Hello, all. We are using fio log files to get more information about disk performance and have face a problem with this approach when run test with a lot of threads - 100+, depending on storage. Some threads starving due to big latency and gets only few IO per second (or even less than one). Fio produce log files with IOPS'es, also fio truncate iops to integers. That produce some difference between json output(total_ios) and iopses from log multiplied for time intervals.
We see three possible workaround, does any of them build in into latest fio (git master): 1) Have a global log instead of log per thread. 2) Write operations instead of operations per second 3) disable rounding of IOPS'es to increase accuracy. There are some examples of fio logs. thread_1_iops.log 610, 6, 1, 0 1090, 4, 1, 0 1519, 3, 1, 0 2053, 9, 1, 0 2615, 12, 1, 0 3192, 12, 1, 0 3753, 10, 1, 0 4263, 11, 1, 0 thread_2_iops.log 592, 5, 1, 0 thread_3_iops.log 5129, 8, 1, 0 Kind regards, Gleb Stepanov. -- To unsubscribe from this list: send the line "unsubscribe fio" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
