Partially complete a simulation and review results and then opt to continue the simulation
Simulations often take a long time to run. I would like to be able to "pause" the simulation and review the results to check that the initialisation is correct, or that an RTC is working properly, and if I am happy then continue the simulation to completion.
The approach which we are looking to leverage here is one which deviates a little from the original request but gives the same ability to interrogate results part way through a simulation. After investigating options, the best appears to be to have a run for the total duration you are interested in, then set a state saving checkbox in the run and when you wish to interrogate simulation results stop the simulation. We will then make use of ICMLive developed concatenation of results to allow us to tie in the “continuation run” with the results which were stopped earlier. There is a lot more detail to this than I can add here, but it should meet the need.
It would be helpful if you could open a specific part of a results file for a completed simulation run. For instance, schedule and complete a 10 year timeseries run but only load the last year of results.
As others have mentioned on here, the option to open part of a bigger results file would limit the unnecessary long regeneration times, you could pick which variables you would like to load e.g. runoff only or depths and flows, and you could perform reports on part of the results.
I appreciate I can run shorter periods i.e. yearly data rather than the full 10 years and use initial/final state files, but as the remote agent can run multiple simulations it will just fail these as the previous model it is dependent on for the initial state has not completed.
I have successfully completed a long timeseries run using this approach by specifying a start time of the next simulation (that is dependent on the results of the previous for its initial state) by the time I know it takes to run the model e.g. simulations staggered by 1.5hrs for a model I know it takes a maximum of 1hr to run, but this is prone to failure and prohibitively inefficient way of working.
Bradley Nissley commented
The ability to view results as they happen would be very helpful particularly when calibrating the model against a flow survey, initializing soil saturation, ensuring proper operation of an RTC and/or checking suitability of future network upgrades. If the simulation is going as expected, continue the simulation without lost time, otherwise, easy to see that something needs adjustment, make the change and re-run.
Having incremental data written to the results file could also ensure that the entirety of a long simulation is not lost in the event of a power failure or no free disk space remaining.