In this presentation, we first talk about the volatility issue in big data analytics, discuss the relationships between statistics, computing and systems science. Then, we emphasize the importance of the reproducibility with examples from our current research, and list four main challenges with the current big data analysis techniques. We use evidence from our research results and/or projects to support our statements. Finally, we conclude that the key to deal with the addressed challenges is to make an adequate tradeoff between engineering and science for the big data analysis.