Abstract
In view of the small molecular model established in the field of high voltage insulation, the actual operation of transformer cannot be fully reflected at micro-level. Therefore, this paper aims to improve the performance of computing environment and expand molecular scale. Firstly, two servers were connected through a high-speed communication network as the initial cluster architecture. Secondly, spatial decomposition and load balancing algorithms were used to improve the operation efficiency of cluster. Meanwhile, the oil-paper composite media model of 105 atoms could be established based on this cluster, but it consumed a lot of time. Therefore, we analysed the relationship between operation efficiency and four characteristic quantities such as central processing unit performance, number of cores, simulation time and number of nodes. Then the highest point of cluster operating efficiency was found through continuous optimisation. It can be summarised that the calculating speed of cluster is nearly 10 times faster than that of the large server. Meanwhile, according to the results based on this cluster, it can be concluded that water molecules would migrate towards the oil during heating. When the initial moisture content in paper is high, the high water region would appear at the oil-paper interface.