Speaker
Dr
markus schulz
(CERN)
Description
The T0 at CERN operates large storage and computing farms for the LHC community. For economic reasons the hardware of the disk servers is, with respect to CPU and memory, virtually identical to the one used in the batch nodes. Monitoring data showed that these nodes are not running anywhere close to their computational limit. Proof of concept tests have been conducted by Andrey Kiryanov showing that more than 80% of the node capacity can be used for computational tasks while creating no detrimental effect on the peak I/O rates. These results have been show at HEPIX 20017.
Our team at CERN is expanding the concept, in the BEER (Batch on EOS Extra Resources ) project, to be ready to be integrated into the production service. The approach to partition the resources, the strategy for configuration management and results with production workloads will be shown.
Primary author
Dr
markus schulz
(CERN)
Co-authors
Mr
Andrey Kiryanov
(PNPI)
Ben Jones
(CERN)
David Smith
(CERN)
Mr
Luca Mascetti
(CERN)
Dr
Massimo Lamanna
(CERN)
