This ticket ended up exploring a the timing of the diaCalculation plugin system using the full HiTS2015 dataset. Runs of the current stack we compared to the updated diaCalculation with multi plugins. The timing plotted is the amount of time taken to update and create new diaObject summary statistics. This time does include some database access, though looking at a finer grained timing of the new multi-plugins, ~95% of this time is taken up by plugins on average.
First is the timing for the current stack with a linear fit to data of timing versus total number of diaObjects that were processed (i.e. updated or created). Errors on the fit are estimated by bootstrapping the datapoints and re-fitting.
And using the multi-plugins:
The multi plugins are extremely linear compared to the single plugins.