ATLAS Distributed Computing perspectives for Run-2 Simone Campana CERN-IT/SDC o n behalf of ADC

IT-SDC : Support for Distributed Computing

ATLAS Distributed Computing perspectives for

Run-2

Simone Campana CERN-IT/SDCon behalf of ADC

IT-SDC

The Challenges of Run2

LHC operation Trigger rate 1 kHz (~400) Pile-up up above 30 (~20) 25 ns bunch spacing (~50) Centre-of-mass energy x ~2

Different detector

Constraints of ‘flat budget’ Both for hardware and for operation and development Data from Run1

8 July [email protected] – WLCG Workshop 2

IT-SDC

Where to optimize?CPU Consumption

Disk usage at T1s & T2s

SimulationCPU

ReconstructionCPU, Memory

AnalysisCPU, Disk Space

8 July [email protected] – WLCG Workshop 3

IT-SDC

Simulation

Simulation is CPU intensive

Integrated Simulation Framework (ISF) Mixing of full GEANT & fast

simulation within an event

Baseline for MC production

More events per 12h job, larger output files, less transfers/merging, less I/O

Or shorter, more granular jobs for opportunistic resources 8 July [email protected] – WLCG Workshop 4

IT-SDC

Reconstruction

Reconstruction is memory eager And requires non negligible CPU

AthenaMP default from 2014 (Almost) all production run on

multicore Analysis on single core

Optimization in code and algorithms

8 July [email protected] – WLCG Workshop 5

IT-SDC

Analysis Model

Common analysis data format: xAOD replacement of AOD & group

ntuple of any kind Readable both by Athena & ROOT

Data reduction framework AthenaMP to produce group data

sample• Centrally via Prodsys

Based on train model• one input, N outputs• from PB to TB

8 July [email protected] – WLCG Workshop 6

ChaoticOrganized (trains)

Organized (reprocessing)

IT-SDC

Computing Model: data processing Optional extension of first pass

processing from T0 to T1s

More flexible usage of resources Reduce the different between T1s and

T2s in the model Loosen the job-to-data colocation

Still one/two full reprocessing from RAW / year, but multiple AOD2AOD reprocessings/year

Derivation Framework (train model) for analysis datasets

Some T2s are equivalent to T1s in term of disk storage & CPU power

Need More Flexibility

8 July [email protected] – WLCG Workshop 7

IT-SDC

Computing Model: Run-2 data placement

Dynamic data placement and reduction of secondary (cache) replicas will continue

More data will be stored on tape beside disk

Only popular primary (pinned) dataset will be kept on disk after some time

All datasets will have a lifetime (from both disk and tape)

Access to data on tape still “organized” and not “chaotic”

8 July [email protected] – WLCG Workshop 8

T0/T1 DATADISK

pinned

pinned

cache

T2 DATADISK

IT-SDC

Preparation for Run-2

Alessandro will now present the work being done in preparation for Run-2

We will have a 3 days Jamboree (Dec 3-5) at CERN focused on sites related aspectsPlaceholder agenda at

https://indico.cern.ch/event/276502/

8 July [email protected] – WLCG Workshop 9