Abstract
Contemporary Online Analytical Processing systems often make extensive use of view materialization in order to minimize the cost of run-time query processing on massive data sets. Due to the cost of generation and maintenance, however, such views have typically been updated only in periodic batches. That being said, users of Big Data analytics platforms are demanding an increasingly shorter update cycle. In this paper, we present a framework for what we call soft real-time OLAP. Specifically, we employ view partitioning and maintain an additional "hot" partition that absorbs incoming update streams. We then augment this strategy with multi-core processing so as to further accelerate view construction and query resolution. Initial experiments with input sets of up to 10 million tuples show that our framework improves update performance by almost two orders of magnitude, while significantly reducing view construction and query costs.