A FELIX Based Architecture for DUNE Giovanna Lehmann

+ A FELIX Based Architecture for DUNE Giovanna Lehmann Miotto

+ 2 Candidate Design for DUNE Detector FE FE … FE ~1500 links; 1. 5 TB/s (per super module) 4 super modules in DUNE FELIX Large distributed temporary store Get data slices DCS Pick data for storage Event selection farm (e. g. +GPUs) Regions of Interest FE Data selected for permanent storage Trigger Save those data To permanent storage

+ 3 Candidate Design for DUNE Detector FE FE … FE ~1500 x 10 Gbps links; 1. 5 TB/s (per super module) 4 super modules in DUNE FELIX host FELIX Large distributed temporary store FE DCS FELIX ++ sw Switched network To permanent storage Event selection farm (e. g. +GPUs) Trigger Data selected for permanent storage

Sequence diagram – Data Flow Detector data FELIX Temp. Store Analyzer Trigger Archiver data net next data Ro. I list Accept list get data

+ What does what n FELIX host n n n Analyzers n n n Check data and extract which ones have signals Provide Ro. I lists to Trigger n n Receives data from electronics Organizes temp store Responds to data requests Distributes timing Exchanges commands/data between DCS and electronics Decides when/which Ro. Is are to be taken and forwards those to Archiver n n Gathers data from FELIX hosts Formats and compresses data for permanent storage

+ Questions n Which data are needed to identify regions (and times) of interest? n long time windows of individual wires? n Matrix of wires in time slices? n How can photon detectors help in identifying interesting activity? n These questions are essential for the implementation of the data organization and exchange (and final costing), but do not affect this design which is flexible to evolution in the understanding of the detector and the Physics we want to study

+ Costing n Costing only makes sense if a coherent model is adopted n For “fun” estimate using model from LHC Phase II… n n Including readout, processing, selection, storage, network 150 APA: 2. 7 – 7 M$ dominated by processing model n 10 Pcs per APA (10 ms processing for 1 ms APA) n 1 PC per APA with 5 GPUs (5 ms processing for 1 ms APA)