how to reduce time of spark application run by "java -jar"

Question

I execute a spark application two ways. Application is Naive bayes training using MlLib.

1. using "spark-submit": then successful execute on a set of data.
2. using "java -jar": then take more time form case 1.

In both case have same data set and same configuration.

Any one can help me to reduce time in execute spark application using "java -jar".