Empirical Evaluation of the Parallel Distribution Sweeping Framework on Multicore Architectures

TL;DR

This paper empirically evaluates a parallel distribution sweeping framework on multicore architectures, demonstrating that simple I/O-efficient algorithms outperform traditional methods in geometric batched problems.

Contribution

It implements and tests a parallel distribution sweeping framework for geometric problems, showing practical efficiency on modern multicore systems.

Findings

Fewer DRAM accesses compared to traditional algorithms

Outperforms plane sweep and divide-and-conquer methods

Effective use of simple I/O-focused models on real hardware

Abstract

In this paper, we perform an empirical evaluation of the Parallel External Memory (PEM) model in the context of geometric problems. In particular, we implement the parallel distribution sweeping framework of Ajwani, Sitchinava and Zeh to solve batched 1-dimensional stabbing max problem. While modern processors consist of sophisticated memory systems (multiple levels of caches, set associativity, TLB, prefetching), we empirically show that algorithms designed in simple models, that focus on minimizing the I/O transfers between shared memory and single level cache, can lead to efficient software on current multicore architectures. Our implementation exhibits significantly fewer accesses to slow DRAM and, therefore, outperforms traditional approaches based on plane sweep and two-way divide and conquer.