I believe the effect you're seeing is that multiple threads will make the IO look more random at the block device. Your configuration uses sequential writes, but with numjobs=4 you'll have each job writing at a different place in the device. Thus, lots of seeks. If the jobs had rw=randwrite, you'd probably see about the same performance in both configurations. One configuration where the 4-job config would run faster is if rw=randwrite and iodepth=1. (Or, ioengine=sync.) In that case, the 4 jobs would keep the IO queue more full than 1 job. But sequential writes with a deep IO queue (like you have) is generally the best case for a single raw block device. Note, even solid-state/flash drives see better performance with sequential writes than with random writes. Best regards, Josh On Jan 16, 2013, at 8:01 AM, Alireza Haghdoost <haghdoost@xxxxxxxxx> wrote: > Hello > > I am running fio on the top of raw block device with following two > configurations and I saw the first config (single threaded) can put > more load on the underlying storage system compared to second config > (with 4 thread). However, I expected to see the opposite side (more IO > load with multiple job compared to single job). Does it make sense to > you ? Is there any locking or handshaking between multiple jobs in the > second config that reduce the IO injection rate of fio ? > > ; -- start job Config 1 file -- > [global] > rw=write > size=5G > > [job1] > ioengine=libaio > iodepth=8192 > bs=4k > direct=1 > numjobs=1 > directory=/dev > filename=sdp > ; -- end job file -- > ------------------------------------------------ > > ; -- start job Config 2 file -- > [global] > rw=write > size=5G > > [job1] > ioengine=libaio > iodepth=8192 > bs=4k > direct=1 > numjobs=4 > directory=/dev > filename=sdp > ; -- end job file -- > -- > 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 -- 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
