d analyzes middle result in Blackboard and then outputs final result via Driver.

We’ll present the reciprocity between the manager and the computing agents in 3.2.

  3.2 SYSTEM ARCHITECTURE

 

Thousands of scheduling algorithms has been proposed so far. These scheduling algorithms can be classified into several categories. The hierarchy is shown in Fig. 3:  

Of course we can not and need not design agents for each algorithm. But we can do that for each class. Our solution is to joint different classes of computing agents into a MASS to realize dynamic integration of scheduling algorithms. Except for a manager, every node of the system is a computing agent, which provides congener algorithms that store in its algorithm base. A starlike architecture of the system is shown in Fig.4.

Fig.4 is only one example of multi-agent based integration of scheduling algorithms system. We may choose either two or more computing agents to build a MASS according to the requirements of a real-life production system. And computing agents can be distributed. They work independently and concurrently.

 The system works as follow:

 1)         The manager searches information from the environment and translates it into a scheduling model. Then it broadcasts a bid request to all computing agents and send an expression of the scheduling mode

l;

 2)         When receiving bid, computing agents retrieve their own Rule Base. Whether or not there are records concerning with the model means whether or not the agents have capability to solve the problem. According to their capabilities, the computing agents decide to bid or not and then send messages to the manager.

 3)         The manager makes a choice from computing agents who make the bids and send information to the selected agents.

 4)         The selected computing agent accomplishes the task and outputs its result to the manager.

 5)         If the current result does not fulfill system’s goal, the manager will invite a new public bid.

 Sometimes, one problem can be divided into several subproblems. Then manager will send bids to several computing agents concurrently. And subproblems are solved independently by computing agents.

 The computing agent in the above scenario is “fat “. Facing requirement of some cases, We can also design a MASS of “thin” computing agent, i.e. computing agents is relatively simple, e.g. it provide only one algorithm, not a algorithms base. Anyway, the idea of building MASS in this paper is flexible.

 4. Conclusion

 Classical scheduling theory is problem-based. It is difficult to be applied directly to practice since it simplifies the real-life scheduling system, which is usually complex and contains several problems. Whereas, in this paper, we proposed a mechanism named integration of scheduling algorithms and built architecture of process-type MASS to support it. The mechanism is system-based. It ca

      [5]