Enabling Scalable Parallel Discrete Event Simulation
Event details
| Date | 15.07.2025 |
| Hour | 14:00 › 16:00 |
| Speaker | Pooria Poorsarvi Tehrani |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Thomas Bourgeat
Thesis advisor: Prof. Babak Falsafi
Co-examiner: Prof. Anne-Marie Kermarrec
Abstract
As computing environments scale out across distributed clusters and scale up with increasingly heterogeneous components, full-system simulation has become more difficult. Simulating these systems requires decomposing the target into interacting discrete event simulators (DES) and coordinating them through Parallel Discrete Event Simulation (PDES). This introduces three challenges: how to partition the states and events, communicate between simulators, and enforce local causality in each one.
We propose two promising directions for future work. First, we consider increasing the way states and events are partitioned using a hierarchical approach to how events are communicated and local causality is enforced. Second, we suggest exploring temporal partitioning through statistical sampling, which enables timing simulation to be parallelized across independent time windows. While this requires extending PDES to earlier simulation phases, functional simulation, functional warming, and detailed warming, we believe this direction can unlock significant speedups without adding local causality enforcement overhead.
Selected papers
1. The Wisconsin Wind Tunnel: virtual prototyping of parallel computers: https://dl.acm.org/doi/10.1145/166962.166979
2. dist-gem5: Distributed simulation of computer clusters: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7975287
3. SimBricks: end-to-end network system evaluation with modular simulation: https://dl.acm.org/doi/10.1145/3544216.3544253
Exam president: Prof. Thomas Bourgeat
Thesis advisor: Prof. Babak Falsafi
Co-examiner: Prof. Anne-Marie Kermarrec
Abstract
As computing environments scale out across distributed clusters and scale up with increasingly heterogeneous components, full-system simulation has become more difficult. Simulating these systems requires decomposing the target into interacting discrete event simulators (DES) and coordinating them through Parallel Discrete Event Simulation (PDES). This introduces three challenges: how to partition the states and events, communicate between simulators, and enforce local causality in each one.
We propose two promising directions for future work. First, we consider increasing the way states and events are partitioned using a hierarchical approach to how events are communicated and local causality is enforced. Second, we suggest exploring temporal partitioning through statistical sampling, which enables timing simulation to be parallelized across independent time windows. While this requires extending PDES to earlier simulation phases, functional simulation, functional warming, and detailed warming, we believe this direction can unlock significant speedups without adding local causality enforcement overhead.
Selected papers
1. The Wisconsin Wind Tunnel: virtual prototyping of parallel computers: https://dl.acm.org/doi/10.1145/166962.166979
2. dist-gem5: Distributed simulation of computer clusters: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7975287
3. SimBricks: end-to-end network system evaluation with modular simulation: https://dl.acm.org/doi/10.1145/3544216.3544253
Practical information
- General public
- Free
Contact
- edic@epfl.ch