We are living in the era of Big Data and witnessing the explosion of data. Given that the limitation of CPU and I/O in a single computer, the mainstream approach to scalability is to distribute computations among a large number of processing nodes in a cluster or cloud. This paradigm gives rise to the term of data-intensive computing, which denotes a data parallel approach to process massive volume of data. Through the efforts of different disciplines, several promising programming models and a few platforms have been proposed for data-intensive computing, such as MapReduce, Hadoop, Apache Spark and Dyrad. Even though a large body of research work has being proposed to improve overall performance of these platforms, there is still a gap between the actual performance demand and the capability of current commodity systems. This paper is aimed to provide a comprehensive understanding about current semantics-aware approaches to improve the performance of data-intensive computing. We first introduce common characteristics and paradigm shifts in the evolution of data-intensive computing, as well as contemporary programming models and technologies. We then propose four kinds of performance defects and survey the state-of-the-art semantics-aware techniques. Finally, we discuss the research challenges and opportunities in the field of semantics-aware performance optimization for data-intensive computing.