High-performance storage and dataflow solutions for the data acquisition system of particle physics experiments
by
The data acquisition system of particle physics experiments is a mission-critical component responsible for the experiments' success. The next generation of large-scale particle physics experiments will have millions of sensors producing a large amount of data at high rates. The DUNE and Phase-II ATLAS experiments are expected to start data-taking in the late 2020s. Data rates from both detectors will reach orders of terabytes per second, posing a significant challenge to the data acquisition chain and, especially, to the storage and dataflow system, which will need to be designed accordingly. Therefore, it becomes essential to investigate methods to collect, store and transport the data efficiently. This seminar presents the work done on the performance characterization of novel event reconstruction algorithms as well as storage technologies and dataflow methods that are suitable for implementing the storage systems of both the ATLAS and DUNE experiments.
Persistent memory devices and novel solid-state devices have been investigated as potential solutions to implement the local storage system of the DUNE experiment. This was achieved using both synthetic benchmarks with an emulated workload and integration testing for each storage technology. In addition, a distributed high-throughput key-value store (DAQDB) was also extensively tested as a solution for the large storage buffer of the ATLAS experiment. Dataflow methods that combine local storage management solutions were also investigated as a possible means to achieve the target rates needed by the detector. Finally, an extensive study on the evolution of the storage system with discrete event simulations was also performed to understand the advantages and limitations of different data acquisition architectures.
This experimental research is completed by the investigation of a novel algorithm (SparseNet) designed to classify track and shower energy deposits across a liquid argon detector. The SparseNet algorithm was tested with both Monte Carlo data and beam data at the ProtoDUNE experiment. Promising results show that the algorithm is capable of discriminating tracks and shower objects with accuracies of over 90%.
Giovanna Lehmann Miotto and Burkhard Schmidt (EP-DT)