start node set V (k) is set to “yes”.

In step S8, the agent processing unit 14 determines whether or not to delete the agent (k) as a deletion candidate from the agent list. However, when k = 0, the agent is the reference agent (0) and is excluded from the deletion target. When k ≧ 1, since it is a derived agent, the tree cost of the derived agent (k) | R (k) | (the sum of the branch costs of all branches of R (k)) and the reference agent (0 ) Tree cost | R (0) |, ENDS (the number of multicast endpoints in E (k) at this point), and its coefficient W (W ≧ 0)
EVA_VALUE 2 = | R (k) |-| R (0) | -W × ENDS
Is calculated. When EVA_VALUE 2 ≦ 0, the deletion candidate agent (k) is not deleted.

  On the other hand, when EVA_VALUE 2> 0, the derived agent (k) is deleted. The reason is that if the value of | R (k) | is sufficiently larger than the reference value | R (0) |, even if the tree creation is further advanced with this derived agent (k), the BBMC This is because it is unlikely that a tree having a smaller cost than | R (0) |

  When the derived agent (k) is deleted, an agent derived from the derived agent (k) is not created, and the process proceeds to step S11. The reason is that it is unlikely that even if a new agent is created by taking over the data of an agent that is so inefficient as to be deleted, a new agent can create a low-cost tree.

  If the agent (k) is not deleted, the process proceeds to step S9. In step S9, the agent processing unit 14 first determines whether a new agent derived from the agent (k) can be created. However, before that, the process of re-registering the attribute information regarding the agent (k) in the agent list storage unit 41, that is, the update process of the agent list storage unit 41 is performed. The reason for this is that the agent cost of agent (k) changes each time a new branch is made, so the arrangement position of agent (k) in agent list storage unit 41 (arranged in ascending order of agent cost) changes. Because there is a possibility of doing.

In this embodiment, the multicast route cost at this point of the agent (k) registered agent is obtained by | R (k) |, and the reference agent (0) obtains the agent agent immediately before the creation of the agent. Multicast route cost based on the determined branch is | R (0) |, the number of derivation stages from the reference agent of agent (k) is DEVS, the coefficient is W1 (W1 ≧ 0), and agent (k) is derived Agent cost AGENT_COST 2 of the above agent (k), where ENDS is the number of unreachable multicast endpoints and W2 (W2 ≧ 0)
AGENT_COST 2 = | R (k) |-| R (0) | + W ₁ × DEVS−W ₂ × ENDS
It is required as follows. Then, based on the calculated agent cost AGENT_COST 2, the registered agent (k) is re-registered in the agent list storage unit 41.

Next, in step S9, whether or not a new agent derived from agent (k) can be created is determined as follows. That is, the evaluation value EVA_VALUE 1 for determination is
EVA_VALUE 1 = | dev_R (k) | − | R (0) | −W × ENDS
If the calculated predicted value is greater than “0”, no new agent is created. On the other hand, when the calculated predicted value is “0” or less, it is determined that there is a possibility of creating a new agent.

  Here, the predicted value | dev_R (k) | of the multicast route cost of the new agent is obtained as follows. That is, since the branch stored last in step S6 is removed from R (k), the number of branches in R (k) is one less than the number of branches in R (0). Therefore, a smaller value is selected from the branch cost of the “quasi-shortest branch” of the branch end node stored in the immediately preceding step S6 and the minimum branch cost in the branch candidate list BCL (k). A value obtained by adding the calculated value to the tree cost obtained by removing the branch stored in step S6 from | R (k) | is determined as the above-mentioned | dev_R (k) |.

  Even when the agent cost of the new agent is “0” or less, that is, when it is determined that there is a possibility of creating a new agent, the new agent is not created unless the following conditions are satisfied. That is, it is determined whether or not the number of derived agents registered in the agent list storage unit 41 has reached a preset upper limit number of derived agents, and the number of derived agents corresponding to the upper limit number has already been determined in the agent list. If it is not yet registered in the storage unit 41, it is determined that a new agent can be created, and the process proceeds to step S10.

  On the other hand, when the number of derived agents registered in the agent list storage unit 41 has reached the upper limit, the maximum value among the agent costs of the derived agents registered in the agent list storage unit 41, The estimated agent cost AGENT_COST 1 of the derived agent to be newly created is compared. If the predicted value AGENT_COST 1 is smaller than the maximum value, the new derivative agent is created, and a pointer to the derivative agent with the largest agent cost registered in the agent list storage unit 41 is created. Replace with a pointer to the derived agent.

Note that the estimated value AGENT_COST 1 in the agent list of the new derived agent to be created is the value of the multicast route cost predicted to be obtained by the new derived agent as | dev_R (k) | The multicast path cost based on the decision branch obtained by the reference agent immediately before the creation of a new derived agent is | R (0) |, the number of derivation stages from the reference agent of the derived agent is DEVS, and the coefficient is W1 (W1 ≧ 0), when the number of unreachable multicast end points at the time when the derived agent is derived is ENDS and its coefficient is W2 (W2 ≧ 0),
AGENT_COST 1 = | dev_R (k) | − | R (0) | + W1 × DEVS−W2 × ENDS
Is calculated by

  Where | dev_R (k) | is the branch cost of the quasi-shortest branch candidate held by the terminal node of the decision branch obtained by the derivation source agent immediately before creation of the derivation agent, and the branch candidate of the derivation source agent Multicast based on the decision branch excluding the decision branch obtained by the derivation agent immediately before creating the derivation agent, with the smaller of the minimum branch costs in the branch candidates stored in the list BCL (k) Calculated by adding to the route cost.

  If it is determined in step S9 that a new agent should be created, a new agent derived from agent (k) is created in step S10. At this time, the branch created in the immediately preceding step S6 is deleted from R (k), and the remaining branch is copied to the branch storage unit dev_R (k) of the new agent derived from the agent (k). Further, “reached attribute”, “shortest branch candidate and its branch cost”, and “quasi-shortest branch candidate and its branch cost” in each node information storage unit 22 are copied from the agent (k) to the memory for the new agent. At that time, the “reached attribute” of the node other than the start point on the branch created in step S6 immediately before is reset to “no”, and the “shortest branch candidate and its branch cost” of these nodes is set to “quasi-shortest branch candidate”. And the value of its branch cost. Thereafter, the “quasi-shortest branch candidate” is set to an empty set, and its branch cost is set to “∞”. Finally, in each node, “shortest branch candidates” whose “reached attribute” is “no” are sorted according to the branch cost and re-registered in the branch candidate list BCL.

  Through the series of processes for transferring data from the agent (k) to the new agent, not only all branches other than the branch created in the previous step S6 are transferred to the new agent, but also the branch candidates of the agent (k) Along with “quasi-shortest branch candidate”, it becomes possible to transfer to the branch candidate list BCL of the new agent.

  The new agent is placed at the end of all existing agents. That is, when agent (m) is the last agent, it is set as agent (m + 1). That is, if it is determined in step S11 that the agent is the last agent, the number of branches in R (k) of all agents is made the same.

  In step S11, the agent (k) determines whether it is the last agent. In the case of the last agent, the process returns to step S1 to request the reference agent (0) for processing. On the other hand, if the agent (k) is not the last agent, k = k + 1 is set in step S12, and processing is requested to the derived agent (k + 1). The processing of this algorithm is repeatedly executed until there is no multicast endpoint of the reference agent (0) in step S2. Since each agent always creates one branch at a time, the absence of the multicast endpoint at the reference agent (0) in step S2 means that the other agents have no multicast endpoint. That is, all agents have finished generating the multicast tree.

  If it is determined in step S2 that the generation of the multicast tree by all the agents including the reference agent and the plurality of derived agents is completed, the process proceeds to step S13. In step S13, the minimum tree is selected from among the multicast trees created by all agents, and the selected multicast tree of the minimum tree is returned as the output of the Multi agent BBMC algorithm.

(Specific example of operation)
Next, a specific example of the operation described above will be described with reference to FIG. In FIG. 4, detailed description of the same processing content as in FIG. 7 (for example, FIG. 4 (1)) is omitted.
Here, the upper limit number of derived agents to be added is set to “1” with respect to the reference agent (0) that performs tree search using the BBMC algorithm. Further, the coefficient W used in EVA_VALUE 1 and EVA_VALUE 2 and the coefficients W ₁ and W ₂ used in AGENT_COST 1 and AGENT_COST ₂ are all set to “2”.

  As shown in FIG. 4 (2), when the first decision branch (s-n2-e1) is created in the reference agent (0), a new derived agent is generated from the reference agent (0) in step S9 in FIG. Whether to create (1) is determined as follows.

That is, first, EVA_VALUE 1 of the derived agent (1)
EVA_VALUE 1 = | dev_R (0) |-| R (0) |-| E (0) |
= 5−3−2 × 2
= -2
It is calculated as follows.

  Here, | dev_R (0) | is the shortest branch candidate s-n3 after the branch candidate s-n2-e1 (3) stored in the branch candidate list BCL (0) of the reference agent (0) is removed. Branch cost = 5. At this time, if the multicast end node e1 holds the quasi-shortest branch, the branch cost is the smaller of the branch cost “5” of the sn3. However, since the multicast end node e1 does not hold the quasi-shortest branch at this time, the branch cost “5” of the s-n3 is selected. | R (0) | is the total branch cost of the decision branch in R (0). At this time, s-n2-e1 (3) is the only branch. | E (0) | is the number of multicast end points remaining in E (0) at this time, and is two, e2 and e3. Since the value of EVA_VALUE 1 is the result of the above calculation −2 and is “0” or less, the derived agent (1) is determined to be valid and is not deleted.

Next, the agent cost of the derived agent (1) to be newly created is calculated. This agent cost is
AGENT_COST 1 = | dev_R (0) |-| R (0) | + 2 × DEVS-2 × ENDS
= 5−3 + 2 × 1-2 × 2
= 0
It becomes.

  Here, since DEVS is the number of derivation steps from the reference agent (0) of the newly derived agent (1) to be created, it is “1” here. ENDS is the number of unreachable multicast end points when the derived agent (1) was last derived, that is, the number of nodes in E (0), and is “2”. As described in the operation condition at the beginning, the upper limit number of derived agents other than the reference agent (0) is set to “1”, and since no derived agent exists in the agent list storage unit 41 at this stage, the derived agent is not described above. (1) is created, and its agent cost = 0 is registered in the agent list storage unit 41.

  In addition, with the creation of the derived agent (1), branches except for the decision branch sn2-e1 are taken over from R (0) to R (1). At this stage, R (0) has s- Since only n2-e1 is registered, R (1) becomes empty as shown in FIG. 4 (1 '). If the multicast end node e1 holds a quasi-shortest branch candidate, the quasi-shortest branch candidate becomes the shortest branch candidate up to the multicast end node e1, and this is the branch candidate list BCL ( Candidate for registration in 1). However, at this stage, since the quasi-shortest branch candidate is not yet registered in the multicast end node e1, the branch candidate list BCL (0) of the reference agent (0) is included in the branch candidate list BCL (1) of the derived agent (1). The same branch candidate is copied.

  As a result, in FIG. 4 (1 ′), the branch candidate s-n3 having the lowest branch cost is deleted from the branch candidate list BCL (1) of the derived agent (1), and the node n3 is not a multicast end node. Branch candidates up to one hop ahead via the node n3 are created in FIG. 4 (2 '). The branch costs of these branch candidates s-n3-e1 and s-n3-e3 are both “6” and are the first branch candidates arrived by the derived agent (1) up to the multicast end nodes e1 and e3. Therefore, it is registered in the branch candidate list BCL (1).

  As a result of FIG. 4 (2 ′), the branch candidate s-n1 has the lowest branch cost in the branch candidate list BCL (1) of the derived agent (1). Branch candidates up to one hop ahead are considered. At this time, the “shortest branch cost” attribute of the node n2 is inherited by the s-n2 blank cost = 2 previously held by the reference agent (0), and sn1-n2 is the branch cost. It is not registered because it is larger. Since s-n1-n4 is the first branch candidate up to the node n4, it is registered in the branch candidate list BCL (1) as the shortest branch candidate. As a result, three branch candidates with a cost of “6” are registered in BCL (1). In this case, one branch candidate is arbitrarily selected. In the example of FIG. 6 (3 ′), sn3-e1 is selected, and the node e1 is stored in R (1) because it is a multicast end node.

At this point in time when the first branch is determined by the derived agent (1), whether to continue to use or delete the derived agent (1) is determined in step S8 as follows. That is, EVA_VALUE 2 of derived agent (1) is
EVA_VALUE 2 = | R (1) | − | R (0) | −2 | E (1) |
= 6−3−2 × 2
= -1
Therefore, EVA_VALUE 2 becomes “0” or less. For this reason, the derived agent (1) is not deleted.

Next, in step S9, it is determined whether or not a new agent replacing the derived agent (1) is to be created. In this case, the agent cost of the derived agent (1) is first updated. The agent cost of the derived agent (1) is
AGENT_COST 2 = | R (1) |-| R (0) | + 2 × DEVS-2 × ENDS
= 6−3 + 2 × 1-2 × 2
= 1
It becomes. Therefore, the derived agent (1) is reset in the agent list storage unit 41 as agent cost = 1 again. At this time, if a plurality of agents are stored in the agent list storage unit 41, these agents are sorted in ascending order of their agent cost values.

If the agent further derived from the above derived agent (1) is agent (2), the agent cost of the new derived agent (2) is
AGENT_COST 1 = | dev_R (1) |-| R (0) | + 2 × DEVS-2 × ENDS
= 6−3 + 2 × 2-2 × 2
= 3
It becomes.

  Here, | dev_R (1) | is the shortest branch candidate held in the branch candidate list BCL (1) of the agent (1) and the branch cost of the quasi-shortest branch candidate up to the multicast end point mode e1. The branch cost of n3-e3 (6) is smaller. However, since there is no quasi-shortest branch candidate up to the multicast end node e1, at the present time, | dev_R (1) | is “6”. DEVS is “2” because the derived agent (1) has already been derived once from the reference agent (0). ENDS is “2” because there are currently two multicast end nodes e2 and e3 in E (1). As a result, the agent cost of the new derived agent (2) is higher than that of the source agent (1), and the upper limit of the number of agents in the agent list is “1”, so that the agent (2) is a new agent. Not created.

  FIG. 4 (4 ′) shows the generation operation of the second branch by the derived agent (1). At this stage, in step S3, the second branch n2-e2 is first created by the reference agent (0) in the same procedure as (3) in FIG. Since the starting node set V (1) stores the nodes n3 and e1 on the first branch sn3-e1 of the derived agent (1), the links originating from these two nodes n3 and e1 are branch candidates. It will be examined whether or not. Since there is a link from the node n3 to the end node e3 and the link cost is “1”, the shortest branch candidate to the multicast end node e3 is replaced with n3-e3.

  Further, since the multicast end node e3 is a multicast end node, the existing shortest branch candidate s-n3-e3 is held at the multicast end node e3 as a quasi-shortest branch candidate. Furthermore, since there is a link from the multicast end node e1 to the multicast end node e2, and the link cost is “1”, the shortest branch candidate to the multicast end node e2 is replaced with e1-e2. Since the node e2 is a multicast end node, the existing shortest branch candidate s-n2-e2 is held at the multicast end node e2 as a quasi-shortest branch candidate. As a result, since n3-e3 is a branch candidate in the branch candidate list BCL (1) with the lowest branch cost, n3-e3 is determined as the second branch and stored in R (1).

Subsequently, in step S8, it is determined whether to continue to use or delete the derived agent (1). In this case, EVA_VALUE 2 of derived agent (1) is
EVA_VALUE 2 = | R (1) | − | R (0) | −2 | E (1) |
= 7−7−2 × 1
= -2
Since EVA_VALUE 2 is equal to or less than “0”, the derived agent (1) is maintained without being deleted.

  Here, | R (0) | has the same result as (3) in FIG. 7 and is “7” because s−n2−e1 and n2−e2 are held in R (0). Also, since only E2 is held in the multicast end point set E (1) of the agent (1), | E (1) | = 1.

Next, in step S9, it is determined whether or not an agent (2) further derived from the derived agent (1) is to be created. However, EVA_VALUE 1 of agent (2) is
EVA_VALUE 1 = | dev_R (1) | − | R (0) | −2 | E (1) |
= 11−7−2 × 1
= 2
Since EVA_VALUE 1 is larger than “0”, it is not created.

  Here, | dev_R (1) | is the branch cost of the quasi-shortest branch to the multicast end node e3 and the shortest branch e1-e2 held in the branch candidate list BCL (1) of the agent (1). Since the smaller value is “5” and this value is added to s−n3−e1 (6), | dev_R (1) | becomes “11”.

  When the process shown in FIG. 4 (4 ′) is completed, the branch candidate e3-e2 exists as a link coming out from the end node e3, and the link cost is “1”. Is stored in R (1) as the final branch. As a result, the multicast tree cost finally generated by the Multi-agent BBMC algorithm is “8”. As described with reference to FIG. 7, this multicast tree cost is significantly smaller than the multicast tree cost “12” when a route search is performed by only one agent using the BBMC algorithm.

(effect)
As described above in detail, in this embodiment, every time a decision branch is created by the reference agent (0) using the BBMC algorithm, the agent (k) (k = 1, 2) derived from the reference agent (0). , ...) is created, and when a predetermined condition is satisfied, a derived agent is created in order, and these reference agents (0) and a plurality of derived agents (k) (k = 1) , 2,. Then, the costs of the multicast trees created by these multiple agents are compared, and the tree with the smallest cost is selected as the final multicast tree. For this reason, it is possible to search a tree missed in the route search by the reference agent (0) by a plurality of derived agents (k) (k = 1, 2,...) Derived from the reference agent (0). This makes it possible to search for a multicast tree with a lower cost than when only the search processing by the reference agent (0) is used.

  Further, when determining whether or not the derived agent (k) (k = 1, 2,...) Can be created, first, the cost difference between the estimated multicast route cost by the derived agent and the multicast route cost by the reference agent (0). If the EVA_VALUE 1 based on is equal to or less than “0” and there is no created agent in the agent list storage unit 41, the derived agent is created. Even when EVA_VALUE 1 by the derived agent is determined to be “0” or less, if the number of derived agents registered in the agent list storage unit 41 reaches the upper limit value, the agent included in the agent list storage unit When the agent cost of the derived agent is smaller than the maximum agent cost among the costs, the derived agent is registered in the agent list storage unit 41 by replacing it with an agent having the maximum agent cost. That is, a new derivative agent is created only when the probability of searching for a multicast tree having a lower cost than any of the agents registered in the agent list storage unit is high. For this reason, it is possible to further increase the probability that a low-cost multicast tree can be searched. In addition, since a new derived agent is created only when the number of registered registered agents does not exceed the upper limit, the processing load required for route search can be reduced compared to when no upper limit is set for the number of registered agents. Can do.

  Further, every time a decision branch is created by a derived agent, the cost difference between the multicast route cost based on the decision branch created by the derived agent and the multicast route cost based on the decision branch obtained by the reference agent is used as a reference. The calculated evaluation value EVA_VALUE 2 is calculated, and when the EVA_VALUE 2 is larger than “0”, the created agent is deleted. For this reason, it is possible to prevent a problem that the route search processing by an agent with a large multicast route cost among end agents that have already been created is executed indefinitely, thereby enabling effective route search using an agent with a low route cost. It can be carried out.

  Furthermore, when calculating the evaluation value EVA_VALUE 2, it is determined whether or not the created derivative agent can be deleted in consideration of the number of multicast end points ENDS that have not yet reached the branch. For this reason, it becomes possible to determine with higher accuracy whether or not the agent to be deleted can be deleted.

  Further, the agent cost of the derived agent registered in the agent list storage unit 41 is updated every time a new branch is generated with AGENT_COST 2 as the agent cost, and the registered derived agent with the highest agent cost is updated from the agent list storage unit. It is selected as a deletion target. For this reason, it is possible to delete in order from the agent with the highest agent cost that does not contribute to the route search with the minimum tree cost, thereby preferentially leaving the agent with a low agent cost and a high contribution to the route search with the minimum tree cost. Search processing can be performed.

[Other Embodiments]
The present invention is not limited to the above embodiment. For example, in the embodiment, the case where the routing server RSV is provided separately from the node group has been described as an example. However, one or more of the nodes may have the function of the routing server.

  In addition, the network configuration, the number of multicast end nodes, the configuration of the routing server RSV, the procedure and contents of the shortest tree creation processing, and the like can be variously modified without departing from the scope of the present invention.

  In short, the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Further, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, you may combine suitably the component covering different embodiment.

  RSV ... Routing server, NW ... Network, 1 ... Control unit, 2 ... Database, 3a0-3an ... Agent memory, 4 ... Shared memory, 5 ... Communication interface unit, 11 ... Network topology management processing unit, 12 ... Branch candidate processing unit 13 ... Minimum tree creation processing unit, 14 ... Agent processing unit, 15 ... Final tree determination unit, 21 ... Link information storage unit, 22 ... Node information storage unit, 31 ... Branch candidate storage unit, 32 ... Branch storage unit, 41 ... Agent list storage unit.

Claims (9)

Multicast route search that is used in a network in which a plurality of nodes can be connected to each other via a communication link, and searches for an optimum route from any start node that performs multicast communication to a plurality of multicast end nodes among the plurality of nodes. A device,
A reference agent searches for a plurality of branch candidates on a route from the start node to the plurality of multicast end nodes, stores the plurality of branch candidates in a reference branch candidate storage unit, and stores a plurality of branch candidates stored in the reference branch candidate storage unit A route search process in which a branch candidate whose end point reaches one of the plurality of multicast end point nodes and whose branch cost is the smallest is stored in the reference branch storage unit as a determined branch for all of the plurality of multicast end point nodes Means to execute until a decision branch is obtained;
Each time a decision branch is obtained by the reference agent, a first derived agent derived from the reference agent is created, and all branch candidates stored in the reference branch candidate storage unit at this time are stored in the first branch agent. A branch that is copied to the first branch candidate storage unit of the derivation agent and that excludes the determined branch from the branches stored in the reference branch storage unit is stored in the first branch storage unit of the first derivation agent. First derivative agent creation means for copying;
The first derivation agent searches the branch candidates on the route to the plurality of multicast end nodes based on the branch candidates stored in the first branch candidate storage unit, and stores the first branch candidate storage A branch candidate whose end point reaches one of the plurality of multicast end-point nodes and has the smallest branch cost among the branch candidates stored in the first branch candidate storage unit is determined as a first branch. Means for executing a route search process to be stored in the branch storage unit until a decision branch is obtained for all of the plurality of multicast end points;
Each time a decision branch is obtained by the first derivation agent, a second derivation agent that is further derived from the first derivation agent is created and stored in the first branch candidate storage unit at this point. All the branch candidates that are present are copied to the second branch candidate storage unit of the second derived agent, and the branches other than the determined branch among the branches stored in the first branch storage unit are Second derivative agent creating means for copying to the second branch storage unit of the derived agent of
The second derivation agent searches the branch candidates on the path to the plurality of multicast end nodes with reference to the branch candidates stored in the second branch candidate storage unit, and stores the second branch candidate storage A branch candidate whose end point reaches one of the plurality of multicast end-point nodes and has the smallest branch cost among the branch candidates stored in the second branch candidate storage unit is determined as a determined branch. Means for executing a route search process to be stored in the branch storage unit until a decision branch is obtained for all of the plurality of multicast end points;
Means for managing attribute information including agent costs of the first and second derived agents created by the first and second derived agent creating means using a derived agent list;
Each multicast path cost based on the decision branch respectively obtained by the reference agent and the first and second derived agents when all agents finally obtain decision branches for all of the plurality of multicast end nodes. And a means for selecting a multicast route having the lowest multicast route cost as the optimum route. The first and second derived agent creation means include:
A cost difference between a multicast route cost based on the decision branch obtained by the reference agent and a multicast route cost predicted to be obtained by a new derived agent is calculated, and the new cost is calculated based on the calculated cost difference. 2. The multicast route searching apparatus according to claim 1, further comprising a first means for determining whether or not a derivative agent can be created. The first and second derived agent creation means include:
If it is determined that the new derived agent can be created, whether or not the number of created derivative agents in the derived agent list reaches a preset upper limit value based on the derived agent list. A second means for determining and creating the new derived agent if the upper limit is not reached;
If it is determined that the number of derived agents already created in the derived agent list has reached the upper limit, the maximum value of the agent costs of the derived agents registered in the derived agent list, and the new A predicted value of the agent cost of the derived agent, and if the predicted value is smaller than the maximum value, the new derived agent is created, and the derived agent with the largest agent cost is created. The multicast route search apparatus according to claim 2, further comprising: a third means for replacing with a different derived agent. The reference agent, the first and second derived agents are:
Means for storing a quasi-shortest branch candidate according to the shortest branch candidate up to the node and its branch cost for each node in the course of the route search process;
Means for storing, in the branch candidate storage unit, the quasi-shortest branch candidate held by the end node of the decision branch selected immediately before by the derivation source agent as a new branch candidate,
The first and second derived agent creation means are obtained by the reference agent immediately before creating the new derived agent, with | dev_R (k) | Multicast path cost based on the determined branch is | R (0) |, the number of unreachable multicast end points is ENDS, its coefficient is W (W ≧ 0), and the evaluation value based on the cost difference is EVA_VALUE 1,
The predicted value | dev_R (k) | is used as the branch cost of the quasi-shortest branch candidate held by the terminal node of the decision branch obtained by the derivation agent immediately before creating the new derivation agent, and the derivation source The smaller one of the minimum branch costs of the branch candidates stored in the branch candidate storage unit of the agent is excluded from the decision branch obtained by the derivation source agent immediately before creating the new derivation agent Calculated by adding to the multicast route cost based on the determined branch,
The evaluation value EVA_VALUE 1 is
EVA_VALUE 1 = | dev_R (k) | − | R (0) | −W × ENDS
Calculated by
3. The multicast route searching apparatus according to claim 2, wherein when the calculated evaluation value EVA_VALUE 1 is 0 or less, it is determined that the new derived agent can be created. The third means of the first and second derived agent creating means is configured to use AGENT_COST 1 as a predicted value of the agent cost of the new derived agent, and a multicast path cost value predicted to be obtained by the new derived agent. | Dev_R (k) |, the multicast path cost based on the decision branch obtained by the reference agent immediately before creation of the new derived agent | R (0) |, the derived agent from the reference agent When the number of stages is DEVS, the coefficient is W1 (W1 ≧ 0), the number of unreachable multicast end points at the time when the derived agent is derived is ENDS, and the coefficient is W2 (W2 ≧ 0),
The predicted value | dev_R (k) | is used as the branch cost of the quasi-shortest branch candidate held by the terminal node of the decision branch obtained by the derivation source agent immediately before creation of the derivation agent, and the derivation source agent A multicast route based on a decision branch obtained by removing a smaller one of the minimum branch costs of the branch candidates stored in the branch candidate storage unit from a decision branch obtained by the derivation source agent immediately before creating the derivation agent. Calculated by adding to the cost,
Estimate AGENT_COST 1 of the agent cost of the new derived agent
AGENT_COST 1 = | dev_R (k) | − | R (0) | + W1 × DEVS−W2 × ENDS
The multicast route search apparatus according to claim 3, wherein the multicast path search device is calculated by:   Each time a decision branch is obtained by the first or second derived agent, the multicast path cost based on the decision branch obtained by the first or second derivative agent and the decision branch obtained by the reference agent A derivative agent deleting means for calculating a cost difference from the multicast path cost based on the calculated agent and deleting the first or second derived agent from the derived agent list based on the calculated cost difference; The multicast route search apparatus according to any one of claims 1 to 3, wherein The derived agent deleting means sets the multicast path cost by the first or second derived agent to | R (k) |, and the reference agent until immediately before creating the first or second derived agent up to this point Multicast path cost based on the determined branch obtained by the above is | R (0) |, the number of unreachable multicast end points is ENDS, its coefficient is W (W ≧ 0), and the evaluation value based on the cost difference is EVA_VALUE 2 and when,
The evaluation value EVA_VALUE 2 is
EVA_VALUE 2 = | R (k) | − | R (0) | −W × ENDS
Calculated by
7. The multicast route search apparatus according to claim 6, wherein when the calculated evaluation value EVA_VALUE 2 is greater than 0, the first or second derived agent is deleted from the derived agent list. Agent cost updating means for updating an agent cost of a derived agent registered in the derived agent list every time a decision branch is obtained by the first or second derived agent,
The agent cost updating means sets the multicast path cost of the registered derivative agent at this time to | R (k) |, and sets the multicast path cost based on the decision branch obtained by the reference agent up to this time to | R (0) |, the number of derivation stages of the derivation agent from the reference agent is DEVS, the coefficient is W1 (W1 ≧ 0), the number of unreachable multicast endpoints at the time the derivation agent is derived is ENDS, and the coefficient is W2 When (W2 ≧ 0), update agent cost AGENT_COST 2
AGENT_COST 2 = | R (k) | − | R (0) | + W1 × DEVS−W2 × ENDS
The multicast route search apparatus according to claim 1, wherein the multicast path search device is calculated by: A multicast route searching apparatus used in a network in which a plurality of nodes can be connected to each other via a communication link, and an optimum route from any start node to a plurality of multicast end nodes performing multicast communication among the plurality of nodes. A multicast route search method for searching for
A reference agent searches for a plurality of branch candidates on a route from the start node to the plurality of multicast end nodes, stores the plurality of branch candidates in a reference branch candidate storage unit, and stores a plurality of branch candidates stored in the reference branch candidate storage unit A route search process in which a branch candidate whose end point reaches one of the plurality of multicast end point nodes and whose branch cost is the smallest is stored in the reference branch storage unit as a determined branch for all of the plurality of multicast end point nodes A process that runs until a decision branch is obtained;
Each time a decision branch is obtained by the reference agent, a first derived agent derived from the reference agent is created, and all branch candidates stored in the reference branch candidate storage unit at this time are stored in the first branch agent. A branch that is copied to the first branch candidate storage unit of the derivation agent and that excludes the determined branch from the branches stored in the reference branch storage unit is stored in the first branch storage unit of the first derivation agent. Copying process,
The first derivation agent searches the branch candidates on the route to the plurality of multicast end nodes based on the branch candidates stored in the first branch candidate storage unit, and stores the first branch candidate storage A branch candidate whose end point reaches one of the plurality of multicast end-point nodes and has the smallest branch cost among the branch candidates stored in the first branch candidate storage unit is determined as a first branch. A path search process stored in the branch storage unit of the plurality of multicast end nodes until a decision branch is obtained for all of the plurality of multicast end points;
Each time a decision branch is obtained by the first derivation agent, a second derivation agent that is further derived from the first derivation agent is created and stored in the first branch candidate storage unit at this point. All the branch candidates that are present are copied to the second branch candidate storage unit of the second derived agent, and the branches other than the determined branch among the branches stored in the first branch storage unit are Copying to the second branch storage of the derived agent of
The second derivation agent searches the branch candidates on the path to the plurality of multicast end nodes with reference to the branch candidates stored in the second branch candidate storage unit, and stores the second branch candidate storage A branch candidate whose end point reaches one of the plurality of multicast end-point nodes and has the smallest branch cost among the branch candidates stored in the second branch candidate storage unit is determined as a determined branch. A path search process stored in the branch storage unit of the plurality of multicast end nodes until a decision branch is obtained for all of the plurality of multicast end points;
Managing attribute information including agent costs of the created first and second derived agents using a derived agent list;
Each multicast path cost based on the decision branch respectively obtained by the reference agent and the first and second derived agents when all agents finally obtain decision branches for all of the plurality of multicast endpoint nodes And a method of selecting a multicast route having the lowest multicast route cost as the optimum route.

Images