di-EOS - 'distributed EOS':Initial experience with split-site persistency in a production service

摘要

In order to accommodate growing demand for storage and computing capacity from the LHC experiments, in 2012 CERN tendered for a remote computer centre.The potential negative performance implications of the geographical distance (aka network latency) within the same "site" and storage service on physics computing have been investigated.Overall impact should be acceptable, but for some access patterns might be significant.Recent EOS changes should help to mitigate the effects, but experiments may need to adjust their job parameters.