Abstract:
To develop common software systems, engineers and designers are currently using an object-oriented approach that helps build distributed object-oriented systems. A distinctive feature of distributed object-oriented systems is the distribution of classes of objects among various nodes. Typically, there is no information about the distribution of classes across servers in applications, which encourages a restructuring procedure that improves productivity. In the proposed approach, software restructuring of distributed object-oriented systems is carried out using the adaptive method using a neural network. At the initial stage, a graph of class dependencies is created, in which nodes represent classes, and relations between nodes correspond to dependencies between classes. Then, the properties of the classes included in this graph are extracted, which are transmitted as input to the neural network for its training. After that, based on the account of class dependencies, clustering is performed, leading to the partition of the set of classes of the distributed object-oriented system into loosely coupled subsets. Next, a graph of clusters is created, the vertices in which correspond to clusters, and the edges correspond to communication channels that may exist between clusters. The k-medoids algorithm is applied to the resulting graph, which is used to collect the clusters in such a way that the number of collected cluster groups becomes equal to the number of available system nodes. The resulting cluster groups are loosely coupled. Finally, cluster groups are assigned to various available nodes in a distributed environment. The simulation results showed that the proposed work gives more effective results compared to existing methods.
Fulltext:
PDF file (642 kB)