Applications are often distributed and accessible on the network for being downloaded when needed. However, the way applications are created does not usually consider the optimization of downloading time and bandwidth usage. This is a relevant problem for applications that are deployed in relatively slow networks. When the download time means cost, like usually for the GSM/GPRS communication network, applications that are packaged for reducing the average downloading time in a set of classical execution scenarios will improve the user experience.

This paper presents an approach and supporting toolset for packaging a Java application as a set of jar files so that the downloading time/cost for a set of execution scenarios is minimized. The approach relies on run-time analysis performed by executing scenarios over a dynamically downloaded and traced application, and uses Genetic Algorithms for determining the sub-optimal packaging of classes/resources in jar files.