Checkpoint

Manages checkpoint/restart capability.

A user may request periodic dumps of ATS Checkpoint Data in the “checkpoint” sublist. The user has no explicit control over the content of these files, but has the guarantee that the ATS run will be reproducible (with accuracies determined by machine round errors and randomness due to execution in a parallel computing environment). Therefore, output controls for Checkpoint Data are limited to file name generation and writing frequency, by numerical cycle number. Unlike “visualization”, there is only one “checkpoint” list for all domains/meshes.

checkpoint-spec

• “file name base” [string] “checkpoint”

• “file name digits” [int] 5

• “single file checkpoint” [bool] true If true, writes all checkpoint to one file. If false, uses a subdirectory with one file per mesh. false is required if meshes exist on other communicators than MPI_COMM_WORLD, but this is toggled if the code detects that this is necessary.

INCLUDES: - [io-event-spec] An IOEvent_ spec

Example:

<ParameterList name="checkpoint">
  <Parameter name="cycles start period stop" type="Array(int)" value="{0, 100, -1}" />
  <Parameter name="cycles" type="Array(int)" value="{999, 1001}" />
  <Parameter name="times start period stop 0" type="Array(double)" value="{0.0, 10.0, 100.0}"/>
  <Parameter name="times start period stop 1" type="Array(double)" value="{100.0, 25.0, -1.0}"/>
  <Parameter name="times" type="Array(double)" value="{101.0, 303.0, 422.0}"/>
</ParameterList>

In this example, checkpoint files are written when the cycle number is a multiple of 100, every 10 seconds for the first 100 seconds, and every 25 seconds thereafter, along with times 101, 303, and 422. Files will be written in the form: “checkpoint00000.h5”.