Opened 2 months ago

Last modified 8 weeks ago

#2637 new Task

VLD-05_Coward_SETTE_inputs

Reported by: acc Owned by: acc
Priority: low Milestone: 2021 WP
Component: MULTIPLE Version: trunk
Severity: minor Keywords: SETTE
Cc: Branch review:
MP ready?: Task progress:

Description

Workplan action

Wikipage: wiki:2021WP/VLD-05_Coward_SETTE_inputs

Collection and rationalisation of SETTE inputs.

A first pass at collecting and cleaning up the SETTE inputs for the current trunk has been made. Provisionally named for r4.2_RC but tried and tested on the current trunk.

Three sets of files are provided in an attempt to reduce the volume of data that has to be moved around if attempting SETTE tests
on new machines:

  • The original files (r4.2_RC_FULL) are chunked and compressed with chunking choices that should make them suitable for most purposes (including extra scaling tests using small processor domains). In some cases these sets include forcing data for much longer time periods than the standard SETTE tests require.
  • The recommended set(r4.2_RC) are the same as above but only sufficient forcing data is provided for roughly twice the length of standard SETTE tests. This reduces the data volume from approximately 5.5GB to 4.5GB but is not as successful as hoped because of two particular restrictions: (a) The ORCA2_OFF_PISCES test, which runs for 95 days but starts from 1st January with climatological inputs. The whole year's worth of data is therefore required in order to provide the end-of December conditions at start-up; (b) The WED025 test, which runs for 10 days but starts on 15th January. Therefore 35 days of forcing are required to support a 20-day integration.
  • The smallest set (r4.2_RC_LITE) (2.5GB) in which the number of significant digits for all variables other than domain or coordinate fields has been reduced. This makes for very much better compression whilst still providing a rigourous enough test for restartability and reproducibility. Note however, that whilst the first two sets will give identical results, answers using the LITE set will differ.

All these files are available from a temporary host: SETTE_inputs but are subject to change before r4.2_RC is tagged. Any updates will be recorded and flagged both in this ticket and on the SETTE_inputs webpage.

Commit History (0)

(No commits)

Change History (2)

comment:1 Changed 2 months ago by acc

The data volume for the r4.2_RC set can be reduced further by lowering the number of significant digits for the time-slices of forcing data that have to be provided for ORCA2_OFF_PISCES and WED015 tests but aren't used. It takes some careful use of ncks to achieve this -but it can be done. The result is a reduction of the overall size from 4.5GB to 3.7GB with no change to any SETTE results. The latest versions have been placed on the SETTE_inputs site and the site's webpage has been updated. The same trick with the LITE version yields only a marginal change in overall volume and so has not been applied to that set.

comment:2 Changed 8 weeks ago by acc

Fully documented the process on: wiki:2021WP/VLD-05_Coward_SETTE_inputs Note some of these files are likely to be changed before the r4.2_RC tag. In at least one instance, for example, the AGRIF nest is too small for tests with halo=2. There should now be sufficient information on this page to repeat the processing steps easily and create replacements as necessary.

Note: See TracTickets for help on using tickets.