To share or not to share vector registers?

Abstract

Query execution techniques constantly adapt to novel hardware features to achieve high query performance, in particular for analytical queries. In recent years, vectorization based on the Single Instruction Multiple Data (SIMD) parallel paradigm has been established as a state-of-the-art approach to increase single-query performance. However, since concurrent analytical queries are executed independently potentially invoking a set of fully vectorized operators, the same data accesses and computations among different queries may be executed redundantly. Various techniques have already been proposed to avoid such redundancy, ranging from concurrent scans via the construction of materialized views to applying multiple query optimization techniques. Continuing this line of research, we now investigate the opportunity of sharing vector registers for concurrently running queries in analytical scenarios. In particular, our core sharing approach is to process data elements of different queries together within a single vector register. As we are going to show, sharing vector registers to optimize the execution of concurrent queries can be very beneficial in many cases. We therefore demonstrate the feasibility of a new work sharing strategy and thus open up a wide spectrum of future research opportunities.