Abstract
The increases of data size and in the geographical distribution of analysis intensify the demand on the I/O subsystem. Over the last year, we greatly improved the I/O throughput, in some case by several factors, when reading ROOT files. ROOT'S improved techniques include improving the pre-existing prefetching, the automatic flushing of data buffers at regular intervals and streaming objects member-wise. These advances reduce the number of transactions with the local disk or the network. We worked in close collaboration with the Large Hadron Collider (LHC) experiments to optimize the I/O access to their use cases and to help adapt their framework to take full advantage of these advances. This presentation will describe in details these improvements and how users can benefit from them.
Export citation and abstract BibTeX RIS