Renewed Exos 20 TB disk - should I be worried?

TrueNAS General
,
1

Some months ago, I bought a renewed 20 TB Seagate Exos X22 hard drive from Amazon. It being a renewed disk, it had only a month of warranty but I thought it could be used as a part of a mirror vDEV.

So I did what I do with all the new drives, I qualified it with @jgreco’s Solnet Array Test script for a full week non-stop, a destructive write disk test a couple of times (diskinfo -wS /dev/da(x)), all four SMART tests (short, long, conveyance, offline). The drive checked out to my eyes - no issues on either of these tests.

Several months later, the drive is performing quite well (as far as I can notice) and SMART output hasn’t failed, but observing scrubbing speed from this night got me thinking - the drive is about half-full (10.33 TB), and scrub is scrubbing only the occupied space, but this drop in speed is considerable?

image
image2296×413 42.1 KB

Pool fragmentation is zero, if that is even relevant.

image
image962×71 3.67 KB

My concerns are:

  • if I interpolate this curve to the full drive, will the sequential [read] speed be close to zero?
  • is this curve drop-off normal or something you would be concerned?
    • I understand that some drop off speed is expected due to the physical geometry of HDD but isn’t this graph a little too steep?
  • would you have any concerns making a mirror with this drive?

I can post additional details if they’re needed.

2

I would assume it is normal, and that the curve will reach a non-zero low point.

Due to the sheer size of the disk I would not use it in anything less than raidz2 or 3-way mirror. But you get to define your own level of paranoia.

1 Like
3

Thank you - I am in the process of procuring another disk of that size and forming up a mirror on the primary storage but in the meantime, there is a backup server that it replicates to.

4

The drop in speed is perfectly normal.
I saw the same behavior when I was performing burnin of my brand new X18 18TB drives.
To understand what is happening, the drive is spinning at a constant angular speed ( or simply put constant RPM).
The heads seem to start at the outter edge (with faster speed relative to the heads) of the platter and progressivelly move forward to the center (with slower speed relative to the heads).
I believe I posted my progress with screenshot on the old FreeNas forum back then.
From recollection, I would loose half the throughput toward the end.

About scrubbing, ZFS only scrub data that is present on the pool, so if the pool is 10% full, then only the 10% will be checked.

For now, a screenshot from one of the user: