I bought an 8TB Seagate ST8000DM004-2CX188 hard drive second-hand and did not test it properly at the time. I know this is an SMR drive, but I have used a similar model for years and expected it to be adequate for archiving large 4K movie files. Large sequential writes initially reach roughly 120 MB/s, then quickly fall below 1 MB/s. The drive is formatted as NTFS, has no important data on it, and changing the allocation unit size has not helped. SeaTools and several other utilities report no problems, including after both short and long tests. Is there anything else I can try before replacing it? Destructive tests or reformatting are acceptable.
4 Answers
Since the disk is empty, you can run a destructive full-disk write/read benchmark or a complete overwrite followed by verification. Watch the transfer graph and SMART data during the test. If it repeatedly drops below 1 MB/s during a simple sequential workload while the connection and health data are normal, there probably is not a filesystem or allocation-unit setting that will fix it—the drive is either exhibiting the limitations of its SMR implementation or is defective.
An 8 KB NTFS allocation unit is not ideal for a drive used almost exclusively for large media files, although it should not by itself cause this dramatic slowdown. Reformatting with a larger cluster size, such as 64 KB, may reduce metadata overhead for that workload, but it will not remove the SMR cache limit. If sustained write performance matters, replacing it with a conventional magnetic recording drive or an SSD is the practical solution.
Check the connection before declaring the disk dead. Try a different SATA data cable and motherboard port, verify that the link is negotiating at the expected speed, and test it outside any USB enclosure or storage pool. Also check SMART attributes such as pending sectors, reallocated sectors, reported uncorrectable errors, interface CRC errors, and the drive temperature. A clean surface test is reassuring, but it does not rule out severe write-performance problems or a failing controller.
This is very typical SMR behavior. The drive can write quickly while its onboard cache is available, but once that fills it has to reorganize overlapping tracks in the background. Sustained writes may then become extremely slow, especially if the drive has previously held data or receives anything other than a perfectly sequential workload. A benchmark that writes a large file directly to the disk can confirm whether the slowdown is simply the SMR cache being exhausted.
For comparison, the other drive reaching about 120 MB/s does not mean this one will sustain that speed. SMR models often look normal for the first few gigabytes and then collapse once the cache is full.