Towards Parallel Learned Sorting

Abstract

We introduce a new sorting algorithm that is the combination of MLenhanced sorting with the In-place Super Scalar Sample Sort (IPS 4 o). The main contribution of our work is to achieve parallel ML-enhanced sorting, as previous algorithms were limited to sequential implementations. We introduce the In-Place Parallel Learned Sort (IPLS) algorithm and compare it extensively against other sorting approaches. IPLS combines the IPS 4 o framework with linear models trained using the Fastest Minimum Conflict Degree algorithm to partition data. The experimental results do not crown IPLS as the fastest algorithm. However, they do show that IPLS is competitive among its peers and that using the IPS 4 o framework is a promising approach towards parallel learned sorting. I would also like to thank my friends Matthew Currie, Sean Roarty, and Kathryn Lecha. Thanks for listening to incomplete parts of this thesis, often when the details were unclear and things went wrong. vi