When SAVE_TIMING is TRUE and SAVE_TIMING_DIR points at a legal path,
performance data and provenance are saved in the directory
<SAVE_TIMING_DIR>/performance_archive
For ACME, SAVE_TIMING is TRUE by default. (However, it is disabled for
jobs created using create_test.)
For current CIME, the default for SAVE_TIMING_DIR is UNSET. However
SAVE_TIMING_DIR is defined in
cime/config/acme/machines/config_machines.xml for a number of systems.
For these systems, these data are copied to the performance archive and
the volume of data will grow over time (unless the indicated directory
is swept periodically) and needs to be managed by the ACME POC for the
system or some other designated person.
The performance team manages these data for the production platforms
(Anvil, Edison, Cori-Haswell, Cori-KNL, Mira, Titan, and Eos) when
generated using the primary ACME allocations. Other systems on which
data is being saved (based on config_machines.xml) are
anlworkstation
blues
cascade
constance
cetus
grizzly
lawrencium-lr2
lawrencium-lr3
melvin
redsky
sandia-srn-sems
skybridge
sooty
theta
wolf
My assumption is that the data for the other systems is never looked at,
and that there is no reason to collect it. In the near future we will be
disabling this collection on all but the production systems, and the
POCs for each systems should check the indicated directories and clean
up what has been saved there (erroneously).
If this is incorrect, and the performance data and provenance being
collected on the above non-production systems is being used and managed,
please send us a note, or you can wait until after config_machines.xml
has been modified and then submit a PR to restore the capability on the
desired development systems.
Thanks.
Pat Worley
