MapReduce Data Skewness Handling: A Systematic Literature Review

One of the most successful techniques in large-scale data-intensive computations is MapReduce programming. MapReduce is based on a divide and conquer approach that uses commodity computers, also known as nodes, for parallel processing. The scalability and performance of this technique are more related to the type of data distribution in map and reduce tasks. Because of many reasons such as node failure, network failure, data skewness, etc. completion time of one task could be longer than other tasks, job completion time is determined by the slowest task. One of the most important reasons for requiring more time to finish one task compared to other tasks is the skewness of data. Despite the widespread use of MapReduce because of its high flexibility and tolerability of the error, with the presence of data skewness, it cannot fully utilize the nodes for parallel processing. The objectives of this study were to review related articles and classify them based on the type of problem addressed and to determine the advantages and disadvantages of them. Open issues were also defined to present guidelines for future research on this subject. In order to achieve the aforementioned objectives, some research questions were defined and answered. In this review, it was concluded that there are important parameters have not been considered in MapReduce data skewness handling approaches.