High dead time at the start of the run
Starting from Allen v4r8 in production, we encountered higher dead time at the start of the run. Note the difference after the change of trigger configuration to a newer stack:
Note also https://lblogbook.cern.ch/HLT/937 .
We drop a large number of MEPs at the beginning of a run, one potential reason for that might be this merge request: !1554 (merged). To test this, we run on the testbench with the production stack_RTA_2024_05_08: MooreOnline/run MooreOnline/MooreScripts/scripts/testbench.py MooreOnline/MooreScripts/tests/options/HLT1/Arch.xml --working-dir /tmp/testConfigTime/ --measure-throughput=120 --data-dir /calib/online/MEP_dumps_20_05_24/ --hlt-type /group/hlt/commisioning/tck.git:0x1000104e --geometry="run3/2024.Q1.2-v00.00" --conditions="master" --partition TESTALLEN
Without this merge request , the configuration time 17 s
emu_wo_MR.log HLT1_0_wo_MR.log
With this merge request, the configuration time 33 s
emu_with_MR.log HLT1_0_with_MR.log
With the initial version of this merge request !1554 (cd501ed0) (only add what is missing), the configuration time 30 s
emu_with_initial_MR.log HLT1_0_with_initial_MR.log
An additional 15 seconds in configuration, lead to discarding MEPs at the start of each run