Main Image

PoS(ICHEP2016)191

LHCb distributed computing in Run II and its evolution towards Run III

A. Falabella, on behalf of the LHCb collaboration

in 38th International Conference on High Energy Physics

Contribution: pdf

Abstract

This contribution reports on the experience of the LHCb computing team during LHC Run 2
and its preparation for Run 3. Furthermore a brief introduction on LHCbDIRAC, i.e. the tool
to interface to the experiment distributed computing resources for its data processing and data
management operations, is given. Run 2, which started in 2015, has already seen several changes
in the data processing workflows of the experiment. Most notably the ability to align and calibrate
the detector between two different stages of the data processing in the high level trigger farm,
eliminating the need for a second pass processing of the data offline. In addition a fraction of
the data is immediately reconstructed to its final physics format in the high level trigger and
only this format is exported from the experiment site to the physics analysis. This concept have
successfully been tested and will continue to be used for the rest of Run 2. Furthermore the
distributed data processing has been improved with new concepts and technologies as well as
adaptations to the computing model. In Run 3 the experiment will see a further increase of
instantaneous luminosity and pileup leading to even higher data rates to be exported. The signal
yield will further increase which will have impacts on the data processing model of the experiment
and the ways how physicists will analyse data on distributed computing facilities. Also connected
to the increased signal yield is the need to produce more Monte Carlo samples. The increase in
CPU work cannot be absorbed by an increase in hardware resources. The changes needed in the
data processing applications will be discussed in the area of multi-processor aware applications,
changes in the scheduling framework of the physics algorithms and the changes in the experiment
data event model to facilitate SIMD instructions.