Data Synchronization Method Between FLPs and EPNs

In the year 2018, the ALICE detectors will be upgraded and the amount of data that will be produced from the detectors will be higher. The data throughput will be approximately 1 TB per second. For this reason, the whole ALICE's system has to be improved. The constraint of the new ALICE experiments is the system should be able to handle a large amount of particles collision events that will be produced from detectors. The data acquisition comprises two computer clusters – First Level Processors (FLP) and Event Processing Nodes (EPN). The two clusters serve different purposes. The FLP cluster is responsible to readout the detector and to reduce by a factor 5x the data before event reconstruction because the data have to be streamed to a permanent storage. Then, the data will be sent through the network to EPNs for event reconstruction and further processing. In the new system, a scheduler between the two clusters is required in order to efficiently propagate jobs and data from FLPs to EPNs. Each FLP node receives collision events which are grouped in time frames spanning 0.1 second. EPN also has to forward the events to an assigned EPN node immediately. Scheduling process and data synchronization must be fast because the delay may cause severe bottleneck and provoke side effects like increasing the buffer space needed on FLPs. In this work, we have adopted Zookeeper, a data synchronization tool, to distribute events schedule (that is created by scheduler) to FLPs.