Workshop: PMBS26: The 17th International Workshop on Performance Modeling, Benchmarking, and Simulation of High Performance Computer Systems
Authors: Kuan-Chieh Hsu, Raees Ahmad Khan, Sairam Sri Vatsavai, Tasnuva Chowdhury, Ozgur Kilic, and Sankha Dutta (Brookhaven National Laboratory); Shengyu Feng (CMU); David Park (Brookhaven National Laboratory); Fatih Furkan Akman (UMASS); Rui Xue and John Rembrandt (Remy) Steele (CERN); Yihui Ren (Brookhaven National Laboratory); Paul Nilsson (CERN); Tatiana Korchuganova and Joseph Boudreau (University of Pittsburgh); Scott Klasky (Oak Ridge National Laboratory); Tadashi Maeno and Verena Ingrid Martinez Outschoorn (CERN); Norbert Podhorszki and Frederic Suter (Oak Ridge National Laboratory); Wei Yang (SLAC National Accelerator Laboratory); Yiming Yang (Carnegie Mellon University); Shinjae Yoo (Brookhaven National Laboratory); Kenny Yo (SLAC National Accelerator Laboratory); and Alexei Klimentov and Adolfy Hoisie (Brookhaven National Laboratory)
Abstract: Modern computing-grid simulators often emphasize system configuration and job allocation while simplifying data-management behavior, despite its stateful impact on performance. In data-centric infrastructures such as the Worldwide LHC Computing Grid (WLCG), replica placement and transfer policies influence network contention, staging delay, storage utilization, and resiliency, yet production-scale evaluation is difficult. We present DMSim, a configurable data-management simulation framework for scientific computing grids. DMSim extends CGSim with a lightweight global file-system manager and SimGrid-based communication modeling to track replicas, simulate contention-aware transfers, and support configurable policies. It models reactive transfers triggered by job inputs, proactive background replication and rebalancing, and explicit drop-in transfers for irregular behaviors. We validate DMSim using historical PanDA and Rucio records, reconstructing initial file locations and comparing simulated transfer-queue times with production observations. Simulated staging times fall above bandwidth-based lower bounds and below observed job queuing times, supporting reproducible policy evaluation at scale in controlled experimental settings.
Back to PMBS26: The 17th International Workshop on Performance Modeling, Benchmarking, and Simulation of High Performance Computer Systems Archive Listing
Back to Full Workshop Archive Listing