Speaker
Lukas Alexander Heinrich
(New York University (US))
Description
During the 2013/14 shutdown of the Large Hadron Collider (LHC) the
ATLAS first level trigger (L1T) and the data acquisition system (DAQ)
were substantially upgraded to cope with the increase in luminosity
and collision multiplicity, expected to be delivered by the LHC in
2015.
To name a few, the L1T was extended on the calorimeter side (L1Calo)
to better cope with pile-up and apply better-tuned isolation criteria
on electron, photon, and jet candidates. The central trigger (CT) was
widened to analyze twice as many inputs, provide more trigger lines,
and serve multiple sub-detectors in parallel during calibration
periods. A new FPGA-based trigger, capable of analyzing event
topologies at 40 MHz, was added to provide further input to forming
the level 1 trigger decision (L1Topo). On the DAQ side the dataflow
was completely remodeled, merging the two previously existing stages
of the software-based high level trigger into one.
Partially because of these changes, partially because of the new
trigger paradigm to have more full event analysis, the high level
trigger (HLT) execution framework and the trigger configuration system
had to be upgraded, tools and data content had to be adapted to the
new ATLAS analysis model. In this paper we describe this work:
The algorithm execution framework was changed to seamlessly work
within the merged HLT, the data access providers were adapted to the
new dataflow. The event building, at which point all data are
retrieved from the readout system, can now dynamically change with
progressing event feature extraction, allowing a more flexible
adjustment to dataflow constraints. The cost monitoring framework
which analyzes data access and CPU consumption, even prior to data
taking, was extended to work within the merged system, several other
improvements followed.
The HLT execution was moved to a memory-saving multi-process
application, in which many event processors are forked after the
system configuration. They thus share common data such as geometry and
conditions information, leading to a dramatic reduction in the overall
memory consumption. Compared to Run 1 many more event processors can
run on each machine.
Upon request from the ATLAS physics groups a new kind of data stream
was implement, in which only a small subset of the reconstructed
trigger objects and no detector data is stored. Since the trigger
reconstruction in Run 2 almost compares to the offline in resolution,
these data make search analyses that require high statistics feasible.
As a consequence of these changes, the new ATLAS data model, and the
new dual environment analysis approach, the tools that are provided
for trigger aware analysis had to be completely restructured. In
particular the reduction and specialization of data content in derived
data sets was posing a challenge for the trigger, a new trigger data
slimming was invented.
The database driven trigger configuration system, which describes the
physics implemented at L1 and HLT, needs to reflect all changes in the
L1 and HLT system. It now incorporates the configuration for the new
L1Topo trigger, has extended the configuration capabilities of the
L1Calo and CT and the describes are merged HLT. A new system for
automatically adjust the trigger prescale factors to the dropping
luminosity during a run was devised and implemented.
We also present measurements of the trigger execution on first data
with the new ATLAS trigger algorithms and selection.
Author
Lukas Alexander Heinrich
(New York University (US))
