Automatic translation from Java to Spark

Abstract

Cloud computing is a powerful infrastructure tool for scientific computing and research and attracts many developers to deploy their products on it. Code migration from local to cloud servers is a significant challenge for all companies and organizations that have many legacy programs. Compared to the de novo development method, the automatic translation is more productive and economical. In this paper, a new Java to Spark (J2S) translator is introduced to achieve the automatic translation from sequential Java code to Spark cloud application, which can translate three types of computing intensive programs. The evaluation experiments demonstrate that these translations significantly improve the source program performance in their domains. It is believed that it is a new step in the automatic code migration era of cloud computing.