Hello,
We have put together few systems with 4x 512GB or 1TB Samsung 850 Pro drives in software RAID-10, in CentOS 6.7, 64-bit. New, functional drives. The results from dd, for all systems, are surprisingly disappointing :
dd if=/dev/zero of=test bs=64k count=16k conv=fdatasync
yields around 400 MB/s. Altough :
hdparm -t /dev/md0
yields about 1200 MB/s (this is for sequential read).
Manufacturer rates the drives for 520 MB/s sequential writes, so we were expecting around 1000 MB/s for RAID-10. We did try different block sizes (8k to 2M).
Why is dd performance so low? Thanks.
We have put together few systems with 4x 512GB or 1TB Samsung 850 Pro drives in software RAID-10, in CentOS 6.7, 64-bit. New, functional drives. The results from dd, for all systems, are surprisingly disappointing :
dd if=/dev/zero of=test bs=64k count=16k conv=fdatasync
yields around 400 MB/s. Altough :
hdparm -t /dev/md0
yields about 1200 MB/s (this is for sequential read).
Manufacturer rates the drives for 520 MB/s sequential writes, so we were expecting around 1000 MB/s for RAID-10. We did try different block sizes (8k to 2M).
Why is dd performance so low? Thanks.