New Persistent Back-End for the ATLAS online information service

The Trigger and Data Acquisition (TDAQ) and detector systems of the ATLAS experiment deploy more than 3000 computers, running more than 15000 concurrent processes, to perform the selection, recording and monitoring of the proton collisions data in ATLAS. Most of these processes produce and share operational monitoring data used for inter-process communication and analysis of the systems. Few of these data are archived by dedicated applications into conditions and histogram databases. The rest of the data remained transient and lost at the end of a data taking session. To save these data for later, offline analysis of the quality of data taking and to help investigating the behavior of the system by experts, the first prototype of a new Persistent Back-End for the Atlas Information System of TDAQ (P-BEAST) was developed and deployed in the second half of 2012. The modern, distributed, and Java-based Cassandra database has been used as the storage technology and the CERN EOS for long-term storage. This paper provides details of the architecture and the experience with such a system during the last months of the first LHC Run. It explains why that very promising prototype has failed and how it was reimplemented using the Google C++ protobuf technology. It finally presents the new architecture and details of the service, which will be used during second LHC Run.