Restarting ES-MDA

Imagine you are conducting an experiment consisting of three iterations of ES-MDA, with each iteration potentially taking hours or even days. For the purposes of this example, assume that the “Relative weights” are set to 4, 2, 1.

As part of running this experiment, ERT creates four ensembles, which by default are named default_0, default_1 and default_2, default_3. It is possible that too many realizations fail during the model evaluation in default_3, leaving no responses to analyse. Various factors, for example license server issues or compute cluster downtime, can lead to such failures. One solution is to restart from default_2 which is straightforward in ERT.

Steps to Restart from `default_2`:

1. Check the “Restart run” checkbox.

2. Use the “Restart from” drop-down list to select the ensemble you wish to restart from.

   In this example, you’ll pick default_2.

3. Click “Run Experiment”.

   This creates a new ensemble which by default is called default_2_3. This ensemble will use the relative weight 1, consistent with the last value specified in the “Relative weights” sequence: 4, 2, 1.

Note:

• When initiating a restart, the PRE_FIRST_UPDATE (see HOOK_WORKFLOW) will only execute if “Restart from” is set to default_0. This means that any workflows that for example manipulate observations, will have run manually before doing the restart.