Designing Alternative Transport Methods for the Distributed Data Collection of ATLAS EventIndex Project

One of the key and challenging tasks of the ATLAS EventIndex project is to index and catalog all the produced events not only at CERN but also at hundreds of worldwide grid sites, and convey the data in real time to a central Hadoop instance at CERN. While this distributed data collection is currently operating correctly in production, there are some issues that might impose performance bottlenecks in the future, with an expected rise in the event production and reprocessing rates. In this work, we first describe the current approach based on a messaging system, which conveys the data from the sources to the central catalog, and we identify some weaknesses of this system. Then, we study a promising alternative transport method based on an object store, presenting a performance comparison with the current approach, and the architectural design changes needed to adapt the system to the next run of the ATLAS experiment at CERN.