[ET Trac] [Einstein Toolkit] #1664: CarpetIOHDF5 should not try to search for grid scalar s and DSITRIB=CONSTANT arrays in all hdf5 files
Einstein Toolkit
trac-noreply at einsteintoolkit.org
Wed Oct 1 12:05:47 CDT 2014
#1664: CarpetIOHDF5 should not try to search for grid scalar s and
DSITRIB=CONSTANT arrays in all hdf5 files
--------------------------+-------------------------------------------------
Reporter: rhaas | Owner: eschnett
Type: enhancement | Status: new
Priority: minor | Milestone:
Component: Carpet | Version: development version
Keywords: CarpetIOHDF5 |
--------------------------+-------------------------------------------------
Currently, if during a checkpoint recovery a grid scalar or
DISTRIB=CONSTANT grid array is not found in the first file, then
CarpetIOHDF5 will eventually open and parse all file looking for this
dataset. This happens eg when one activates a new thorn with a
checkpointed scalar or also when a new code version of a thorn that
declares new checkpointed variables is used.
Also, the current behaviour is that grid variables that are not found at
all (as compared to grid variables for which only some but not all
components are found) are silently ignored.
Finally, the current checkpointing code will write a copy of grid scalars
and DISTRIB=CONSTANT grid arrays in every single checkpoint file, which
means that, for a valid set of checkpoint files, if the grid variable is
not found in the first file it will also never be found in any of the
other files.
The ticket is thus an enhancement request to have CarpetIOHDF5 stop
searching for the grid variable after the first file. This will not change
the behaviour of any current checkpoint files but will speed up the case
where a grid variable is missing. This will either speed up a final error
report (if eg a thorn was using the missing data and reports poisoned
data) or the restart with extra thorns significantly.
--
Ticket URL: <https://trac.einsteintoolkit.org/ticket/1664>
Einstein Toolkit <http://einsteintoolkit.org>
The Einstein Toolkit
More information about the Trac
mailing list