Improving the speed of global parallel optimization on PDE models with processor affinity scheduling

Abstract

Parallel global optimization of expensive simulation models like nonlinear partial differential equations (PDEs) can speed up model calibration or project design decisions, but the impact of memory management on the efficiency of using parallel global optimization methods has not been previously studied. This paper quantifies cache memory limitations arising during parallel optimization of expensive PDE models. An efficient parallel optimization algorithm is applied to model calibration for two different, expensive real-world PDEs (i.e., hydrodynamic and water quality analysis for a 250-hectare lake). One of these two lake models takes 4.5 h per simulation in serial, but that PDE simulation time per simulation increases to 12 h with parallel optimization if default processor scheduling strategy is used on a modern nonuniform memory access multicore platform. We proposed a novel mixed affinity scheduling strategy for parallel simulation optimization that increases computational efficiency by as much as 20% over the default affinity strategy.

Surrogate Global Optimization for Identifying Cost‐Effective Green Infrastructure for Urban Flood Control With a Computationally Expensive Inundation Model

“…Besides the use of surrogate models, parallel computing techniques would also be very helpful in reducing computational time. There are efficient parallel surrogate optimization methods developed recently, which use parallel computing and surrogate techniques to reduce the computational time (Xia & Shoemaker, 2021; Xia et al., 2021; Xia & Shoemaker, 2022). These parallel surrogate optimization methods could be considered for problems in a much larger scale (e.g., models take hours or days to run for a single simulation).…”

A Framework to Calibrate Ecosystem Demography Models Within Earth System Models Using Parallel Surrogate Global Optimization

“…In step (b) of the above optimization processes (Section 2.5, Figure 2), there are multiple objective functions (i.e., the computationally expensive ED models) running simultaneously on multiple processors. To overcome the cache memory limitations arising during parallel optimization of expensive models, we apply the mixed affinity processor scheduling strategy developed by Xia and Shoemaker (2021) to schedule the simulation evaluation tasks over multiple processors on the supercomputer platform. The mixed affinity scheduling strategy assigns each simulation task to one processor so they cannot switch around multiple processors.…”

“…The mixed affinity scheduling strategy assigns each simulation task to one processor so they cannot switch around multiple processors. Xia and Shoemaker (2021) shows that this scheduling strategy can improve the usage of memory resources and increase the speed of the optimization processors over the default processor scheduling strategy of the operation system. In this study, this mixed affinity processor scheduling strategy increases computational efficiency of the whole optimization process on ELM‐FATES by up to 50%.…”

“…Xia et al. (2021) have demonstrated that PODS outperforms other state‐of‐art parallel surrogate optimization algorithms on eight well‐known test problems. They also successfully applied PODS to calibrate computational expensive 3D hydrodynamic lake models with high speed and accuracy (Xia et al., 2020).…”

“…PODS also uses some ideas from the (nonsurrogate) dynamical coordinate search (DDS) (Tolson & Shoemaker, 2007) to choose which decision variables to perturb, which has proven to be efficient for optimization problems with many decision variables (Regis & Shoemaker, 2013;Tolson & Shoemaker, 2007). Xia et al (2021) have demonstrated that PODS outperforms other state-of-art parallel surrogate optimization algorithms on eight well-known test problems. They also successfully applied PODS to calibrate computational expensive 3D hydrodynamic lake models with high speed and accuracy .…”