Yannis Chronis (University of Wisconsin-Madison): Analytic Query Processing using Associative Computing

Speaker:
Yannis Chronis (University of Wisconsin-Madison)

Title:
Analytic Query Processing using Associative Computing

Abstract:
We are in the midst of a “Cambrian” hardware evolution in which a variety of architectures are being invented with a flurry that we haven’t seen in a long time. The associative computing paradigm enables designs that utilize memories in new ways and packs storage and computing closely together. Associative computing processors offer massive data-level parallelism and single-cycle search and update primitives. The focus of our work is on exploring how database analytic workloads can run efficiently on this new computing substrate. Our vehicle for this research is the Castle project in which we are building a data processing engine to run on the CAPE hardware, which is a specific SRAM-based associative processor. To fully exploit CAPE’s potential, we adapt data query processing operators and query optimization methods to leverage the style of data-parallel computing that can be carried out natively inside CAPE. We experiment with both individual query processing operators and end-to-end queries and show that the performance gains using our methods are significant, and up to 128X in some cases. Tentatively, these results point to the exciting potential in accelerating database analytic workloads using associative computing.

Bio:
Yannis is a fifth-year Ph.D. student in the Computer Sciences Department at the University of Wisconsin-Madison advised by Prof. Jignesh Patel. His research focuses on efficient data processing using modern and emerging hardware.