Hybrid Network Simulation for the ATLAS Trigger and Data Acquisition (TDAQ) System

The poster shows the ongoing research in the ATLAS TDAQ group in collaboration with the University of Buenos Aires in the area of hybrid data network simulations. he Data Network and Processing Cluster filters data in real-time, achieving a rejection factor in the order of 40000x and has real-time latency constrains. The dataflow between the processing units (TPUs) and Readout System (ROS) presents a “TCP Incast”-type network pathology which TCP cannot handle it efficiently. A credits system is in place which limits rate of queries and reduces latency. This large computer network, and the complex dataflow has been modelled and simulated using a PowerDEVS, a DEVS-based simulator. The simulation has been validated and used to produce what-if scenarios in the real network. Network Simulation with Hybrid Flows: Speedups and accuracy, combined • For intensive network traffic, Discrete Event simulation models (packet-level granularity) soon becomes prohibitive: Too high computing demands. • Fluid Flow simulation models can approximate the network behaviour with aggregated flows of coarser granularity. RESEARCH TOOL: PowerDEVS [3] Computationally more efficient, but does not capture essential stochastic features (e.g. bursts). DEVS -based hybrid modeling and simulation tool • Hybrid simulation models mix fluid and discrete flows simultaneously, offering an appealing solution. Fluid “background traffic” + Discrete “probe flows”: Qualitatively good results at low computing demands