Requirements for Certification/Verification
of Level 3 Code
Certification:
-
Must be fully tested on a Linux system.
(Inconsistencies in performance between Linux and other operating
systems are sometimes observed.)
-
Must work "out of the box".
(Private modifications to code or RCP files, no matter how minor, are
not acceptable.)
-
Must run without crashing on samples of order
100K events:
-
On real data and Monte Carlo.
-
Single tool on its own and with
all other released tools.
-
Memory leaks must be checked for.
-
Particularly when new or modified tools and/or filters are being
proposed, the filter code and the related
trigger database entries
must be tested, as well as that of the associated tools.
-
The performance must be measured in Trigsim, running a properly constituted
trigger list including the proposed new triggers.
-
It is not enough to "simulate" this stage by applying
private cuts on the objects produced by the tool.
-
Help with entering tool/filter/trigger definitions into the trigger
database can be provided by the DØ triggermeisters or Terry Wyatt.
-
Tool performance studies/plots: efficiency, rejection, etc. must be
presented to the L3 algorithms group.
(Interaction with and approval of the relevant ID group(s) obviously
essential also.)
-
Must test persistency of physics_results:
-
Write out events.
-
Read back in (without simulator being run) and check physics
distributions OK.
-
Up-to-date documentation of tool and filter algorithms and results on data and
Monte Carlo must be provided.
(We really mean it!).
-
Once approved by the L3 algorithms group the performance studies
must be presented to the DØ trigger board and receive approval before they are able to
run online.
Verification:
-
As soon as possible after the code is run online: run the
simulator on data offline and check consistency with online results:
-
Event selection/trigger statistics.
-
Distributions from debug and physics results blocks.