CERN Accelerating science

ATLAS Slides
Report number	ATL-SOFT-SLIDE-2019-188
Title	ATLAS job submission system for Salomon HPC based on ARC-CE
Author(s)	Svatos, Michal (Institute of Physics of the Czech Academy of Sciences) ; Chudoba, Jiri (Institute of Physics of the Czech Academy of Sciences) ; Vokac, Petr (Czech Technical University in Prague)
Corporate author(s)	The ATLAS collaboration
Collaboration	ATLAS Collaboration
Submitted to	High Performance Computing in Science and Engineering, Karolinka, Czech Republic, 20 - 23 May 2019
Submitted by	[email protected] on 14 May 2019
Subject category	Particle Physics - Experiment
Accelerator/Facility, Experiment	CERN LHC ; ATLAS
Free keywords	HPC, computing
Abstract	The ATLAS Experiment at CERN is using HPCs opportunistically to extend its computing capacity for years. To the Salomon HPC, ATLAS jobs come via ARC-CE machines located in the computing center of the Institute of Physics of the Czech Academy of Sciences. The ARC-CE serves as an interface between job management systems of the ATLAS and the HPC. Commands of the PBSpro batch system are submitted via ssh. Scripts and input files are shared between the ARC-CE and shared file system located at the HPC via sshfs. There are several aspects of interaction between ARC-CE machines and Salomon's batch system which are important for performance of the whole system. First, the allowed amount of requests to PBSpro is limited and the ARC-CE needed to be adapted to this fact. Second, the sshfs connection speed seems to be a limiting factor for job turnaround. Some possibilities of sshfs parameters tuning were investigated. Moreover, monitoring allows quick detection of issues and therefore helps the performance of the system. The ARC-CE based job submission system has adapted to conditions of the Salomon HPC and utilizes successfully its resources.