Smoothing currently balances the CUs required to complete background tasks across 24 hours. This often means that overnight we are under-utilising our capacity and offsetting some of the load to our peak usage period during the day. Since the peak period also includes many interactive processes, smoothing is actually creating usage peaks right when we are using the processes which are most impacted by those peaks.
We should be able to control smoothing by setting a percentage of capacity that can be utilised by background processes. If background usage is below this figure the spare capacity should then be used to speed up background processing.
Example: Consider a 6 CUhr background operation starting at midnight. Currently, it would consume 0.25 CUhr per hour for 24 hours. Running an F2 Sku, I set my Background Processing Limit to 50% between 00.00 and 06.00. Now the operation consumes 1 CUhr per hour for 6 hours and no consumption for the remaining 18 hours.
