1 of 10

ATLAS status report

​

L Rinaldi, A De Salvo

​

INFN Tier-1 CdG, September 28th 2018

2 of 10

ATLAS data taking

LHC Data taking just restarted. Integrated luminosity already above 2017 (53.9 vs 50.2)

3 of 10

Global Processing activities

Average 320K jobs/day, peak 420K jobs/day

4 of 10

INFN-T1 processing activities (cumulative)

Average 25K jobs/day, peak 77K jobs/day

Mainly simulation activity (58%) and user analysis (23%)

5 of 10

Harvester and Panda unified queue (UCORE)

Analysis

Production

Evgen

Simul

Reproc

Group production

  • ATLAS wants to decide how many resources to assign to each share
  • Impossible to control through independent SCORE and MCORE queues at each site
  • ⇒ Unified queues with a centrally controlled SCORE and MCORE ratio depending on share status
  • For unified queues to work, sites need to:
    • Not have static SCORE-MCORE partitions and all machines be able to run S and MCORE
    • Allow submission of the different types to the same batch queue and follow the CPU and memory requirements in the JDL

...

SCORE

MCORE

6 of 10

Usage of UCORE (-->MCORE) at INFN-T1

CINECA

CNAF

Expected 1-core/8-core ratio: 10 % (1-core used for MC simulation only)

Now ratio is ok (before 9/15 mcore nodes swithed off for intervention),

BUT in the present system resource allocation is faster for 1-core (8-core filling could be slower)

HTCondor would behave better…, still in preliminary testing phase

7 of 10

Tape Carousel (greater use of tape-based workflow)

8 of 10

9 of 10

10 of 10

the StoRMy summer

Frequent crashes of the ATLAS StoRM FE (details in Lucia’s slides)

mainly due to high number of SRM requests (mainly by WNs)

​

Several fixes on StoRM (thanks to many), now tests show that the system can manage high load (compatible with production load)