Thanks for your reply Tim. If anything it makes me feel worse. I was
hoping it was something easily fixed.
I just tried transferring a 50 Mb file to the OBSD samba box from win
using SCP. Again very slow writes but much faster reads. The 50 Mb file
took about 7 mins to transfer to the OBSD box and about 30 seconds to
read from the OBSD box.
Perhaps this isn't a samba smb issue at all.
My fstab...
# cat /etc/fstab
/dev/wd0a / ffs rw 1 1
/dev/wd1a /data1 ffs rw 1 2
/dev/wd2a /data2 ffs rw 1 2
same result with either data disk. I've been googling all evening and
found many many forum posts with similar problems but no solutions. Some
posts date back to 2002!
If I have to go back to RH7.3 I'll be bummed. Especially as I spent ages
setting up all my families accounts and softlinks for the data store.
Waste of a day!
Tim Hammerquist wrote:
Gary Clemans-Gibbon wrote:
David Gwynne wrote:
Gary Clemans-Gibbon wrote:
Everything is working fine except that when I copy files to the
box from a Windows XP box the transfers are very slow, like
9 minutes for a 48 Mb file. Copying the same file back to the win
box is quick - a couple of seconds as you'd expect.
I would suggest looking at the socket options parameter in
/etc/samba/ smb.conf. I have the following in my smb.conf and
transfer speeds seem to perform a lot better now:
socket options = TCP_NODELAY SO_RCVBUF=8192 SO_SNDBUF=8192
I just tried that line but it seems to be the same or if anything it
seems even slower.
Gary,
I've seen this same phenomenon when copying to from my OSX Powerbook and
my fileserver (running both FreeBSD 5 and Gentoo Linux), with the OSX
acting as samba client.
The transfer speeds are not "slightly" slower, they are slower by orders
of magnitude, with normally 20sec transfers taking 10-20 minutes.
I watch the progress meter slowly incrementing at the rate of 32-64k/sec
over a 100bTX link. Does this sound like your issue?
In my setup, I had limited success merely unmounting and remounting the
share; that worked maybe 50% of the time. Also, the rate seemed to be
normal more often if I had a simultaneous ssh connection between the
same two machines, even if the ssh connection were idle. I was not able
to find any consistently effective solution.
After googling many times over several months, finding nothing more than
the same advice you got about TCP_NODELAY and the SO_*BUF settings
(which did not affect performance in my case either), I finally gave up,
switching to NFS and/or scp.
For what it's worth, I haven't noticed this since I upgraded my
powerbook to OSX 10.4, so it might have something to do with the client
OS, network stack, or Samba version.
I apologize for not having anything solid to recommend. But I wanted to
let you know that this *has* happened to others; you're not imagining
it.
Tim Hammerquist
.