Log data traffic characterization for packet loss estimation in ALICE O2 system

At CERN, the ALICE O
²
computing system has nodes, called First Level Processors (FLPs), which collect particle interaction data from the ALICE detectors and carry out local processing. Log data generated by tasks running on FLPs are sent over a network to the ELK stack server to monitor and identify anomalies in FLPs. A large amount of the log-data traffic from FLPs over this network could lead to packet losses. In this paper, we create FLPs in a testbed environment to characterize the log-data traffic generated by tasks in FLPs and fit these to time-series models and probability distributions assuming independent interarrival times. The fitted models are then used to study the end-to-end packet loss with input traffic from a large number of FLPs in a network of switches. The simulation results will help to guide the number of FLPs and traffic intensity that the network can sustain for different kinds of tasks running on FLPs.