CERN Openlab: my transition from science to business CERN Openlab alumni talk CERN Openlab: my transition from science to business Maaike Limper 21/09/2017

Outline (why I’m not at work right now…) Before Openlab My CERN Openlab fellow My career since Openlab CERN Openlab: lessons learnt

Before Openlab One of the many physicists at CERN… Master in Particle Physics at University of Amsterdam 1999-2004 First time at CERN as Summer Student in 2003! PhD in Particle Physics 2004-2009 Building the ATLAS SemiConductor Tracker (SCT) Testing and installation of SCT at CERN Software development for track and vertex reconstruction Post-doctoral researcher for University of Iowa based at CERN First physics analysis with LHC data in ATLAS Prompt Reconstruction Coordinator for ATLAS “Co-discovered the Higgs-boson” along with others thousands of physicists…

My CERN Openlab fellow SQL-based approached to physics analysis: How to discover the Higgs boson in an Oracle database SQL-based approached to physics analysis: Databases allow us to do much more than store calibration constants and conditions Complex SQL with analytics-functions or precompiled procedures allow to do event filtering and calculations directly in the database SQL-analysis of basic physics data potentially faster than ROOT-based analysis Physics data has many attributes per row, making it more suitable to be kept in column-storage and partitioned in files across the grid for easy parallel processing

My CERN Openlab fellow A + B + C = D D What I actually discovered ROOT/PROOF is optimized for physics: Hard to improve using database analysis! Though users don’t always make optimum use of its features But great test-case for complex DB-analysis: Sequential scan vs index look-up Column vs row-based storage Love-hate relationship with SQL optimizer Use of partitions/distributed processing Hadoop vs PROOF vs Oracle RAC: Any distributed data processing ultimately comes down to the same thing, optimize your partitioning to push down your joins! A6 B6 C6 D6 E6 A6 B6 C6 D6 E6 A + B + C = D A1 B1 C1 A2 B2 C2 D A3 B3 C3 A4 B4 C4

My career since Openlab The final frontier of connectivity SITA OnAir: Business data analyst Inmarsat: Head of Aviation Service Performance

Aviation Service Performance Data-driven approach to performance monitoring Inmarsat Aviation has a dedicated team for Service Performance Our mission: ‘to turn data into information’ Data collected from all IFC systems is analysed using advanced custom data- processing solution and dedicated tools provide monitoring, customer analytics and enable data-sharing with our customer

The all devouring data-engine “Any sufficiently advanced technology is indistinguishable from magic” Automatic processes extract, consolidate and aggregate data from every possible data source Correlating data from different sources allow unique opportunities to study performance dependencies between our systems … pre-processed data is presented through a variety of internal and external reporting tools. On-Board portal-logs Portal content usage User-agent: Device/OS System events The all-devouring data-engine Allows Inmarsat support teams to proactively troubleshoot problems Inmarsat Internal Performance Monitoring ISP server-logs ISP portal usage User-agent: Device/OS Inmarsat network data Terminal status Terminal statistics Terminal messages ISP sessions data Payment method Session type User id Unique device ID Remote Maintenance View Allows external maintenance teams to monitor IFC status in real-time Airline: movement data OOOI timestamps Customer Monthly Reports SNMP Polling Discrete system status ARINC MIBs Flexible solutions to share pre-processed data with our Customer Ping utility Customer Analytics Portal Latency Outage timestamps Netflow Bytes/packets per IP Unique device IP Customer API Sandvine DPI Usage by category Whitelisted sites usage

CERN Openlab: lessons learnt What working at CERN has taught me No Higgs boson without IT Multi-culturalism Lateral thinking DIY Don’t trust technology Love, trust & respect your data