There are no hard and fast rules for this. You must test. It will depend on many variables. How many indexes do you have? How wide are they? What order is the data in when it arrives? The work to drop all indexes before loading, then building them after loading can be less than keeping all of them organized while loading. Try both ways. Try a mix of just dropping all nonclustered but keeping the clustered, especially if your data set can be sorted in the same order as the clustered index before loading.