The paper Physical Oscillator Model for Supercomputing by Ayesha Afzal, Georg Hager, and Gerhard Wellein received the Best Short Paper Award at the 14th International Workshop on Performance Modeling, Benchmarking and Simulations of High Performance Computer Systems (PMBS23), which is to take place in conjunction with the SC23 conference in Denver, Colorado, on November 13.
In this paper, a novel idea for modeling parallel programs with regular compute-communicate cycles was suggested: a physical system comprising set of coupled harmonic oscillators. Motivated by the iconic Kuramoto Model, which had been used to describe, among others, the synchronization of blinking fireflies and coupled metronomes, the researchers modified the nonlinear interaction among the oscillators to mimic resynchronizing and desynchronizing behavior of processes in parallel programs running on supercomputers. Describing a parallel program as a nonlinear mechanical system opens exciting new possibilities for program characterization and for understanding the complex dynamics of parallel code. The typical emerging patterns depend mainly on the bottleneck structure of the hardware and how the software interacts with it.
https://doi.org/10.1145/3624062.3625535
Focus to learn more
