so is that a cumulative query? I'd be more interested in what your latency is during normal business hours. Lots of things can trash tempdb and you don't really care because they happen off hours. Index maintenance, ETL processes, etc. Take a look at Paul's script that collects latencies over time, change the waitfordelay to something that works for you. https://www.sqlskills.com/blogs/paul/capturing-io-latencies-period-time/ But yes, over 1000ms for writes is very bad in most situations.
19 People are following this question.