Abstract:Scientific applications often contain large, computationally-intensive, and irregular parallel loops or tasks that exhibit stochastic characteristics. Applications may suffer from load imbalance during their execution on high-performance computing (HPC) systems due to such characteristics. Dynamic loop self-scheduling (DLS) techniques are instrumental in improving the performance of scientific applications on HPC systems via load balancing. Selecting a DLS technique that results in the best performance for different problems and system sizes requires a large number of exploratory experiments. A theoretical model that can be used to predict the scheduling technique that yields the best performance for a given problem and system has not yet been identified. Therefore, simulation is the most appropriate approach for conducting such exploratory experiments with reasonable costs. This work devises an approach to realistically simulate computationally-intensive scientific applications that employ DLS and execute on HPC systems. Several approaches to represent the application tasks (or loop iterations) are compared to establish their influence on the simulative application performance. A novel simulation strategy is introduced, which transforms a native application code into a simulative code. The native and simulative performance of two computationally-intensive scientific applications are compared to evaluate the realism of the proposed simulation approach. The comparison of the performance characteristics extracted from the native and simulative performance shows that the proposed simulation approach fully captured most of the performance characteristics of interest. This work shows and establishes the importance of simulations that realistically predict the performance of DLS techniques for different applications and system configurations.

An experimental analysis environment for logical process simulation algorithms

Computation platform for automatic analysis of embedded software systems using model based approach

SyPSE: A Symbolic Computation Toolbox for Process Systems Engineering Part II─Design for PSE Applications

LLM and Simulation as Bilevel Optimizers: A New Paradigm to Advance Physical Scientific Discovery

BeSimulator: A Large Language Model Powered Text-based Behavior Simulator

An Accurate Multi-processing Simulator Based on ADL

On the Design and Analysis of LLM-Based Algorithms

An Approach for Realistically Simulating the Performance of Scientific Applications on High Performance Computing Systems

Automated discovery of business process simulation models from event logs

Simulating Petri nets with Boolean Matrix Logic Programming

Simulation Modeling of a Large-Scale Formal Verification Process

Code Simulation Challenges for Large Language Models

AgentSimulator: An Agent-based Approach for Data-driven Business Process Simulation

Search Strategies for Self-driving Laboratories with Pending Experiments

An Extension of the Systems Modeling Language for Physical Interaction and Signal Flow Simulation

Enhancing LLMs for Power System Simulations: A Feedback-driven Multi-agent Framework

Constructing Hybrid Software Process Simulation Models.

Can Language Models Pretend Solvers? Logic Code Simulation with LLMs

A Systematic Methodology for Developing Discrete Event Simulation Models of Software Development Processes

LLM experiments with simulation: Large Language Model Multi-Agent System for Simulation Model Parametrization in Digital Twins

Process Simulation for Software Engineering Education