Systematically Applying High-Level Mutations for Fuzz Testing Big Data Applications

Abstract

As the amount of data worldwide continues to grow the big data field is becoming increasingly important. Fuzz testing has shown to be an effective testing tool, and recent work has applied fuzz testing to big data applications. This study aims to contribute to knowledge on fuzz testing big data applications by extending on BigFuzz, a state-of-the-art fuzzing framework for big data applications. Our study offers an alternative mutation approach by systematically applying combinations of seven high-level mutation types, instead of selecting mutations randomly. Our findings show that 1) for three out of five benchmarks, systematic exploration finds a higher number of failures; 2) the amount of trials needed to find an equal number of failures is not increased by testing systematically for the majority of the benchmarks; 3) our configuration returns the best results when it explores with increased exhaustiveness; Thus, we show that systematically applying high-level mutations can find a higher number of unique failures in an equal number of trials.