Parallel Algorithm Mobile Agent-based Data Set Of Equilibrium Strategies

To solve the problem of distributing data in parallel arithmetic, the method of dividing dataset into several equal parts and sending every part to a special computing node was adopted. If the system was constructed by the same type of computing nodes, then the method can work efficiently. On the other hand, it can make the system unbalanced.In this thesis, the dataset balancing strategy based on mobile agent in parallel arithmetic was proposed. The mobility of mobile agent enabled the parallel computing to be true and the dataset balancing to be easy. The strategy consisted of two part, DBSMAV(DataSet Balancing Strategy Based on Mobile Agent Velocity ) and DBSMAS(DataSet Balancing Strategy Based on Mobile Agent Scheduling). By using PLUM (Parallel Load balancing for Unstructured Meshes), SBN (Symmetric Broadcast Network) , MinEX (Minimal Extra overhead) , the thinking in Mobile Agent for reference, we proposed this strategy.It was proved that MAV(Mobile Agent Velocity) could reflect the performance of the computing node when the memory of the node was large enough. DBSMAV could balance the system when it divided the dataset according to MAV. In order to resolve the unbalanced problem led by the memory, the DBSMAS was proposed. In the DBSMAS strategy, the over load first scheduling principle was applied. When the DBSMAS strategy judged that scheduling a Mobile Agent could reduce the total time of the system, it scheduled the Mobile Agent to move to another node with it's dataset. By this way, the unbalanced system can be balanced.In our experiment, the dataset balancing strategy based on mobile agent in parallel arithmetic show high performance.