1. Workflows and Data Management

2. Workflow and DM Run3 and after: conditions m LHCb major upgrade is for Run3 (2020 horizon)! o Luminosity x 5 (2 10 33 ) P Lumi levelling: higher pileup, from 1.1 to 5.5 o Trigger rate… x 5 (at least, dominated by charm physics) o RAW data size x 2 (pileup) o Online reconstruction = offline reconstruction P Allows direct analysis from online data (TURBO stream) d TURBO data format is directly analysis data (no RAW!) o Output from DAQ: P Any linear combination from TURBO data to full reconstruction output (reco + RAW) P Use year “n” data to tune TURBO for year “n+1” ! o Throughput between 6 and 10 GB/s (GPDs of today) o Trigger (SW only) == offline selection P Stripping and streaming are no longer effective (all events are for physics!) 2

3. Workflow and DM Online Calibration & Alignment m Novel concept of detector alignment & calibration done in between the two stages of HLT processing o Successfully exercised in 2015 o Part of online reconstructed events immediately available for user analysis o Enabling HLT2 processing for better signal yield o Same constants used for offline processing o Concept will be further exploited in Run 3 2 Feb '163

4. Workflow and DM Trains and indices m Using event indices for analysis o Replace “stripping + streaming” with “selection + indexation” P Because stripping retention will be high (more selective trigger) o Event set query to central (or local) index P Download a local event collection (i.e. direct access addresses) o Random access to local or remote data P Using a local replica catalog (Gaudi Federation) m R&D can start now (2016/17) for: o Setting up train analyses P framework similar to stripping o Data indexing P Select technology (central vs distributed, DB vs files) P Index content to be defined P Event set queries to be defined for jobs o Optimizing random access through ROOT m Not to forget for analysis data access: o Network bandwidth is not all: disk spindles is equally important 4

5. Workflow and DM Analysis job using event index 5 Central Event Index Application Local Event Catalog Local Replica Catalog Replica Catalog Remote storage Local storage Event set query Job One size does NOT fit all: we will have different event formats for different analyses (microDST, DST, sth in between?)