[Samba] About SOCKET options
Ok, as some of you read I was the guy who experimented with SOCKET options. In my last post, where I figured that I needed to use the right syntax I found out that SO_SNDBUF=4096 and SO_RCVBUF=4096 worked way better than with 8192 as value. Today I verified this behavior. I used SO_SNDBUF=4096 and SO_RCVBUF=4096 and copied a 780 MB file from Windows to Linux. ETA of Windows was 4 mins and it did indeed happen within 4 mins (13:58-14:02). I switched to SO_SNDBUF=8192 and SO_RCVBUF=8192, rebooted (just to be sure), and copied the same file again. ETA of Windows started at 45 mins and rose up to 101 mins. I waited for 2 mins and it was copying the hell slow. So I assume it would indeed take around 80-100 mins to copy the same file. I aborted. Swichted back to 4096 et voila, was at 4 mins again. Checking the Internet for socket options it does say that windows SO_RCVBUF and SO_SNDBUF are at 16k on default. You can increase the number but there is a point where the results do get worse again, e.g. picking too high or too low a number. I assume the same is true for the Linux system, since the socket stuff belongs to the network protocol and is not OS dependant. My physical network configuration is: Windows XP on an AMD XP2000+ Intel Ethernet Express 100b Debian Linux on AMD Athlon 650 3Com 905C-TX on a switched 100 MBit Network. (via a hardware SMC Barricade broadband router). Can anyone explain this weird behavior to me? Especially since almost all internet sites recommend 8192 nowadays and not 4096. Why should 4096 be faster for me? Am I a rare case (since everyone seems to recommend 8192)? -- Benjamin Weber -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
Re: [Samba] About SOCKET options
Le jeu 26/09/2002 à 14:47, Benjamin Weber a écrit : > Ok, as some of you read I was the guy who experimented with SOCKET options. > > In my last post, where I figured that I needed to use the right syntax I > found out that SO_SNDBUF=4096 and SO_RCVBUF=4096 worked way better than with > 8192 as value. Today I verified this behavior. > so did I! 4069 is much faster here (Win XP pro for the clients, 2.4.19(gentoo) kernel, samba 2.2.5. David -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
Re: [Samba] About SOCKET options
On Thu, 2002-09-26 at 10:50, David Morel wrote: > Le jeu 26/09/2002 à 14:47, Benjamin Weber a écrit : > > Ok, as some of you read I was the guy who experimented with SOCKET options. > > > > In my last post, where I figured that I needed to use the right syntax I > > found out that SO_SNDBUF=4096 and SO_RCVBUF=4096 worked way better than with > > 8192 as value. Today I verified this behavior. > > > > so did I! 4069 is much faster here (Win XP pro for the clients, > 2.4.19(gentoo) kernel, samba 2.2.5. Are either of you using "large readwrite = yes" as well? -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
AW: [Samba] About SOCKET options
Yep I am using "large readwrite=yes" as it is set on by default. That might have anything to do with it? -- Benjamin -Ursprüngliche Nachricht- Von: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]Im Auftrag von Chris Smith Gesendet: Freitag, 27. September 2002 19:05 An: [EMAIL PROTECTED] Betreff: Re: [Samba] About SOCKET options On Thu, 2002-09-26 at 10:50, David Morel wrote: > Le jeu 26/09/2002 à 14:47, Benjamin Weber a écrit : > > Ok, as some of you read I was the guy who experimented with SOCKET options. > > > > In my last post, where I figured that I needed to use the right syntax I > > found out that SO_SNDBUF=4096 and SO_RCVBUF=4096 worked way better than with > > 8192 as value. Today I verified this behavior. > > > > so did I! 4069 is much faster here (Win XP pro for the clients, > 2.4.19(gentoo) kernel, samba 2.2.5. Are either of you using "large readwrite = yes" as well? -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
Re: AW: [Samba] About SOCKET options
On Fri, 2002-09-27 at 13:24, Benjamin Weber wrote: > Yep I am using "large readwrite=yes" as it is set on by default. That might > have anything to do with it? Not sure. I didn't think that was the default, maybe I'm mistaken or that it changed at a certain release. -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
Re: AW: [Samba] About SOCKET options
On Fri, 2002-09-27 at 13:24, Benjamin Weber wrote: > Yep I am using "large readwrite=yes" as it is set on by default. That might > have anything to do with it? Just checked my docs for 2.2.5 claim that the default is "large readwrite = no". Maybe it's just the way your distribution set up the smb.conf. -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
AW: AW: [Samba] About SOCKET options
My samba help file says: large readwrite (G) This parameter determines wheather or not smbd supports [...]. [...] Can improve performance by 10% with Windows 2000 clients. Defaults to on. Not as tested as some other Samba code paths. Default: large readwrite=yes I am using Samba 2.999+30cvs, directly from the cvs tree it seems. You guys are putting it into the debian testing/unstable tree pretty fast. But I am not sure that this must be related to my "discovery". The socket values of 4096 and 8192 respectivly are quite distant form the large readwrite streaming value of 64k. Anyy ideas? I got emails from 2 list users who will try and run benchmarks on their M$ systems with 4096 and 8192 just to find out if they can reproduce the weird transfer behavior I experienced. Not sure of the results, but I am quite curious of course. Anyone else encountered higher transfer speeds when using 4096 instead of 8192 for the SO_SNDBUF and SO_RCVBUF? Maybe its a Samba 3.0 thing. -- Benjamin -Ursprüngliche Nachricht- Von: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]Im Auftrag von Chris Smith Gesendet: Freitag, 27. September 2002 19:58 An: [EMAIL PROTECTED] Betreff: Re: AW: [Samba] About SOCKET options On Fri, 2002-09-27 at 13:24, Benjamin Weber wrote: > Yep I am using "large readwrite=yes" as it is set on by default. That might > have anything to do with it? Just checked my docs for 2.2.5 claim that the default is "large readwrite = no". Maybe it's just the way your distribution set up the smb.conf. -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
Re: AW: AW: [Samba] About SOCKET options
On Fri, 2002-09-27 at 15:15, Benjamin Weber wrote: > My samba help file says: > > large readwrite (G) > > This parameter determines wheather or not smbd supports [...]. > [...] > Can improve performance by 10% with Windows 2000 clients. Defaults to on. > Not as tested as some other Samba codepaths. > > Default: large readwrite=yes > > I am using Samba 2.999+30cvs, directly from the cvs tree it seems. You guys > are putting it into the debian testing/unstable tree pretty fast. Ah, so it's a change in defaults for possible future releases. I have been running with it enabled. But I have also seen no performance issues with the 8192 buffer sizes. -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
