MEP V2, a new event building protocol for the LHCb experiment

During the long shutdown of the LHC in 2018, the LHCb collaboration plans to significantly upgrade the LHCb detector. This upgrade foresees a trigger-less read-out. The detector data is sent directly into a read-out network at the LHC rate of 40 MHz. The data is then reconstructed and processed by a computing farm. The expected data rate will be in the order of 40 Tbit/s, making it the largest read-out network in HEP to date. The data flow in this network is rather atypical for HPC networks. The network is divided into O(500) source and O(5000) target nodes. All data that constitutes an event is initially distributed as fragments among the sources. These fragments have to be sent to one particular target node for assembly and further processing. One of the biggest challenges is to control the data flow through the network to ensure that the output port to a single target node is not being overloaded, because the fragments of one event all become available at exactly the same time. We will present the lessons learned from the MEP v1 protocol currently in use and a first draft of the MEP v2 protocol, which will be used in the upgraded network. Since we do not yet know with certainty which technologies will be available for such a network in the future, the new version of MEP will adhere more strictly to the OSI model. The protocol will be constrained to one layer only and will be a more generic protocol for aggregating data fragments from many sources on one particular target.