Alan Stern wrote:
On Mon, 6 Feb 2006, [EMAIL PROTECTED] wrote:

Hi everybody,
I recently bought 2 large USB disk for doing backups. New disks are approx. 400 GB in large, usb 2.0. My backup program (Bacula 1.36.3) sucessfully writes to one of that disks and everything goes well. I bought usb disks for quick solution for doing backups as my streamer eaten another tape :? , and I would be happy to replace it with usb disks for convienience, reliability and - price.

The problem occurs when I need to copy a bigger file from one disk to another: no matter if it is normal SCSI disk in the computer or second USB drive. While copying bacula archive (a single file, which is 8GB in large now and will grow up to 50GB) afer some time (300-900MB) copying process slows down until it stops, slowly restart and continue copying. Because of that slow average speed of copying to and from usb 2.0 is about 2-3 MB/s only. While this process goes I can see in logs:

[kernel] usb 1-2: reset high speed USB device using ehci_hcd and address 5

This line repeats continously until I finish copying. After starting copying again, I still can see the message, but now since the very beginning. Unregistering the device and restarting the backup machine solves a problem until next copy.

I've checked md5 sums of copies and they are correct (uf!), I can also restore backups. I assume the problem lies somewhere in kernel, because it happens also on another computer.

Not necessarily. The problem could lie in the drive or in the USB cables. Be sure to check or replace them, and don't forget the cables inside the
computer case (if you're using one of the front-side USB ports that isn't
directly adjacent to the motherboard).

I have to say that I really need a solution, because I feel unsafe to perform backups to a device which is not managed correctly by my kernel. Maybe I miss something in the kernel configuration or I need another solution? The filesystem I use is ext3.

To get more information, you should turn on the USB Mass Storage verbose debugging option in the kernel configuration (CONFIG_USB_STORAGE_DEBUG) and rebuild the usb-storage driver.

For testing, use only one of the USB drives; unplug the other one. The usb-storage debugging option will generate a _lot_ of debugging messages, so make sure your /etc/syslog.conf is set up not to store debug-level messages from the kernel. Then go ahead and start writing a large amount of data to the drive. When you see those resets start to occur, run dmesg and post a copy of the output.

Alan Stern


Am I really getting insane? I recompiled kernel (and upgraded to 2.6.15-gentoo-r3) but that didn't help either. I've tested cables, and so on. Still nothing, still have errors and the slow.

Debug messages say the same things as about 4 years ago when I was happily checking out first usb mass storage driver in kernel (then I was trying home-made usb disk made with Seagate 4GB and interface based on Cypress chip, but instead of working I had my machine hung). During my visit in servers' room I did nothing but switched USB cables each other, and that partialy helped - I successfully managed to copy 57GB of data between 2 USB disk with only 11 such errors. Md5 sums are correct.

But by my mistake (I've got Big Blue machine which can't boot from scsi but only floppy) I booted to previous kernel, so the solution was to switch USB cables each other? I thought I'm experienced in hardware, but I'm surprised.

Neverthless I still think that the issue of poorly written usb driver in the kernel should be considered. I use about 7 of IBM 206 and 226 machines and on every Linux one I found the same problem using all kind (long, short, old, new, expensive and cheap) of cables. Under Windows there is no such problem at all!

Maybe usb driver needs some throttling when a cheap device does not meet all specific to USB requirements? I can understand that resetting high speed device performed by kernel could be perceived as a solution but it is way too rough for me.




-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems?  Stop!  Download the new AJAX search engine that makes
searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642
_______________________________________________
Linux-usb-users@lists.sourceforge.net
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-users

Reply via email to