[Dirvish] USB Connected Drives - speed and reliability

[Keith Lofstrom]

On Mon, May 14, 2007 at 10:02:59PM +0200, Gert Brinkmann wrote:
> Gert Brinkmann wrote:
> 
> > hard disc is connected to an USB2 controller. The Problem: data transfer
> > is very slow. It makes approx. 1 MByte/s only, but it should be more
> > than 20 MByte/s. The hard disc is formatted as ext3fs and is mounted as

IDE drives connected through a USB2/IDE translator may have problems.
Gert's question encouraged me to to my own experiments.  I have been
using IDE hotswap with a customized 2.4.22 linux kernel.  IDE hotswap
is no longer in the kernel, so I have been experimenting with USB2
hotswap on a 2.6.18 kernel (Scientific Linux 5, an RHEL5 clone).

My experiment does multiple passes of filling and emptying a 422GB
ext3 partition on a 500GB Seagate Barracuda 7200.9 PATA-100 drive.
The computer is a 1.7GHz P4 with 1GB of RAM, using a no-name-chinese
3 port USB controller with a VIA Technologies VT6202 (date code 0301)
USB2 interface chip.

I wrote a little C program to write 100MB files, which sits in a 
shell script that does this 4300 times, reporting timing. 

I tried a bunch of different USB2 to IDE translators.  Here are
the results, from awful to OK:
---------------------------------------------------------------------
0) "Control" - direct IDE connection to Promise Ultra133TX2
   PCI controller with PDC20269 controller (date code 0409).
   Results: Wrote 430GB at 30.6MB/sec (4 hrs), as fast as it gets.

1) No-name-chinese, USB to IDE smart cable.  2005 swap meet.
   Claimed Linux compatability.  Genesys GL811E bridge chip .
   Results:  Doesn't even register with "lsusb".  Doorstop,
   probably a defective individual unit.

2) SanMax/inClose USB2DOCK swap cage .  2003 Fry's Electronics.
   Cypress C768013 chip, 0304 date code.
   With a 2.4.22 kernel this locked up after an hour or so.
   2.6.18 Results: locks up immediately

3) No-name-chinese, USB to IDE smart cable.  May 2007 PCCLUB.
   Looks like the picture of a SABRENT device at Tiger Direct.
   Cypress AT2LP Demo RC17 chip according to lsusb -vv .
   Results:  Worked for a while, moved 60GB in about an hour
   (16.7MB/s), then hung and locked up the kernel.

4) ViPower VP-1028LSF swap cage.  2003.03.17 design. 2004 purch.
   Genesys GL811USB chip,  there are multiple versions :-(
   Results:  Wrote 233GB at 11.0MB/sec (6 hrs), slow.

5) UltraSum TZ350B2 external enclosure.  2004 Local computer store.
   Myson Century CS88136-103 chip, 0410 date code. 
   Bypassed wimpy power supply.
   Results:  Wrote 430GB at 17.0MB/sec (7 hrs), tolerable.

6) ViPower VP-1028LSF swap cage.  2003.07.04 design.  2006 purch.
   Cypress C768300A chip 0404, there are multiple versions :-(
   Results:  Wrote 430GB at 19.6MB/sec (6 hrs), tolerable.

--------------------------------------------------------------------
As you can see, a random sampling of USB2 drives and enclosures
revealed scattered results, from broken to flaky to slow to tolerable.
All of the drives are slow, compared to the 48MB/sec one might expect
for a 480MB USB2 bus or thw 30MB/sec of a direct IDE connection.
Perhaps handshaking and such slows USB to IDE connections down a lot.

This will be on the wiki: http://wiki.dirvish.org/index.cgi?USB2drives
Feel free to add to it.  I am especially interested in USB2 to IDE
smart cables, cheap/fast/good .

Keith

-- 
Keith Lofstrom          keithl at keithl.com         Voice (503)-520-1993
KLIC --- Keith Lofstrom Integrated Circuits --- "Your Ideas in Silicon"
Design Contracting in Bipolar and CMOS - Analog, Digital, and Scan ICs