ALICE O 2 | 2015 | Pierre Vande Vyvre O 2 Project Pierre VANDE VYVRE

ALICE O 2 | 2015 | Pierre Vande Vyvre Online and offline computing (Runs 1 and 2) 2 Compressed data Information Knowledge Offline computing  Extract the particle properties from raw data.  Store the results for analysis.  Produce knowledge by analyzing the information Offline computing  Extract the particle properties from raw data.  Store the results for analysis.  Produce knowledge by analyzing the information Raw data OnlineOffline Triggering and online computing  Fast inspection of all interactions.  Reduce rate by event selection.  Reduce data volume by compressing the selected events.  Store the compressed data. Triggering and online computing  Fast inspection of all interactions.  Reduce rate by event selection.  Reduce data volume by compressing the selected events.  Store the compressed data. O (kHz) O (100 Hz)

ALICE O 2 | 2015 | Pierre Vande Vyvre Physics programme and data taking scenarios 3 YearSystem√s NN L int N collisions (TeV)(pb -1 )(nb -1 ) 2020 pp · Pb-Pb · pp · Pb-Pb · pp · pp5.564 · pp · Pb-Pb · pp · Pb-Pb · p-Pb pp · Pb-Pb · 10 10

ALICE O 2 | 2015 | Pierre Vande Vyvre Requirements after LS2 1.After LS2, LHC will deliver min bias Pb-Pb at 50 kHz –100 x more data than today 2.Physics topics addressed by ALICE upgrade –Very small signal-to-noise ratio and large background –Triggering techniques very inefficient if not impossible –Needs large statistics 3.Support for continuous read-out (TPC) –Detector read-out triggered or continuous by Time Frames of ~ 20ms 4.Running scenarios –Goal: 13 nb −1 for Pb–Pb collisions at 5.5 TeV (minimum bias)  Too much data to be stored  Compress data intelligently by processing it online 4

ALICE O 2 | 2015 | Pierre Vande Vyvre O 2 Project Requirements for Run 3 (2020) 5 -Handle >1 TByte/s detector input -Online reconstruction to reduce data volume -Common hw and sw system developed by the DAQ, HLT, Offline teams -Handle >1 TByte/s detector input -Online reconstruction to reduce data volume -Common hw and sw system developed by the DAQ, HLT, Offline teams O2O2

ALICE O 2 | 2015 | Pierre Vande Vyvre Paradigm shift for Run 3 6 Compressed data Information Knowledge ALICE computing  Optimize the distributed analysis.  Produce knowledge by analyzing the information ALICE computing  Optimize the distributed analysis.  Produce knowledge by analyzing the information Raw data 1 common computing environment ALICE computing  Read-out all interactions.  Reduce data volume by extracting the particle properties and compressing the events.  Store only processed data. ALICE computing  Read-out all interactions.  Reduce data volume by extracting the particle properties and compressing the events.  Store only processed data. 50 kHz

ALICE O 2 | 2015 | Pierre Vande Vyvre Design strategy Iterative process: design, benchmark, model, prototype 7 Design Model Prototype Technology benchmarks

ALICE O 2 | 2015 | Pierre Vande Vyvre Functional Requirements 8 -Functional requirements of the O2 system -Data fully compressed before data storage -Reconstruction with calibrations of better quality -Grid capacity will evolve much slower than the ALICE data volume -Data archival of reconstructed events of the current year to keep Grid networking and data storage within ALICE quota -Needs for local data storage higher than originally anticipated Asynchronous and refined calibration, reconstruction Event extraction Quality control Compressed Sub-Time Frames Continuous and triggered streams of raw data Data aggregation Synchronous global reconstruction, calibration and data volume reduction Quality control Data storage and archival Compressed Time Frames Reconstructed events Compressed Time Frames Readout, split into Sub-Time frames, and aggregation Local pattern recognition and calibration Local data compression Quality control Detectors electronics 1.1 TB/s 90 GB/s

ALICE O 2 | 2015 | Pierre Vande Vyvre Hardware Architecture 9 Detectors 8000 Read-out Links 250 FLPs First Level Processors 1500 EPNs Event Processing Nodes Input: 250 ports Output : 1500 ports 1.2 TB/s Switching Network 500 GB/s 90 GB/s Storage 34 Storage Servers Storage Network Input: 1500 ports Output : 34 ports

ALICE O 2 | 2015 | Pierre Vande Vyvre Data flow & processing (1) O 2 /T0/T1 FLPs EPNs FLPs Detectors electronics Detector data samples interleaved with synchronized heartbeat triggers Buffering TPC Detector reconstruction e.g. track finding Raw data input Data Reduction 0 e.g. clustering Local processing Time Frame building Frame dispatch Global processing TRD Sub-Time Frames Partially compressed sub-Time Frames Full Time Frames Compressed Time Frames Data Reduction 1 … Trigger and clock Calibration 0 on local data, ie. partial detector Calibration 1 on full detectors e.g. space charge distortion T0/T1 Archive Time slicing O(100) O(1000) Local aggregation Storage Quality Control Sub-Time Frames Time Frames Compressed Time Frames AOD QC Tagging Synchronous QC data ITS … CTF AOD Storage Load balancing & dataflow regulation

ALICE O 2 | 2015 | Pierre Vande Vyvre Data flow & processing (2) O 2 /T0/T1 Reconstruction passes and event extraction Compressed Time Frames T0/T1 Archive Analysis Storage T2 Simulation CTF Simulation AODO(10) Condition & Calibration Database Quality Control Sub-Time Frames Time Frames Compressed Time Frames AOD CCDB Objects QC Asynchronous QC data CTF AOD Event Summary Data Analysis Object Data Analysis Facilities Storage Histograms, trees O(1) Analysis AOD Storage Reconstruction Event building AOD extraction Compressed Time Frames O 2 /T0/T1 O(1) Event extraction Tagging Global reconstruction QC ESD, AOD AOD extraction Calibration 2 ESD, AOD

ALICE O 2 | 2015 | Pierre Vande Vyvre Calibration and reconstruction flow 12 EPN: synchronous asynchronousAll FLPs Raw data Local Processing E.g. Clusterization Calibration Detector Reconstruction CTF AOD Step 1Step 2Step 3Step 4 Matching procedures Final calibration and 2nd matching Final matching, PID, Event extraction… Step 0 FPGA GPU CPU

ALICE O 2 | 2015 | Pierre Vande Vyvre Quality control 13 Data storage Dataflow Quality Control System Data Collection Generation of Monitoring Objects Quality Assessment Storage Visualization Raw data Reconstructed data Condition & Calibration API Access

ALICE O 2 | 2015 | Pierre Vande Vyvre Control, configuration and monitoring 14 Control, Configuration and Monitoring LHCTrigger Status/ Monitoring data Status DCSGrid Commands/ Configuration data Status Commands/ Configuration data Status/ Monitoring data Grid Jobs Status Commands

ALICE O 2 | 2015 | Pierre Vande Vyvre Technology: Input/Output with PCI Express 15 PCIe Gen2 -Up to 3.4 GB/s -Device independent from each other PCIe Gen3 -Up to 6 GB/s -FPGA-based DMA controller

ALICE O 2 | 2015 | Pierre Vande Vyvre Technology: hw acceleration with FPGA and GPU 16 FPGA -Acceleration for TPC cluster finder versus a standard CPU -25 times faster than the software implementation GPU -TPC Track Finder based on the Cellular Automaton principle to construct track seeds. -It is implemented for OpenMP (CPU), CUDA (Nvidia GPU), and OpenCL (AMD GPU). -1 GPU replaces 30 CPU cores and uses 3 for I/O

ALICE O 2 | 2015 | Pierre Vande Vyvre Facility design 17

ALICE O 2 | 2015 | Pierre Vande Vyvre Facility design Network layout 2 18 FLP … EPN … 10 Gb/s FLP … EPN … 10 Gb/s … Leaf Switch 1 Leaf Switch 7 Cluster 1 out of 4 40/56 Gb/s

ALICE O 2 | 2015 | Pierre Vande Vyvre Facility design Network layout 3 19 FLP EPN FLP EPN /56 Gb/s SEPN 1 10 Gb/s 1 EPN Gb/s 50 FLP X 40/56 Gb/s 50 2 X 40/56 Gb/s

ALICE O 2 | 2015 | Pierre Vande Vyvre Simulation: Scalability 20 Network layout 2 -Based on Eth. Technology -Scales up to 90 kHz Network layout 2 -Based on IB Technology -Scales up to 140 kHz

ALICE O 2 | 2015 | Pierre Vande Vyvre Software framework ALFA –Prototype developed in common by experiments at FAIR and ALICE –Base on the ZeroMQ package –Data transport Performance (see plot) –Dynamic Deployment System ALFA devices (FLPs, EPNs and one sampler). Propagated about 77 millions key-value < 200 seconds 21

ALICE O 2 | 2015 | Pierre Vande Vyvre O 2 Project PLs: P. Buncic, V. Lindenstruth, P. Vande Vyvre PLD:.M. Krzewicki, T. Kollegger Computing Working Group(CWG)Chair 1.ArchitectureS. Chapeland 2.Tools & ProceduresA. Telesca 3.DataflowT. Breitner 4.Data ModelA. Gheata 5.Computing PlatformsM. Kretz 6.CalibrationC. Zampolli 7.Reconstruction R. Shahoyan 8.Physics SimulationA. Morsch 9.QA, DQM, VisualizationB. von Haller 10.Control, Configuration, MonitoringV. Chibante 11.Software LifecycleA. Grigoras 12.HardwareH. Engel 13.Software frameworkP. Hristov Editorial Committee L. Betev, P. Buncic, S. Chapeland, F. Cliff, P. Hristov, T. Kollegger, M. Krzewicki, K. Read, J. Thaeder, B. von Haller, P. Vande Vyvre Physics requirement chapter: Andrea Dainese 22 Project Organization O 2 CWGs

ALICE O 2 | 2015 | Pierre Vande Vyvre O 2 Project Institutes 23

ALICE O 2 | 2015 | Pierre Vande Vyvre Schedule (1) 24

ALICE O 2 | 2015 | Pierre Vande Vyvre Schedule (2) 25

ALICE O 2 | 2015 | Pierre Vande Vyvre ALICE O2 Technical Design Report 26