Simulation-Based Evaluation of Big Data Caching Mechanisms

This paper provides a simulation-based evaluation that addresses memory management problems throughout Big Data processing. A significant problem occurs with in-memory computing when there is not enough available memory for processing the whole chunk of data, and hence some data must be selected for deletion to make room for new ones. The selected research strategy is to use different cache selection and replacement algorithms, such as Adaptive Replacement Cache (ARC) and Low Inter-Reference Recency Set (LIRS) algorithms, besides the default one, which is Least Recently Used (LRU). A simulator was built by the authors to assess the use of different caching approaches on Big Data platforms. The evaluation showed that the LIRS and the ARC algorithms gave a better hit ratio for different workloads than the LRU algorithm.