Performance Impact of Aggregator Configuration on Parallel HDF5 Writes in the MULTIPAC Data Acquisition Software

This report investigates the impact of ROMIO aggregator configuration on the performance of parallel HDF5 writing in the MULTIPAC data acquisition software at CERN. Synthetic benchmarks were conducted using MPI-IO and h5py to emulate the high-throughput data flows encountered in real experiments. Performance was measured across a range of aggregator configurations, rank counts, and data sizes, with a focus on realistic DAQ scenarios. While the collected data revealed clear trends in the small and large data regimes, the results in the intermediate regime (k∼107) were inconclusive, and no single optimal aggregator configuration could be identified based on the available measurements. To address these ambiguities, a detailed plan for future evaluation is presented, including systematic exploration of advanced ROMIO tuning hints, fine-grained I/O tracing, and the implementation of an automated aggregator selection algorithm. These steps are expected to yield the information necessary to fully understand ROMIO’s behavior in the MULTIPAC context and to identify the optimal configuration for future high-rate data taking.