Ordonnancement et routage intelligent pour les réseaux d’acquisition de données

Large scientific instruments are typically composed of tens of thousands of sensors and a data acquisition (DAQ) system whose objective is to collect fragments of data from each sensor and to assemble all the fragments corresponding to each collision event into a single data set. This process, called Event Building, involves an all-to-all collective exchange between one or more network computers. However, the traffic associated with Event Building tends to create congestion in the event of a failure due to the nature of the all-to-all exchange, which requires using almost all available network bandwidth. Congestion can severely degrade the performance of the DAQ system and lead to the loss of highly valuable experimental data. This thesis presents approaches to minimizing network congestion in the event of failures, using the DAQ network of the LHCb (Large Hadron Collider beauty) detector as a case study.