Re: Slow USB3 drive (Intel amd64)

[Mark Carroll <mtbc%ixod.org@localhost>]

On 30 Oct 2018, Greg Troxel wrote:

> You are using the block device, not the raw device.   This traditionally
> mattered on Unix and I think it might not on modern Linux.
>
> Use /dev/rsd0d instead. I am hopeful that this is your only problem.

Aha, thanks to you and Michael I have learned something.

(snip)
> Also, I recently tried out a USB3 samsung external SSD (again, not
> exotic - just basic decent stuff).  I got around 100 MB/s, which is not
> really fast, but way better than USB2.

That's decent indeed. Following your advice:

sd0: 1073741824 bytes transferred in 395.850 secs (2 712 496 bytes/sec)
sd1: 1073741824 bytes transferred in 393.434 secs (2 729 153 bytes/sec)
wd0: 1073741824 bytes transferred in 18.443 secs (58 219 477 bytes/sec)
rsd0d: 1073741824 bytes transferred in 13.787 secs (77 880 744 bytes/sec)
rsd1d: 1073741824 bytes transferred in 12.329 secs (87 090 747 bytes/sec)
rwd0d: 1073741824 bytes transferred in 2.560 secs (419 430 400 bytes/sec)

with
sd0 at scsibus0 target 0 lun 0: <Seagate, Expansion, 0706> disk fixed
sd1 at scsibus1 target 0 lun 0: <WD, Elements 25A2, 1019> disk fixed

so significantly better. I put in extra spaces for readability.

> But, I got mysterious read errors about 100GB into the drive when
> reading back.

On a separate but almost identical system running OpenBSD 6.1 I get rare
intermittent read errors on these drives. As they are not all that old
and with cables from different sources, I'm wondering if it is not the
drives themselves at fault: I am curious to see if I get the same errors
from NetBSD.

>>> xhci driver is at best experimental.
>
> I'm not sure that's really true in 8.

Aha, I had hoped so, at least given that my hardware's probably now
neither very new nor rare.

-- Mark