I'm still on 2.6.18-3-ixp4xx. Yesterday, I have had I/O errors over
samba. These were my first samba incidents, but then again I just
started using samba recently. I have been trying to reproduce the error
with samba with no luck.
My conclusion is that errors occur much less frequently with samba than
with nfs. The first explanation that comes to my mind is that samba puts
less load on the network than nfs does, as samba transfers at about 40%
the rate of nfs, thus supporting the 'ethernet driver bad behavior under
load' argument.
This also explains why rdiff-backup did not result in any errors
although it managed to transfer files nfs used to corrupt -- because it
transfers at a much lower speed.
Probably someone should try transfers over http or ftp to confirm it's
not specific to nfs as they are capable of transferring at similarly
high speeds.
This leaves the issue of other kernels. To be honest, I cannot rely on
my memory. I am sure I have had nfs trouble with all the configurations
I ran (d-i beta2, debootstrapped slugOS kernel, and 2.6.18-3-ixp4xx),
but cannot confirm the 'reproducibility'/frequency and the type of the
errors at this time on the old configurations. As far as I remember, d-i
beta2 did not have the nfs kernel modules, so it probably was the
nfs-user-server only. On debootstrapped slugOS kernel though, I
installed the nfs kernel modules (ipkg files) and ran into some trouble
with nfs too, but I cannot be more specific on that. I have not tried
Debian's 2.6.17 at any time, as I upgraded from slugOS kernel to
Debian's 2.6.18.
- Ahmad
Martin Michlmayr wrote:
* Martin Michlmayr <[EMAIL PROTECTED]> [2006-12-29 22:36]:
That's odd. I tried Debian's 2.6.17 kernel and didn't see any
problems, I think. I'll try again.
I copied a 840 MB file 100 times over NFS this night with 2.6.17
without any problems.