Voir en


CERN data storage gets ready for Run 3

Extensive consolidation of the data storage infrastructure culminated in 2020 when the new CERN Tape Archive software (CTA) entered production

On 29 June, the CERN Tape Archive (CTA) officially entered production after 83 petabytes of ATLAS data initially stored in CASTOR were successfully migrated to CTA

On 29 June, the CERN Tape Archive (CTA) officially entered production after 83 petabytes of ATLAS data initially stored in CASTOR were successfully migrated to CTA

The output, or product, of CERN’s experimental programme is its data: vast amounts of data are produced by the detectors in the LHC experiments and elsewhere in the accelerator complex. These data are curated by the IT department’s Storage Group for reconstruction and analysis by physicists across the world, using the Worldwide LHC Computing Grid (WLCG).

The custodial copy of all of CERN’s physics data – amounting to about 340 petabytes –  is stored on magnetic tapes at the CERN Data Centre, also called the WLCG “Tier-0”.

During Runs 1 and 2 of the LHC, the software system used to manage the archival storage of physics data was CASTOR, the CERN Advanced Storage manager, which was conceived as a system to manage both disk and tape storage. Over the last ten years, requirements have evolved and a new disk system – EOS – was developed for online storage and data analysis. As EOS does not itself provide offline storage and data archival, a new project, the CERN Tape Archive (CTA), was conceived as the tape back-end to EOS. CTA is an evolution of the CASTOR tape system that removes the necessity to maintain a second disk-management system.

In early 2020, the CTA team started commissioning tests with the ATLAS experiment as part of a reprocessing campaign on all of their experimental data from Run 2. During this exercise, raw data are recalled from Tier-0 tapes or from the WLCG Tier-1 data centres (which are national data centres that curate a proportional share of the LHC data) in order to be “reconstructed” into meaningful physics data that can be analysed. CTA replaced CASTOR for the tape-recall part of the exercise and had the lowest error rate of all sites, demonstrating performance and reliability for large-volume transfers.

Production for CTA was delayed by the disruption caused by COVID-19 and the subsequent move to teleworking, compelling the CTA team and ATLAS data-management team to adjust to new ways of communicating and scheduling tests. Nonetheless, the integration work and final commissioning tests were marked by a determined spirit of cooperation, which eventually allowed the migration of ATLAS’s data to take place during the last two weeks of June. This involved transferring the metadata (as opposed to the physical data, which does not move) of 86 million files – the entirety of ATLAS’s physics output – from CASTOR to CTA. After the migration of ATLAS data was completed, CTA entered into operation on 29 June.

All other CERN experiments, from the large LHC detectors to the smallest experiments, will migrate to CTA from October onwards, with ALICE the first to follow. This will allow CASTOR disk servers to be recovered and repurposed.

Besides the introduction of CTA, CERN is also preparing for Run 3 by installing a new tape library in the Data Centre and by upgrading FTS, its File Transfer Service. FTS, which distributes the majority of LHC data across the WLCG infrastructure, benefitted from several significant performance improvements. It is now also supporting CTA and is used by more than 25 experiments at CERN and in other data-intensive sciences.

Michael Davis

As part of the data storage consolidation, a new tape library was installed in August 2020 in the CERN Data Centre. (Video: CERN)