Write latency after dropping an index

Eleven indexes removed from production tables, and the write path measured before and after.

  orders INSERT p50    1.9ms → 1.1ms
  orders INSERT p99    8.4ms → 3.2ms
  order_lines INSERT   2.4ms → 1.2ms

  index storage        14.2 GB reclaimed
  buffer pool pressure hit rate 99.1% → 99.6%

at 41,000 inserts a day the p50 saving is about 33
seconds of database time daily, which is nothing.

the p99 halving is not nothing, and the buffer pool
change is the one that matters — those pages are now
holding data.

The median saving is trivial and the tail is not, which is the usual shape of an index-removal measurement. The buffer pool hit rate is the second-order effect and is the reason this was worth doing at all — eleven indexes were occupying memory that the working set now has.