Scalable Distributed String Sorting
Florian Kurpicz, Pascal Mehnert, Peter Sanders, Matthias Schimek
TL;DR
String sorting on distributed-memory machines faces scalability limits due to latency and communication overhead. The authors propose a multi-level string sorting framework that recursively partitions processors into groups and performs LCP-aware merging, with a prefix-doubling variant that drives per-level work toward the sum of distinguishing prefixes $D$. They introduce building blocks such as an $r$-way LCP loser-tree merge, an improved robust hypercube quicksort for strings, and a multi-level Bloom filter, along with analyses demonstrating near-$N$ (or near-$D$ for the prefix-doubling variant) per-level work and communication. Empirically, the approach scales to $p=49152$ cores and achieves up to 5x speedups over the prior distributed string sorting methods, enabling efficient large-scale indexing and text-processing pipelines on HPC systems.
Abstract
String sorting is an important part of tasks such as building index data structures. Unfortunately, current string sorting algorithms do not scale to massively parallel distributed-memory machines since they either have latency (at least) proportional to the number of processors $p$ or communicate the data a large number of times (at least logarithmic). We present practical and efficient algorithms for distributed-memory string sorting that scale to large $p$. Similar to state-of-the-art sorters for atomic objects, the algorithms have latency of about $p^{1/k}$ when allowing the data to be communicated $k$ times. Experiments indicate good scaling behavior on a wide range of inputs on up to 49152 cores. Overall, we achieve speedups of up to 5 over the current state-of-the-art distributed string sorting algorithms.