On Wed, Feb 27, 2008 at 8:07 PM, Volker Lendecke <[EMAIL PROTECTED]> wrote:
> On Wed, Feb 27, 2008 at 07:43:42PM +0100, Alex Still wrote: > > I think I found it. > > Samba-3.0.28 calls set_filetime() from real_write_file(), which 3.0.8isn't > > doing. > > set_filetime -> utimes -> nfs SETATTR (in my case) > > After removing that bit from real_write_file, I get exactly the same > > performance I had before. Now, I realise this has probably been put > there > > for a reason... > > > > Stuck now, any advice on the matter appreciated > > It was put there to properly support the so-called sticky > write time feature that for example Excel depends upon. > Stefan Metzmacher has written a patch that needs to be > merged that avoids those utime calls by putting the relevant > data into locking.tdb. We need to take some time to shape up > that patch a bit and merge upstream though. Many thanks for that explanation. I just read about that sticky write time "feature" in www.nasconf.com/pres05/allison.pdf , nice quirk in the protocol ! I think I will still have a performance hit once the patch is in tho, because samba 3.0.8, in addition to not supporting the "sticky write time", doesnt seem to call set_filetime at all within real_write_file() . This is introduced in samba 3.0.11 I guess the real fix would be to stop serving NFS shares. Best regards, -- Alex -- To unsubscribe from this list go to the following URL and read the instructions: https://lists.samba.org/mailman/listinfo/samba
