[Bridge] slow network performance when using bridged interfaces in 2.6.13 compared to 2.6.12.

[Stephen Hemminger]

On Fri, 28 Oct 2005 21:33:39 +0000
"Josh Burke" <kernelbridgeissue at iokui.com> wrote:

> (originally filed as a bug in Fedora's bugzilla, see  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=171933)
> 
> Greetings,
> 
> Using Fedora Core 4 on a Dell PE 420sc.  Malfunctioning kernel is smp-2.6.13-1.1532_FC4.  Properly functioning kernels included smp-2.6.12-1.1456_FC4.
> 
> Network performance is extremely poor when using bridged network interfaces.  When not using brctl, the interface gives ~800kbps.  When using brctl to bridge two interfaces, the speed drops to ~1.0kbps.
> 
> Reverting to kernel-smp-2.6.12-1.1456_FC4 returns the speed to normal.
> 
> The two interfaces have IP addresses when not bridged.  They do not have IP addresses when they are bridged.  (As expected.)
> 
> No unusual output in dmesg.  
> 
> Traffic is not slowed that originates on a bridge interfaces and goes out a bridge interface, only traffic from the host that the bridge is on, is slowed.
> 
> Making these connections unbridged returns normal traffic performance.  Reverting to 2.6.12 also returns normal performance.
> 
> Tcpdump indicates increasing pauses returning packets. (From bridge host to the outside.  Bridge host is a web server, though even SSH connections are slow.)
> 
> For example:
> 14:25:53.387001 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:25:53.648768 IP 10.0.0.1.32752 > 192.168.0.1.https: tcp 0
> 14:25:53.648817 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:25:53.648859 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:25:53.770239 IP 10.0.0.1.32752 > 192.168.0.1.https: tcp 0
> 14:26:03.242674 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:26:03.462594 IP 10.0.0.1.32752 > 192.168.0.1.https: tcp 0
> 14:26:03.462675 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:26:03.462714 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:26:03.581169 IP 10.0.0.1.32752 > 192.168.0.1.https: tcp 0
> 14:26:22.526025 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:26:22.784218 IP 10.0.0.1.32752 > 192.168.0.1.https: tcp 0
> 14:26:22.784325 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:26:22.784379 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:26:22.895638 IP 10.0.0.1.32752 > 192.168.0.1.https: tcp 0
> 14:27:00.780762 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:27:01.045112 IP 10.0.0.1.32752 > 192.168.0.1.https: tcp 0
> 14:27:01.045209 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:27:01.045263 IP 192.168.0.1.https > 10.0.0.1.32752: tcp 1260
> 14:27:01.160083 IP 10.0.0.1.32752 > 192.168.0.1.https: tcp 0
> (192.168.0.1 = bridge host, 10.0.0.1 = external host.  This happened while trying to download a text file.)
> 
> Any suggestions, or places to look for information, is appreciated.

What kind of ethernet interfaces. The bridge code changes have been very
small.  I would suspect the TSO changes in the base stack are causing
problems.  Try turning TSO off on the interfaces.

Another possibility is netfilter.

-- 
Stephen Hemminger <shemminger at osdl.org>
OSDL http://developer.osdl.org/~shemminger