Skip to content

Draft: Throughput test for calo decoding and electron line

Maxime Schubiger requested to merge maxime_test_branch into master

This MR should not be merged !

  • This is a merge request to get an official throughput test with the gitlab ci. According to my latest tests the whole sequence (i.e. calo decoding + selective matching + brem recovery + 2 electron trigger lines) shows only a 6.7% decrease in throughput compared to the master branch ! Let's double check that.
  • This branch should already contain a commit from Roel that allows the throughput test to use MDF files.
  • The minimum bias and bs->phiphi samples needed for the throughput tests are here :
    • /eos/lhcb/user/m/maxime/public/forAllen/upgrade_mc_minbias_scifi_v5_000.mdf
    • /eos/lhcb/user/m/maxime/public/forAllen/upgrade_mc_bsphiphi_scifi_v5_000.mdf

Throughput results for hlt1_pp_default with calo:

GeForce RTX 3090      │████████████████████████████████████████████       224.19 kHz
GeForce RTX 3080      │█████████████████████████████████████              188.70 kHz
Quadro RTX 6000       │████████████████████████████████                   160.14 kHz
GeForce RTX 2080 Ti   │██████████████████████████████                     153.59 kHz
Tesla V100-PCIE-32GB  │███████████████████████████                        137.28 kHz
AMD EPYC 7502 32-Core │███                                                19.32 kHz
Intel Xeon E5-2630 v4 │▌                                                  3.61 kHz
                      ┼────┴────┼────┴────┼────┴────┼────┴────┼────┴────┼
                      0         50       100       150       200       250

This is about 4.7% slower than master (great!).

Edited by Daniel Campora Perez

Merge request reports