External disk drive write performance decreases after some time. What could be the reason?

Question

I have an external USB 3 "spinning rust" 2.5'' Harddisk.

Serial number lookup at Western Digital reveals this is a model WDBYNN0010BBL, "WD My Passport (Ultra) Portable Storage, Blue". 1 TB = 931.48 GB, cheap and cheerful. No data sheet. WD instrumentation software provides no way to find out more about the actual harddisk in the enclosure.

Although USB 3, I connect it to an USB 2 interface on the PC (it's an old PC).

I'm filling the disk with semi-random data, directly writing to the partition (i.e. /dev/sdd1 in Linux).

Problem:

Initially, the writing program reports a write performance of ~28 MiB/s (way below the max performance of USB 2, 50.7 MiB/s).

The writing program does nothing except write a chunk of data, then sync.

Once around 7 GiB have been written, write performance becomes flaky and starts to pogo between 8 MiB/s and 28 MiB/s.

After about 115-120 GiB have been written, write performance decays further to consistently hover at around 6 MiB/s, temporarily going as low as 2 MiB/s for a few hundred MiB.

However, once you suspend the program for half a minute (CTRl-Z) and then resume it, write performance goes back up to 28 MiB/s. And then after some time, the performance decays again.

Question:

Is there something in the construction of this disk which allows only short bursts of acceptable write speed? Does it pretend to be fast by having a large internal (nonvolatile) cache?

I have a number of other external USB disks (some from WD, although none that are of the exact same model) and they all manage to sustain 10+ MiB/s write speed, which makes backups at least viable as they can run in an afternoon instead of 24h.

Update #1

It is also running the latest firmware v1.019, as per "WD Firmware Updater" for Windows)

Update #2

"WD Drive Utilities" for Windows says it passes the S.M.A.R.T. test, the "quick drive test", and the "complete drive test". Nice!

Update #3

Filling the disk on another Linux machine via an USB3 interface upgrades the write throughput to 100 MiB/s. However, the decay is still there, going down to 33 MiB/s or even 9 MiB/s, so that at the end of the operation, the mean write throughput for the whole disk was just 30 MiB/s.

Update #4

smartctl --all /dev/sdd reveals:

Device Model:     WDC WD10SDZW-11UMGS0
Serial Number:    WD-WXT1A57EU75S
LU WWN Device Id: 5 0014ee 607b8ce4a
Firmware Version: 01.01A01 <--- HUH? I thought it was 01.019?? 
User Capacity:    1,000,171,331,584 bytes [1.00 TB]
Sector Sizes:     512 bytes logical, 4096 bytes physical
Rotation Rate:    5400 rpm
Form Factor:      2.5 inches
Device is:        Not in smartctl database [for details use: -P showall]
ATA Version is:   ACS-3 T13/2161-D revision 5
SATA Version is:  SATA 3.1, 6.0 Gb/s (current: 6.0 Gb/s)
Local Time is:    Sun Dec  8 12:58:26 2019 CET
SMART support is: Available - device has SMART capability.
SMART support is: Enabled

All offline tests passed, reallocated sector count is 0 (though this may not mean much).

Answer 1

There may be many reasons for decreasing performance, but I understands it decreases steadily the more data is written to the drive.

My first thought was that it can have some bad sectors, but I would only impede the speed at a few points and not be so consistent.

Try connecting the drive to another computer and test it's performance. USB 2.0 can't have too much of a throughput though. 40MB/s is practically a max and half that is typical for cheap drives.

Test it on USB 2.0 and 3.0 (if you can) but at least on a different machine (or two). If the issue will occur on another machines then I would think there's something wrong with the drive itself. Check WD's page if there isn't any mention about new firmware fixing the issue you're having.

Answer 2

There are many possible root causes:

• flacky driver or controller: USB protocol (and cheap implementations) has a somewhat bad reputation regarding stability. Maybe something got overflooded and needs some time to recover;

• SMR HDD: hard disk based on shingled magnetic recording generally have some fast (or landing) PMR zones intermixed with larger slow or shingled ones. If a landing zone is completely filled it needs to be flushed to the shingled zone, which take time (with the disk appearing as "freezed" or very slow in the meantime);

• bad sectors spread around the disk surface.

Can you post the output of smartctl --all <dev>? Please also try the same disk on another computer: does it behave differently?