On Sun, May 3, 2020 at 2:29 PM Mark Knecht <markkne...@gmail.com> wrote:
>
> I've used the WD Reds and WD Golds (no not sold) and never had any problem.
>

Up until a few weeks ago I would have advised the same, but WD was
just caught shipping unadvertised SMR in WD Red disks.  This is going
to at the very least impact your performance if you do a lot of
writes, and it can be incompatible with rebuilds in particular with
some RAID implementations.  Seagate and Toshiba have also been quietly
using it but not in their NAS-labeled drives and not as extensively in
general.

At the very least you should check the model number lists that have
been recently released to check if the drive you want to get uses SMR.
I'd also get it from someplace with a generous return policy and do
some benchmarking to confirm that the drive isn't SMR (you're probably
going to have to do continuous random writes exceeding the total
capacity of the drive before you see problems - or at least quite a
bit of random writing - the amount of writing needed will be less once
the drive has been in use for a while but a fresh drive basically acts
like close to a full-disk-sized write cache as far as SMR goes).

> Build a RAID with a WD Green and you're in for trouble. ;-)))

It really depends on your RAID implementation.  Certainly I agree that
it is better to have TLER, but for some RAID implementations not
having it just causes performance drops when you actually have errors
(which should be very rare).  For others it can cause drives to be
dropped.  I wouldn't hesitate to use greens in an mdadm or zfs array
with default options, but with something like hardware RAID I'd be
more careful.  If you use aggressive timeouts on your RAID then the
Green is more likely to get kicked out.

I agree with the general sentiment to have a spare if it will take you
a long time to replace failed drives.  Alternatively you can have
additional redundancy, or use a RAID alternative that basically treats
all free space as an effective spare (like many distributed
filesystems).

-- 
Rich

Reply via email to