Recommending just enough memory for analytics
Charles Reiss, Randy Howard Katz · 2013
MapReduce was designed by Google for large-scale data analysis on slow but cheap disk-based storage. Nevertheless, memory has declined in price to where cost-effective machines offer ever larger memory capacity. Furthermore, a more diverse data analyst community, with smaller datasets, has emerged. These trends motivate new parallel processing frameworks, like Spark [2], with better support for in-memory data analysis.