smbmount read/write block size stuck at 4k

[Dag Wieers]

On Wed, 8 Dec 2004, Dan Singletary wrote:

> When using smbmount to mount a share from WinXP, I observed
> less-than-optimal throughput when copying a file from the remote share
> to a local disk.  Observation of the protocol dialog with Ethereal
> revealed that smbmount was making read requests for 4k bytes at a time.
>  Similar observation of protocol dialog between two WinXP boxes showed
> that the typical read request was for 64k.
> 
> I then tried to use smbclient to 'get' the file, and again observed 4k
> SMB read requests.  I found that using the '-b <buffersize>' option with
> smbclient is effective in REDUCING the buffer size as small as 1 byte
> (reflected in Ethereal showing 1 byte SMB read requests), however I am
> unable to use '-b' to increase the read requests beyond 4k.
> 
> Why does smbmount/smbclient only read  data at 4k per block?  This
> results in horribly poor performance (~4 MBytes/sec) on a gigabit
> network that should permit transfer rates that challenge the speed of
> the disk drives involved...

Yes, I've reported the same problem a week ago. Both the smb client as the 
cifs client have this problem, on 2.4 and 2.6 kernels. I know Steve French 
is working on it since recently. You may have better luck in getting a 
technical explanation from the linux-cifs-client ml.

Even though we all need a solution right now, this thread looks promising:

	http://lists.samba.org/archive/linux-cifs-client/2004-December/000543.html

Kind regards,
--   dag wieers,  dag at wieers.com,  http://dag.wieers.com/   --
[all I want is a warm bed and a kind word and unlimited power]