artdaq: An Event-Building, Filtering, and Processing Framework

Several current and proposed experiments at the Fermi National Accelerator Laboratory, Batavia, IL, USA, have novel data acquisition needs. These include 1) continuous digitization, using commercial high-speed digitizers, of signals from the detectors, 2) the transfer of all of the digitized waveform data to commercial off-the-shelf (COTS) processors, 3) the filtering or compression of the waveform data, or both, and 4) the writing of the resultant data to disk for later, more complete, analysis. To address these needs, members of the Accelerator and Detector Simulation and Support Department within the Scientific Computing Division at Fermilab are using parallel processing technologies in the development of artdaq, a generic data acquisition toolkit. The artdaq toolkit uses Message Passing Interface (MPI) and art, an established event-processing framework shared by new experiments at Fermilab. In an artdaq program, the digitized data are transferred into computing nodes using commodity Peripheral Component Interconnect Express (PCIe) cards, and event fragments are transferred between distributed processes using MPI and assembled into complete events. These events are then processed by a configurable selection of user-specified algorithms, commonly including filtering and compression algorithms, using the art event-processing framework. This paper describes the architecture and implementation of the first phase of the artdaq toolkit and shows early performance results with configurations that match upcoming experiments both at Fermilab and elsewhere.