Seagate shipping 8tb drives
- Thread starter oxygene
- Start date
Associate
I'm getting great speeds with two ST8000AS0002's on Linux. Filling the disks with random data to perform a surface test completed at 150MB/s
And a full read test with dd completed at 190MB/s average.
Creating a ZFS pool over them and filling with 7TB of files averaged about 112MB/s, which I suspect was limited by the source disks, as their maximum read is about that rate.
ZFS / RAIDZ appears to be a much more friendly filesystem for SMR drives as it is essentially a log-structured filesystem which avoids rewriting data, batching random writes in memory and committing to disk as long linear writes.
NTFS/EXT3/4 and traditional journalled filesystems are almost a worst-case for SMR drives because they rewrite data continuously via the journal. Similar effect to how if one uses a journalling filesystem on NAND flash without a FTL, it'll quickly physically destroy the cells.
Code:
dd if=/dev/zero bs=1M \
| openssl enc -aes-256-cbc -nosalt -pass pass:$(openssl rand 32 -hex) \
| pv \
| dd bs=1M of=/dev/disk/by-id/ata-ST8000AS0002-...
8001563222016 bytes (8.0 TB) copied, 53292.3 s, 150 MB/sAnd a full read test with dd completed at 190MB/s average.
Creating a ZFS pool over them and filling with 7TB of files averaged about 112MB/s, which I suspect was limited by the source disks, as their maximum read is about that rate.
ZFS / RAIDZ appears to be a much more friendly filesystem for SMR drives as it is essentially a log-structured filesystem which avoids rewriting data, batching random writes in memory and committing to disk as long linear writes.
NTFS/EXT3/4 and traditional journalled filesystems are almost a worst-case for SMR drives because they rewrite data continuously via the journal. Similar effect to how if one uses a journalling filesystem on NAND flash without a FTL, it'll quickly physically destroy the cells.