Kernel updated to 2.6.7, browsing almost impossible

[phantom]

Hi guys.
Two days ago, my linux box updated (up2date) the kernel from 2.6.6 to 2.6.7. After that, I have been experiencing some problems. The first one that I noticed was that browsing is very slow for some sites. I haven't found a pattern. For instance, if I go to google, it works fine, but going to other places is very slow (like this one) and some others, almost impossible (www.mural.com). I have checked with ping and traceroute, and I cannot see any difference when i do the same running the old kernel. For the old kernel, I had deactivated ipv6, as recommended by someone in the forum. This is my current /etc/modprobe.conf:
alias eth0 e1000
alias scsi_hostadapter ata_piix
alias snd-card-0 snd-intel8x0
install snd-intel8x0 /sbin/modprobe --ignore-install snd-intel8x0 && /usr/sbin/alsactl restore >/dev/null 2>&1 || :
remove snd-intel8x0 { /usr/sbin/alsactl store >/dev/null 2>&1 || : ; }; /sbin/modprobe -r --ignore-remove snd-intel8x0
alias usb-controller ehci-hcd
alias usb-controller1 uhci-hcd
alias net-pf-10 off
alias ipv6 off
Now, up2date also wants to update gaim and gimp, but the first time when actually running the up2date, it said that everything was ok and that there was no need to update anything. Next time I loggin, it again said that it had to update the same packages, but stayed there until I killed the process. :rolleyes:

Has anybody experienced similar problems? Any suggestions in what to do? ( :confused: )

For the moment, I think that I will continue using the old kernel, but I guess that the new one should eventually work.
Thanks for your help!

[ats-tech]

I've experienced the same thing. I'd had to kill up2date several times because it appears to "lock up". Generally I've found that when upd2date sayst aht packages need to be updated, and then when you run it, says everything is fine... just wait a few minutes, or run it via the cli (up2date -u) generally works.

[blammo]

Hi guys.
Two days ago, my linux box updated (up2date) the kernel from 2.6.6 to 2.6.7. After that, I have been experiencing some problems. The first one that I noticed was that browsing is very slow for some sites.

I've seen a couple posts in this forum with some ethernet cards poofing out with the new 2.6.7 kernel. This might be the cause of your slow browsing, but then again it might be your ISP! From your post I'm understanding that when you go back to the old kernel, everything is ok... am I right? As for up2date, I'd dump that monstrosity in favor of yum. I've had plenty of trouble using up2date and all that vanished when I switched to yum for updating my system.

[phantom]

Thanks ats-tech and blammo.
Actually, the up2date thing is not what really worries me, but the access to the network, since I cannot really work comfortably. You know, even when you are programming, or using the PC as a standalone workstation, some times it is required to access an on-line reference or so. And as of now, I cannot do it. My guess is that the new kernel has some kind of internal ipv6 or something like that. Eventhough, when I was having problems in the 2.6.6 kernel, the browser was delayed when resolving IPs. I disabled ipv6 and everything went fine. Now, in kernel 2.6.7, it basically stops in 'Transferring from...' Sometimes, it displays part of the page, but some parts are missing, and continue like that forever. For instance, when trying to connect to www.hotmail.com, it downloads parts of the page, but stops transferring info from cb.msn.com or something like that.
:p

[Micha_Silver]

I'm experiencing the same problem as Phantom with browsing (only!) in the 2.6.7 kernel. The browser gets "stuck" when retrieving any page. i.e. the tcp connection is opened, but nothing arrives. This started after updating to the newer kernel with yum. Other network services - retrieving mail, ssh etc are all working fine. But nothing works going to port 80. Not even lynx. And when I reboot into the older 2.6.6 kernel browsing returns to normal. So I'm quite sure the problem is in the kernel.

I tried to follow the packet exchange with tcpdump when trying to access a web page. The initial tcp exchanges where there, going and coming, but, as I said above, once the tcp connection opened, nothing followed. The browser just sits there waiting for pages from a site it knows it found, but nothing arrives.

BTW, I'm on a thinkpad T41, Fedora Core 2, pretty standard setup. The only non-rpm I put in so far was Thunderbird. And that, as I said, works fine in both kernels.
Hope someone can put a finger in this. It's quite disturbing.

--Micha

[phantom]

Hi guys. I have bad news. It seems that the problem is still there in the new kernel (2.6.8). So, I guess that I will continue using the old 2.6.6. I hope that somebody can help us to solve this :confused:

[Ned]

I'm using the 2.6.7 kernel and don't experience the browsing problems you've described. I tried the problematic websites mentioned above and all work fine for me. Not much help I know, sorry, but maybe at least it's removed any doubts about the websites from your equation.

Ned

[benno]

I'm having the same problem since kernel 2.6.7. Today I updated to 2.6.8, same problem still there. So I switched back to 2.6.6 which works ok. I have an Intel EtherXpress PRO 100 onboard in my Compaq DeskPro workstation. One of the most common network cards ever.

[ssaady]

I experienced pretty much the same symptoms, slow browsing, yum, and POP3 e-mail. I took the NIC out, boot, let Kudzu remove the configuration, shutdown, ad the card back (making sure it was not a shared PCI/ISA slot, let Kudzu configure it, off to the races.

[Prometheus]

mine goes kind of slow, but not agonizingly so. More like, loading, loading, loading ,then poof, the whole page loads at once in about half a sec. Kind of annoying, but i can live with it. Thing with me though, is my card is onboard, so i cant just pop it out like ssaady did... but it works, thats the most important thing

[cricket]

Kernels 2.6.7 and 2.6.8 come with changes to the tcp window size that (should) increase the efficiency of tcp. What it HAS done is expose a problem with the way some routers and firewalls handle requests for large windows. Those routers and firewalls just refuse to handle them; which is why you may be able to access some sites, but not others.

The quick fix is to su to root and do two things:

#echo "net.ipv4.tcp_moderate_rcvbuf=0" >> /etc/sysctl.conf
#echo "net.ipv4.tcp_default_win_scale=0" >> /etc/sysctl.conf

..and then reboot. Alternately, you can just issue:

#/sbin/sysctl -w net.ipv4.tcp_moderate_rcvbuf=0
#/sbin/sysctl -w net.ipv4.tcp_default_win_scale=0

...and save yourself the reboot, but your settings will be lost when you DO reboot.

-cricket

[Micha_Silver]

Cricket's suggestions was right on target. Fixed my problem.
What I had experienced was a total stalling of all tcp traffic (udp worked fine. i.e. name resolution, and vpn traffic that was encapsulated in udp packets) when working thru my new ADSL router.
But when I took the laptop to another network everything was OK. So I was sure the problem was some mismatch between the ADSL router and the 2.6.7 kernel.
With the two changes Cricket suggested to kernel params:
tcp_moderate_rcvbuff=0 and tcp_default_win_scale=0
browsing and tcp traffic is back to normal.

Many thanks, and keep chirping...

[phantom]

Hey cricket!!
Thanks for the help. Now it works perfectly!!!
:D

[benno]

Thanks cricket. This fixed it indeed.

[smidoid]

Cricket said:

The quick fix is to su to root and do two things:

#echo "net.ipv4.tcp_moderate_rcvbuf=0" >> /etc/sysctl.conf
#echo "net.ipv4.tcp_default_win_scale=0" >> /etc/sysctl.conf


I'm totally confused (but impressed) how did you figure that one out?

And what does it do?

Smid (who should have used a quotation)!
____________
Experts are not born, they are hewn from the granite of experience and the bedrock of hard work :D

[jkj5301]

[QUOTE=cricket
The quick fix is to su to root and do two things:

#echo "net.ipv4.tcp_moderate_rcvbuf=0" >> /etc/sysctl.conf
#echo "net.ipv4.tcp_default_win_scale=0" >> /etc/sysctl.conf

..and then reboot. Alternately, you can just issue:

#/sbin/sysctl -w net.ipv4.tcp_moderate_rcvbuf=0
#/sbin/sysctl -w net.ipv4.tcp_default_win_scale=0


-cricket[/QUOTE]

I'm having the same problems with web browsing. Some sites work fine, some - including Yahoo and Google - are contacted but time out without displaying anything. Everything works with kernel 2.4.27. I get the same behavior with different network cards, so the problem doesn't appear to be the NIC or the driver.

Cricket's explanation is the most plausible I've found, but the solution above doesn't work for me. This seems to be a serious bug in the 2.6 kernel -- I wonder if there are any other solutions forthcoming?

[jcoles]

Kernels 2.6.7 and 2.6.8 come with changes to the tcp window size that (should) increase the efficiency of tcp. What it HAS done is expose a problem with the way some routers and firewalls handle requests for large windows. Those routers and firewalls just refuse to handle them; which is why you may be able to access some sites, but not others.

The quick fix is to su to root and do two things:

#echo "net.ipv4.tcp_moderate_rcvbuf=0" >> /etc/sysctl.conf
#echo "net.ipv4.tcp_default_win_scale=0" >> /etc/sysctl.conf

..and then reboot. Alternately, you can just issue:

#/sbin/sysctl -w net.ipv4.tcp_moderate_rcvbuf=0
#/sbin/sysctl -w net.ipv4.tcp_default_win_scale=0

...and save yourself the reboot, but your settings will be lost when you DO reboot.

-cricket
Is it necessary to restart firefox after issuing the commands?
(If they work, I'll add them to sysctl.conf.)
The commands didn't help, but I kept firefox open on this forum page.

The second command returns an error:
# /sbin/sysctl -w net.ipv4.tcp_default_win_scale=0
error: "net.ipv4.tcp_default_win_scale" is an unknown key

Is there a typo in the command?

Thanks.

[jcoles]

On FC5, the second command
# /sbin/sysctl -w net.ipv4.tcp_default_win_scale=0
is
# sbin/sysctl -w net.ipv4.tcp_window_scaling=0

This solves the problem. Thanks, cricket.

[new-user]

Hi All,
I am using Fedora 7 and having the same problem. anyone knows what is the equivallent command to the following command for FC 5:
# sbin/sysctl -w net.ipv4.tcp_window_scaling=0

thanks,