JPH07287692A - Method and system for distributing request in cooperative problem solution system - Google Patents

Abstract

PURPOSE:To provide a distributing method and a distributing system for a request in a cooperative problem solution system in which the variation of load at the time of execution can be suppressed and the occurrence of the competition of a resource can be reduced and processing time can be shortened. CONSTITUTION:A number of agents determining means 100 which determines the optimum number of agents to meet a scale based on the resource quantity and the request quantity of a problem by using information 130 related to the resource and the information 140 related to the individual request, a clustering means 110 which clusters the request to the request group of the number of the agents determined before by the number of agents determining means 100, and a request distributing means 120 which distributes the request group clustered by the clustering means 110 to the agents are provided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a request distribution method in a cooperative problem solving system and a request distributing system in a cooperative problem solving system, and more particularly, to route setting on a network, LSI wiring design, facility allocation, and other resource allocation. In a cooperative problem solving system that distributes requests to multiple agents and solves problems by coordinating between agents, a request distribution method in a cooperative problem solving system that distributes requests to agents and a request in a cooperative problem solving system A demand distribution system in a distribution system.

[0002]

2. Description of the Related Art In a collaborative problem solving system for solving a resource allocation problem using a plurality of agents, (1) request distribution to agents; (2) local solution calculation unique to each agent; (3) inter-agent negotiation Resolution of resource conflicts;

In such a collaborative problem solving system, the above-mentioned conventional method for distributing requests to agents in (1) is based on a preset number of agents. The number of requests for is evenly distributed.

The local solution calculation of (2) is to divide into a request and a subproblem, distribute them to agents, and obtain a local solution for each agent independently, and when there is competition with other agents in resource allocation. , The resource is reallocated by performing the negotiation of (3).

[0005]

However, the above-mentioned conventional distribution method has the following problems.

The first problem is that since the number of agents is set in advance, the number of agents may not be appropriate depending on the scale of the problem.

The second problem is that the distribution to the agents satisfies only the request number balancing, and the load for processing individual requests and the proximity of requests (positional relationship between requests) are required. Therefore, there is a factor that prolongs the processing time, such as variations in the load between agents at the time of execution and the possibility of frequent resource competition.

The present invention has been made in view of the above points, and solves the conventional problems, determines the number of agents according to the scale of the problem, and balances the number of requests.
Coordination that distributes to agents in consideration of the load required for processing requests and the proximity of requests, suppresses load variations at execution, reduces resource contention, and shortens processing time An object is to provide a demand distribution method in a problem solving system and a demand distribution system in a cooperative problem solving system.

[0009]

A request distribution method in a cooperative problem solving system according to the present invention is a method for distributing a request to a plurality of agents in response to a resource allocation problem in which a limited resource is allocated to a plurality of requests. When requests are distributed to agents of a cooperative problem solving system that solves problems by coordinating tasks, the optimum number of agents according to the resource amount of the problem and the scale of the request amount is used by using the information about resources and the information about individual requests. Then, the requests are clustered into a request group of the number of already-determined agents, and the clustered request group is distributed to the agents.

FIG. 1 is a diagram for explaining the principle of the present invention.

According to the present invention, a request distribution system in a cooperative problem solving system distributes a request to a plurality of agents for a resource allocation problem in which limited resources are allocated to a plurality of requests, and the problem is solved by cooperation among the agents. In a system that distributes requests to agents of a cooperative problem solving system that performs the above-mentioned method, the optimum number of agents according to the resource amount of the problem and the scale of the request amount is determined using the information 130 and 140 on the resource and individual requests. Number of agents determining unit 100, clustering unit 110 that clusters the request into a request group of the number of agents already determined by the agent number determining unit 100, and request to distribute the request group clustered by the clustering unit 110 to the agents Distribution means 12 Having. Further, the clustering unit 110 distributes requests having a high degree of proximity with a small variation in load during execution among the agents having the number of agents determined by the number-of-agents determining unit 100 to the same agent.

[0012]

FIG. 2 is a diagram for explaining the operation of the present invention. In the method of distributing requests to agents according to the present invention, first, the number of agents is determined according to the scale of the problem (step 1). The scale of the problem is estimated using the information about the resource and the information about the demand, and the number of agents is changed according to the scale of the problem. This makes it possible to determine the optimal number of agents according to the scale of the problem.

Next, the requests are clustered into a request group of agents (step 2). Clustering is performed in consideration of load balancing required for processing requests and proximity of requests. As a result, it is possible to perform clustering for each request having a small load variation between clusters and high proximity.

Finally, the request group clustered to the number of agents is distributed to the actual agents (step 3). Which request group is distributed to which agent is arbitrary.

By using the procedure as described above, the present invention collects requests having a high degree of proximity with a small variation in execution load among agents with an optimum number of agents according to the scale of the problem. By distributing the requests to the same agent, it is possible to reduce the occurrence of resource competition and distribute the requests to the agents, which leads to a reduction in processing time.

[0016]

Embodiments of the present invention will now be described in detail with reference to the drawings. According to the present invention, request distribution of a cooperative problem solving system for a resource allocation problem on a network is performed by changing the number of agents according to the scale of the problem, clustering the requests into a request group of the agent, and actually executing the request group. This is done by distributing to the agents.

The details will be described below.

FIG. 3 shows an example of a resource allocation problem on a network according to an embodiment of the present invention. The network shown in the figure is composed of nodes a to p and links connecting them. The request has a start point node and an end point node. In the figure, there is a request that the start point node is "a" and the end point node is "i", and that the start point node is "m" and the end point node is "k". The resource allocation problem on the network is one in which a plurality of requests are used as links and resources are set between nodes indicated by the requests.

Hereinafter, a method of determining the number of agents according to the scale of the problem, a load required for request processing, and a clustering method considering the proximity of requests will be described in order.

[Determination of Number of Agents] First, the number of agents is determined to be an optimum number according to the scale of the problem.

Here, the scale of the problem is represented by a resource amount and a request amount, and the number of nodes n and the number of links m of the network are used as the parameter of the resource amount, and the request number r and the request length L _i are used as the parameter of the request amount.

FIG. 4 is a diagram for explaining parameters of the resource amount and the required amount according to the embodiment of the present invention.

The request length L _i represents the size of the request, and the larger the request, the longer the processing of the request. Here, the size of the request is represented by the distance between the start and end points (the minimum number of via links).

First, the problem size is approximated by using these four parameters (number of nodes, number of links, number of requests, request length). In FIG. 4, it is assumed that the number of nodes n = 16, the number of links m = 24, the number of requests r = 2, and the required length L _i = 5, 3.
For example, the scale of the problem is approximated by the following formula. The fourth term on the right side is based on the assumption that the search range is proportional to the square of the required length, and the processing time of the request is also proportional to this.

[0025]

[Equation 1]

In FIG. 4, when the calculation is performed with α = β = γ = δ = 1, the scale of the problem is 16 + 24 + 2 + (25 + 9) = 76.

Next, the relationship between the approximated scale and the optimum number of agents is obtained. Using the problem size and the number of agents as parameters, the number of agents that minimizes the processing time is experimentally determined for each problem size. However, the scale of the problem is set discretely using the representative value for each certain section.

Then, if it is obtained from the relation of the optimal number of agents according to the scale of this problem, thereafter, for each problem,
The optimal number of agents can be determined by calculating the size of the problem.

For example, the optimum number of agents is simulated by using the number of agents and the scale of the problem as parameters, and the number of agents that minimizes the processing time is experimentally obtained. From this preliminary experiment, for example, if the following relationship is obtained,

[0030]

[Table 1]

In the example of FIG. 4, since the problem scale is 76, "3" is acquired as the optimum number of agents from the nearest problem scale 80.

[Request Clustering] A plurality of given requests are clustered into a request group (cluster C _i ) of the number of agents obtained by the above method.

FIG. 5 is a flow chart showing the procedure of clustering according to an embodiment of the present invention.

Step 101) Each request is set as each cluster (the total number of clusters is equal to the number of requests).

Step 102) If the number of agents is equal to or larger than the number of clusters, the process proceeds to step 106, and if the number of agents is smaller than the number of clusters, step 10
Move to 3.

Step 103) The similarity between clusters is calculated for all the clusters.

Step 104) Merge the largest cluster verb into either one. This reduces the total number of clusters by one.

Step 105) If the number of clusters is equal to the number of agents, the process is terminated. That is, step 10
The process of 3 and step 104 is repeated until the total number of clusters reaches the number of agents, and when the number of clusters reaches the number of agents, clustering ends.

Step 106) If the number of clusters is less than or equal to the number of agents, the number of agents is set to the number of clusters, and the process ends.

Here, the similarity between the clusters is defined in consideration of the proximity between the clusters and the balancing of the total processing load among the clusters. Proximity between clusters aims to reduce the occurrence of resource competition with other agents by collecting requests that are close in position and distributing them to the same agent. On the other hand, the balancing of the total processing load among the clusters aims at reducing the occurrence of resource competition with the agent at the time of execution.

Inter-cluster proximity: The inter-cluster proximity represents how close the positional relationship between clusters is. For example, the distance between the closest points between the clusters or the distance between the centers of gravity between the clusters is used as the degree of proximity. Figure 6
It is a figure for demonstrating the proximity degree between clusters. This figure shows an example in which the distance between the centers of gravity is used as the degree of proximity. Each request is represented by a straight line connecting the start point and end point (request line), and its center is the center of gravity. When there are multiple requests in the cluster, the center of gravity of the respective requests is set as the center of gravity of the cluster. The proximity between the clusters is represented by the distance between these centers of gravity.

Processing load: The load required to process a request is
Assuming that it is proportional to the search range, it is approximated by the square of the required length as when determining the number of agents. Therefore, the total processing load in the cluster is the sum of squares of the individual request lengths L _i .

From the above, the similarity between the clusters C _i and C _j is defined as follows.

[0044]

[Equation 2]

The clustering will be specifically described below. As described above, for the similarity, the similarity between the clusters is calculated from the positions of the start and end points of the request and the request length, and the clusters with the highest similarity are validated. And The number of requests = 5 and the number of agents = 2.

Cluster Setup Individual requests are set to the cluster. C1 = {r1} C2 = {r2} C3 = {r3} C4 = {r4} C5 = {r5} ... Cluster number = 5.

Calculation of similarity between clusters Since the number of clusters> the number of agents, the similarity between clusters is calculated. The degree of similarity between clusters is as follows.

C1 C2 C3 C4 C5 C1-3 4 5 1 C2-2 4 2 C3-4 2 C4-3 Cluster fusion The maximum similar clusters are fused. In this case, "C
1 ”and“ C4 ”are merged into a new C1. As a result, C1 = {r1, r4} C2 = {r2} C3 = {r3} C5 = {r5} ... Number of clusters = 4. Becomes

The above process is repeated until the number of clusters below becomes the number of agents.

[0050] Here, C3 and C5 are fused to be new C3. As a result, C1 = {r1, r4} C2 = {r2} C3 = {r3, r5} ... Cluster number = 3.

Furthermore, C1 and C2 are fused together to newly create C1.
And As a result, C1 = {r1, r2, r4} C3 = {r3, r5} ... Cluster number = 2.

End of clustering When the number of clusters is equal to the number of agents, the process ends. Here, the clustering result is C1 = {r1, r
2, r4} and C3 = {r3, r5}.

By performing clustering using such a similarity, it is possible to collect requests with small variations in load between clusters and high proximity.

By distributing the group of requests thus clustered to each agent, it is possible to reduce the variation in the load of each agent at the time of execution, reduce the occurrence of resource competition, and shorten the processing time. Become.

[0055]

As described above, according to the present invention, in a cooperative problem solving system that distributes requests of a plurality of agents to a resource allocation problem and cooperates between the agents to solve the problem, the scale of the problem ( The number of agents can be used according to the resource quantity and the request quantity), and the processing time can be reduced by distributing the requests to the agents so as to suppress the variation in the load of each agent at the time of execution and reduce the occurrence of resource competition. Effective in shortening.

[Brief description of drawings]

FIG. 1 is a diagram for explaining the principle of the present invention.

FIG. 2 is a diagram for explaining the operation of the present invention.

FIG. 3 is a diagram showing an example of a resource allocation problem on a network according to an embodiment of the present invention.

FIG. 4 is a diagram for explaining parameters of a resource amount and a request amount according to an embodiment of the present invention.

FIG. 5 is a flowchart showing a procedure of clustering according to an embodiment of the present invention.

FIG. 6 is a diagram for explaining proximity between clusters.

[Explanation of symbols]

 100 Agent Number Determining Means 110 Clustering Means 120 Request Distributing Means 130 Resource Information 140 Request Information

Claims (3)

[Claims] 1. For a resource allocation problem in which limited resources are allocated to a plurality of requests, the requests are distributed to a plurality of agents, and the requests are distributed to agents of a cooperative problem solving system that solves problems by cooperation among the agents. In this case, the optimum number of agents according to the scale of the problem is determined using the information about the resource and the information about each request, the requests are clustered into a request group of the number of already-determined agents, and the clustered request group Request distribution method in a collaborative problem solving system, which is characterized by distributing requests to agents. 2. For a resource allocation problem in which limited resources are allocated to a plurality of requests, the requests are distributed to a plurality of agents, and the requests are distributed to agents of a cooperative problem solving system that solves problems by cooperation among agents. In this system, the number of agents determining means for determining the optimum number of agents according to the resource amount in question and the scale of the request amount using the information about the resource and the information about the individual request, and the request number determining means Request distribution in a collaborative problem solving system, characterized by having a clustering means for clustering into a request group of the number of agents already determined by the above, and a request distribution means for distributing the request group clustered by the clustering means to the agents. system. 3. The clustering means distributes requests having a high degree of proximity and having a small variation in load during execution among agents of the number of agents determined by the number-of-agents determining means, to the same agent all together. A request distribution system in the cooperative problem solving system described in 2.