Abstract

Exploiting parallelism is the name of the game for executing analytical queries with low latency in in-memory database engines. Most prominently, modern general-purpose CPUs, which continue to dominate the area of computing units, offer high-computational power through two data-oriented parallel paradigms: MIMD and SIMD. Unfortunately, as both parallel paradigms exhibit different programming models and memory access patterns, exploiting both in a combined fashion is challenging. However, recent hardware advances for SIMD on CPUs have relaxed the restrictions on SIMDfriendly memory access patterns. The availability and performance of alternative access patterns has improved significantly compared to the state-of-the-art of a pure linear access pattern. As we will demonstrate in this paper, these advancements pave the way for a unified parallelization approach that harnesses both MIMD and SIMD in a combined fashion, offering a novel and promising way for an efficient analytical query processing.