我們設計了一個以Mobile Agent為基礎的平行處理Framework,目標在提供可以在執行期間動態地改變運算單位數量與位置的能力,以便可以隨時增加運算效能。在實作上,我們使用一個常見的平行處理模型:Master/Slave Model,加以改進(在這裡我們把它叫做Active Worker Model)使其可以適合在Mobile Agent平台上執行,並透過繼承自Java與Mobile Agent的長處,如跨平台特性、Dynamic Class Loading、Dynamic Proxy Generation、Code Mobility等等,使得我們可以在異質的環境下輕易的執行。同時藉由以不同的實作方法配合實際測試結果,找出較好的實作方式,使得我們可以在彈性與效能上取得平衡點。
We designed a java-based mobile agent framework that has the capability of changing the location and number of computation units at runtime dynamically. In the implementation, we use a common model of parallel processing, Master/Slave Model. The model is modified for the java-based mobile agent and is called Active Worker Model. By taking advantages of Java and mobile agent characteristics such as heterogeneity, dynamic class loading, dynamic proxy generation and code mobility, the framework can be used in heterogeneous environment more easily. The framework was implemented in two different ways where their efficiencies are compared.