HDD Exos X16 16TB ST16000NM001G-2KK103 CrystalDiskMark iops/4K on SATAII, USB 2 and USB 3

[postcd]

Hello,

Hello, can anyone please comment on the comparison of the HDD Seagate Exos X16 16TB ST16000NM001G-2KK103 results between SATAII, USB 2 and USB 3?

from left to right: SATAII, USB2, USB3

The drive should have not been utilized by any other process and was empty during the tests.

In short, i am curious about 4K random operations differences between the interfaces. Differences seems to be significant.. The Q1T1 (Q=queued,T=transactions) is the same for USB2 and USB 3, yet the multithreading 4K (Q32T16) is way different. Why is it so? Is the 3.68MB/s maximum bandwith at Q32T16 for the USB2.0 interface or the interface has some strange limitation for threaded operations when comparing to single thread?

Here i am comparing this drive performance with WD Elements 5TB and 8TB drives. All via USB 3. You can see this drive is a performing significantly better than older drives:

 

[drescherjm]

Is the Q32T16 read score repeatable on the USB3 and SATA II?

 

[postcd]

Is the Q32T16 read score repeatable on the USB3 and SATA II?

I tried to repeat it on USB3. Old vlaues was:
1.64 18.11
new values are (did 5 runs for each instead of 2):
2.80 18.93
(so the read value is similar to SATA and write value stays similar as before which is 1/3 more than SATAII - this is not a problem for me as i will be using the drive as USB3)

 

[sphinx99]

Transactional latency on the USB 2.0 interface?

 

[drescherjm]

I think the results are a combination of latency of the interface and OS cache.

 

[UhClem]

...
In short, i am curious about 4K random operations differences between the interfaces. Differences seems to be significant.. The Q1T1 (Q=queued,T=transactions) is the same for USB2 and USB 3, yet the multithreading 4K (Q32T16) is way different. Why is it so? Is the 3.68MB/s maximum bandwith at Q32T16 for the USB2.0 interface or the interface has some strange limitation for threaded operations when comparing to single thread? ...

I agree with you; there is something interesting at play here.
[You've heard the expression: "The devil is in the details.", right?] Well, it is possibly pertinent to know just how you've "achieved" your USB2 connection. I am prompted to ask because your "Seq 1M, Q1T1" numbers for USB2 [44/44] look too high to me. For a conventional USB2 connection, they should have been ~35/~35. [By conventional, I mean USB2-port/USB2-cable/USB2-SATA-bridge/SATA-HDD.]

Awaiting details ...

I think the results are a combination of latency of the interface and OS cache.

I don't believe the OS cache (ie, system buffers) is involved. CrystalDM may not be perfect, but it is not guilty of such a novice gaffe as using buffered I/O.