Thank you very much, Vincent! I guess the difference between FIO clat and the time delta from blktrace is caused by the start time for clat. What is interesting to me is that every action combination I tried as the time delta in blkparse is generally lower than what FIO clat reported. On Tue, Sep 19, 2023 at 6:38 AM Vincent Fu <vincentfu@xxxxxxxxx> wrote: > > On 9/18/23 12:54, Jay Han wrote: > > Hello, > > I am seeing a stiff delta between the fio clat vs blktrace data and > > wondering how fio clat is measured in terms of the I/O processing > > step. In blktrace, I measured the time delta between the complete(C > > action) and issued(D action) per LBA basis. > > > > Can you tell me how fio clat is calculated please? > > > > Thank you, > > Jay. > > Here is an extended discussion of fio's latency calculations: > > https://github.com/vincentkfu/fio-blog/wiki/How-does-fio-measure-latency
