On 6 June 2016 at 22:41, Karan Singh <karan@xxxxxxxxxx> wrote: > > I am running a FIO job and at the same time monitoring bandwidth / > IOPS using iostat command. What i found is FIO reports total IOPS as > 431 ( 215 read , 216 write, see below output ) , but iostat r/s and > w/s reports in the magnitude of a few thousands. > So my query is > > #1 : Is r/s and w/s fields in iostat output corresponds to read and > write IOPS ? ( https://serverfault.com/questions/342273/disk-iops-count-in-gnu-linux > mentions r/s and w/s as IOPS) > #2 : If yes then why does they dont match with FIO IOPS results ? See below for a guess. > #4 : Whats your favourite command / tool to monitor system IOPS ? It depends on what level you want to record the statistics. The diagram on http://www.brendangregg.com/linuxperf.html gives some good starting tools. > # fio --filename=/dev/nvme0n1 --name=rw-4M --rw=rw --ioengine=libaio > --bs=4M --numjobs=1 --direct=1 --randrepeat=0 --iodepth=1 > --runtime=300 --ramp_time=5 --size=100G --group_reporting Your block size is huge - won't the kernel wind up breaking this up and submitting multiple smaller requests to the device? What's are nvme0n1's queue/max_* values? Do you have better correlation at smaller block sizes (e.g. 4k)? The information on https://www.ibm.com/developerworks/community/blogs/58e72888-6340-46ac-b488-d31aa4058e9c/entry/exploring_the_linux_storage_path_tracing_block_i_o_kernel_events?lang=en will allow you to trace I/Os going through the block layer so you can get a better idea of what's going on (note that fio often tells you if the kernel did merges in its block device output). -- Sitsofe | http://sucs.org/~sits/ -- To unsubscribe from this list: send the line "unsubscribe fio" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
