Declarative Adaptive Optimization of Task-Based Applications on Heterogeneous Architectures
Emanuele De Angelis, Guglielmo De Angelis, Romolo Marotta, Federica Montesano, Alessandro Pellegrini, and Maurizio Proietti
Published in: Proceedings of the 2025 International Symposium on Computer Architecture and High Performance Computing Workshops
Abstract:
This paper presents a knowledge-based technique for mapping task-based applications onto heterogeneous computing resources using Answer Set Programming (i.e., ASP) for dynamic, multi-objective task allocation. Our method models applications through the Actor Model, considering device constraints, task workloads, and performance factors like computational overload and inter-actor communication costs. By formulating these elements as logical rules, our ASP-based method adapts allocations to changing workloads and system dynamics, nearing the theoretical optimum achievable by an oracle with complete knowledge. Simulation experiments show that our approach significantly outperforms (up to 45%) traditional static partitioning techniques by maximizing throughput and preventing unfruitful migrations. These results highlight the effectiveness of declarative optimization for online allocation in heterogeneous architectures, and suggest that a clear syntax for modelling non-functional metrics ease the extrapolation of a broad set of optimization scenarios.
BibTeX Entry:
author = {De~Angelis, Emanuele and De~Angelis, Guglielmo and Marotta, Romolo and Montesano, Federica and Pellegrini, Alessandro and Proietti, Maurizio},
title = {Declarative Adaptive Optimization of Task-Based Applications on Heterogeneous Architectures},
booktitle = {Proceedings of the 2025 International Symposium on Computer Architecture and High Performance Computing Workshops},
year = {2025},
month = oct,
publisher = {IEEE},
series = {SBAC-PADW},
location = {Bonito, Brazil},
note = {To Appear}
}