Abstract
The core component of Statistical Model Checking (SMC) is the repeated sampling of a given system as to estimate statistical measures. To obtain probabilistic estimates with high confidence a significant number of simulations is required, in particular in the presence of rare events. In this paper we explore the use of Graphical Processing Unit (GPU) for accelerating SMC for Networks of Stochastic Extended Timed Automata (SXTA). We discuss the many challenges and solutions required to achieve significant speedups on a GPU architecture. In collaboration with NVIDIA we develop a prototype tool for parallel SMC using both GPU and multi-core CPU. Experimental results demonstrate trade-offs in the computation time when utilizing either CPU or GPU. In one case we observed the GPU using 20% of the power of the CPU equivalent while delivering a 2.73 time speedup.
Original language | English |
---|---|
Title of host publication | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
Number of pages | 26 |
Publication date | 2025 |
Pages | 267-292 |
DOIs | |
Publication status | Published - 2025 |