Thesis: Interactive Data Analytics using GPUs
Modern GPUs can provide an order-of-magnitude higher memory bandwidth compared to CPUs which can enable interactive query runtimes on analytical queries. However, the massively parallel architecture of GPUs requires rearchitecting in-memory data analytics systems in order to achieve optimal performance. In my PhD, we developed Crystal, a framework for writing high performance SQL query implementations for GPU. We showed that using GPUs can lead to significant performance improvement (~24x in our setup). Finally, we developed theoretical models for estimating query runtime and showed that Crystal achieves these lower bounds.
