Add this to your /etc/sysctl.conf:
#Temporary workaround for tcp issue with 2.6.27
net.ipv4.tcp_timestamps = 0
EDIT:
I can't find the patch in http://www.kernel.org/pub/linux/kernel/ … g-2.6.27.4 so I assume it will be in the next release.
However, it is still a work-around and on the patched kernel it is not necessary to turn it of. I doubt that you will ever see any differences when you just use this option and forget about it.
I think when you have to disable TCP_SACK with your hardware, this can possibly have a small performance impact. Nothing you would care of as a normal user I guess.
I would personally recommend setting on of these options if it helps, rather than patching the kernel, but I just not installing the new kernel for now and waiting for the fixed one is also an option. This is the way I am going, because I don't need the new kernel today. It's just that I will run into problems on arch when I update everything except the kernel, so I need to have a working 2.6.27 kernel somewhen.
]]>handy wrote:What patch?
http://bugzilla.kernel.org/show_bug.cgi?id=11721#c58 (listed some hours after your post ;-) )
This will be included in 2.6.27.x (with x beeing quite close to 2 i hope)
So either apply the patch or just wait for the next release. I have no time for a long compile so I just wait. Didn't test it myself, but it seems to relate very closely to my problem, referring from the comments and the problem (ordering of tcp options in the packet)
Thanks for the info'.
As it stands my system is working as well as it ever was, on the current kernel after taking the following actions:
Add this to your /etc/sysctl.conf:
#Temporary workaround for tcp issue with 2.6.27
net.ipv4.tcp_timestamps = 0
So what ill effect does turning off timestamps have on tcp?
]]>What patch?
http://bugzilla.kernel.org/show_bug.cgi?id=11721#c58 (listed some hours after your post ;-) )
This will be included in 2.6.27.x (with x beeing quite close to 2 i hope)
So either apply the patch or just wait for the next release. I have no time for a long compile so I just wait. Didn't test it myself, but it seems to relate very closely to my problem, referring from the comments and the problem (ordering of tcp options in the packet)
]]>http://lwn.net/Articles/92727/
One time fix as root run:
sysctl -w net.ipv4.tcp_window_scaling=0
To make it persistent across reboots:
As root edit /etc/sysctl.conf
Add a line as follows:
net.ipv4.tcp_window_scaling=0
cheers Barde
]]>Great work handy.
Add this to your /etc/sysctl.conf:
#Temporary workaround for tcp issue with 2.6.27 net.ipv4.tcp_timestamps = 0
I wondered whether the above command would be permanent.
Thanks for that info' I'll add it to the new thread:
]]>Add this to your /etc/sysctl.conf:
#Temporary workaround for tcp issue with 2.6.27
net.ipv4.tcp_timestamps = 0
Anyway I read the bug report that Misfit138 posted a link to above & after using su I entered the following at the root prompt:
echo 0 > /proc/sys/net/ipv4/tcp_timestamps
I am able to use the current kernel, & my internet problems are no more!
I have no idea what that command does, but it came up as a worker in the bug report below:
]]>thank you very much. i will try this out as soon as i get home. i'll file a bug at the main kernel bugtracker, that's why i wanted more information.
cheers Barde
Thanks for doing that.
]]>