----- Forwarded message from Tobias Göller <tob...@brabbel.ch> -----
Date: Thu, 23 Jul 2009 11:47:38 +0200 From: Tobias Göller <tob...@brabbel.ch> To: Christian Perrier <bubu...@debian.org> Subject: Re: Bug#538079: [Pkg-samba-maint] Bug#538079: samba: Dead Slow READ performance - WRITE Performance OK X-Mailer: Apple Mail (2.935.3) X-CRM114-Status: Good ( pR: 27.0851 ) Hi Christian, When I found this performance problem I was playing a bit with the socket options... as you already mentioned: no performance impact what so ever. It wasn't there before - I just copy and pased the files into the mail... Fun thing is that when I write to the server the performance figures are very good - up to 500MBit/s over GBit channel - write. But read (and only read) is dreadfully slow. I have a switched network (layer 3 switches) and have tried about everything (duplex, flow-contrl, etc) - without luck so far. NFS works like a charm (even from my Mac Systems). CU Tobias On 23.07.2009, at 11:44, Christian Perrier wrote: > Quoting Steve Langasek (vor...@debian.org): > >> These are all part of the *stock* smb.conf that we ship in Debian >> and will >> have no impact on file sharing speed. > > I agree in general but wanted to reduce the investigation as much as > possible. I don't think that problems come from the smb.conf settings > (except maybe "socket options"). > >>> socket options = TCP_NODELAY SO_RCVBUF=8192 SO_SNDBUF=8192 >> >> This could have a performance impact. > > Agreed. And, according to upstream, there is no reason to use that > anymore. > > >> >>> winbind enum users = Yes >>> winbind enum groups = Yes >> >> These could have a performance impact, but should not in the >> context of file >> serving. > > > Yep. Agreed too. > > ----- End forwarded message ----- --
signature.asc
Description: Digital signature
