Re: [linux-usb-devel] 2.6.0-test1: random errors for USB disk

Re: [linux-usb-devel] 2.6.0-test1: random errors for USB disk

[Alan Stern]

On Thu, 17 Jul 2003, Harald Dunkel wrote:

> Harald Dunkel wrote:
> > Alan Stern wrote:
> > 
> >>
> >> Did you say that this worked correctly under 2.4.22?  If it does, 
> >> could you send the equivalent system log (with usb-storage debugging 
> >> enabled) showing what happens when you create a filesystem and 
> >> everything works right?
> >>
> > The rebuild is running. But the last working version I have is 2.4.21
> > plus the USB patches of June 20th. Is it OK to start with this version?
> > 
> 
> See attachment.

The information in your system logs is very clear.  Your USB drive worked 
just great until it received a big data transfer.

Under 2.4, the largest WRITE transfer was 130048 bytes and it worked fine.  
Under 2.6, the largest attempted WRITE transfer was 524288 bytes and it
crashed the drive.  The two commands were otherwise identical.

Presumably differences in the block I/O systems account for the difference
in buffer sizes.  But I don't know of any way to tell Linux that the drive
is unable to accept data transfers larger than some fixed limit.

Can anyone else suggest a way to solve this?

Alan Stern

Re: [linux-usb-devel] 2.6.0-test1: random errors for USB disk

[David Brownell]

Alan Stern wrote:

> The information in your system logs is very clear.  Your USB drive worked 
> just great until it received a big data transfer.
> 
> Under 2.4, the largest WRITE transfer was 130048 bytes and it worked fine.  
> Under 2.6, the largest attempted WRITE transfer was 524288 bytes and it
> crashed the drive.  The two commands were otherwise identical.

Just to clarify:  there's another difference, and that's that 2.4 does
doing that smaller write one page at a time (write, wait, write, wait,...)
while 2.6 does it all at once (write, write, write, write, ... wait).

That difference is how I've seen usb-storage top 30 MByte/sec on USB
with 2.6 kernels, when 2.4 may not reach 10 MB/sec on the same hardware.

- Dave


> Presumably differences in the block I/O systems account for the difference
> in buffer sizes.  But I don't know of any way to tell Linux that the drive
> is unable to accept data transfers larger than some fixed limit.
> 
> Can anyone else suggest a way to solve this?
> 
> Alan Stern
> 
> 

Re: [linux-usb-devel] 2.6.0-test1: random errors for USB disk

[Harald Dunkel]

David Brownell wrote:
> Alan Stern wrote:
> 
>> The information in your system logs is very clear.  Your USB drive 
>> worked just great until it received a big data transfer.
>>
>> Under 2.4, the largest WRITE transfer was 130048 bytes and it worked 
>> fine.  Under 2.6, the largest attempted WRITE transfer was 524288 
>> bytes and it
>> crashed the drive.  The two commands were otherwise identical.
> 
> 
> Just to clarify:  there's another difference, and that's that 2.4 does
> doing that smaller write one page at a time (write, wait, write, wait,...)
> while 2.6 does it all at once (write, write, write, write, ... wait).
> 
> That difference is how I've seen usb-storage top 30 MByte/sec on USB
> with 2.6 kernels, when 2.4 may not reach 10 MB/sec on the same hardware.
> 
I am using the disk in question to do backups of my $HOME. On 2.4.21
it was written with an amazing speed of about 20 MByte/sec, which I had
never expected. I could live with 10, if it is reliable.


Regards

Harri