FastLanes is a project initiated at CWI, intended as a foundation for next-generation big data formats. In this first paper on FastLanes, we focus on significantly improving data decoding performance over the state-of-the art. in a follow-up we will discuss improvements in compression ratio, as well. It introduces a new layout for compressed columnar data, that increases the opportunities for data-parallel decoding, improving performance by factors. It does so in a way that works across the heterogeneous and evolving Instruction Set Architectures (ISAs) landscape, is future-proof, and minimizes technical debt by relying on scalar-only code.
Afroozeh, A. (2023). cwida /FastLanes.