Take the following 3 commands:
fio --name=write_throughput --numjobs=8 \
--size=10G --time_based --runtime=60s --ramp_time=2s --ioengine=libaio \
--direct=1 --verify=0 --bs=1M --iodepth=64 --rw=write \
--group_reporting=1
This results in a write speed just over 1GB/s
Difference: 1 job instead of 8.
fio --name=write_throughput --numjobs=1 \
--size=10G --time_based --runtime=60s --ramp_time=2s --ioengine=libaio \
--direct=1 --verify=0 --bs=1M --iodepth=64 --rw=write \
--group_reporting=1
This results in a write speed around 255MB/s
Difference: 1 job, 4K bs
fio --name=write_throughput --numjobs=1 \
--size=10G --time_based --runtime=60s --ramp_time=2s --ioengine=libaio \
--direct=1 --verify=0 --bs=4K --iodepth=64 --rw=write \
--group_reporting=1
This results in 8MB/s
I find this confusing. I'm basically looking for a scenario that tells me "how fast is writing a single 10GB file" and these different options give me wildly different output.
I'm not looking for a theoretical maximum. I'm looking for true to life performance. For an application that writes a file as it is generating it. So not having first created a section on the filesystem and then filling it up with bytes, either.
What am I misunderstanding here? Is fio not the tool for this?
