Multi-agent Based Integration of Scheduling Algorithms

nager.

 3.1 DEFINITION OF AGENTS

 A whole computation consists of several steps: environment analysis, goal setting, evaluation of computation capability, decision making, computing, output conclusion, etc. We integrated these steps into a general model of computing agent as Fig.1

 Elements of a computing agent are detailed as follow:

 1)  Algorithm Base

 Stores algorithms that belong to the same type, e.g. scheduling rules. Each algorithm can be used inside the agent according to condition of their being used. Also, new algorithms belong to the type can be added in.

 In fact, the contents in the base may be information as: ID of an algorithm, Input, etc. It points to a program of an algorithm.

 2)  Rule Base

 Stores knowledge of using algorithms: applicable conditions, capability, efficiency, etc. New rule can be added in at any time.

 The rules use the format of 4-vector: (ID, Condition, Capability, Efficiency), thereinto: 

l     ID: ID of the algorithm;

 

l     Condition: relation between the algorithm with some scheduling models, i.e. if the algorithm can use in one of algorithms.

 

l     Capability: degree of optimization. It is a relative value of an algorithm we select as a standard.

 

l     Efficiency: the time of finishing computation.

 

3)  Analyzer 

Analyzes information from the sensor and makes decision of whether or not responding to the information.

 Information from sensor is m

ainly ID of scheduling model. It then be use to retrieve ID of algorithms in Rule Base. Finding a suited ID of an algorithm means agents can response the scheduling model.

 4)  Reason Machine

 If there are several algorithms in the Algorithms Base that are suited with the scheduling model, selects the best one according to capability and efficiency under the support of the rule base.

 5)  Computing Cell

 Finishing computation with selected algorithm.

 6)  Sensor

 Receives information such as jobs, resources and scheduling models from the Manager and responds with bidding or not.

 7)  Driver

Outputs result.

 In order to harmonize computing agents, we need a manager. It’s a special agent as shown in Fig.2 and is responsible for following functions:

 1)         Registers each computing agent with Register Model and stores their properties in Database.

 2)         Searches information from environment through the Sensor and translates it into appropriate scheduling model in the Modeler under the support of the Knowledge Base.

3)         Communicates with computing agents via Communicator.

4)         Records and anal

《Multi-agent Based Integration of Scheduling Algorithms(第4页)》