Description of problem: after kernel update the internet performance using squid is very bad. Pictures don't load, sometimes text don't load either. When I configure any browser to connect directly to the internet the performance is good. Same performance problem with getting email with evolution+spamassasin. I am using ppp and a dsl modem to connect to the internet. Version-Release number of selected component (if applicable): kernel-2.6.16-1.2107_FC5.i686 evolution-2.6.1-1.fc5.2 squid-2.5.STABLE13-1.FC5 spamassassin-3.1.1-1.fc5 How reproducible: allways. when I boot kernel-2.6.16-1.2096_FC5 everything is OK Steps to Reproduce: 1. 2. 3. Actual results: Expected results: Additional info:
update: When I uncheck external tests from the spam section in evolotion, recieving email is fast. There was no problem with external tests with 2.6.16-1.2096_FC5
The problem with Squid is related to the object size. Smaller objects (max. 1-2 kB) are okay, but larger objects won't work (like images or web pages > 20 kB). Objects are loaded successfully into the cache (this can be verified with ethereal and by looking into /var/spool/squid), but the objects are not correctly sent to the calling application. It's not the fault of the web browser, because "telnet localhost 3128" shows the problem as well. With the previous kernel 2.6.16-1.2096_FC5, everything is fine. "sysctl -a" shows no significant differences of kernel settings. There might be more applications affected by this, but Squid is a very obvious one and easy to reproduce. Don't know if this is a problem in the kernel, or in certain applications (like Squid) that is just triggered by the kernel update.
does booting with pci=nomsi make this go away?
Actually, that option is also broken in this kernel (sigh). 2108, available from http://people.redhat.com/davej/kernels/Fedora/FC5/ has this disabled by default. Give that a try ?
it doesn't work with 2108 either
With this kernel 2.6.16-1.2107 fetchmail cannot fetch my mail. I returned to kernel 2.6.16-1.2096 and I could then get my mail as usual.
Same experience as Martin Dnald; fetchmail to exim to procmail/spamassassin mail chain, working for years, gets through the first email download (pop3 protocol) and hangs on 2006-05-03 21:25:50 SMTP data timeout (message abandoned) on connection from +localhost (grithr.localdomain) Reverting to 2096 and the problem goes away. Didn't get anything in the exim logs, mail log, procmail log, or /var/log/messages.
Same here, but for me kernel-2.6.16-1.2111_FC5 fixed the problem.
yes, kernel-2.6.16-1.2111_FC5 fixed it
I'm seeing this exact same problem with squid and 2.6.17-1.2139_FC5smp. Some objects it loads fine and fast, others take forever or come up broken. Doesn't matter if I browse from firefox/linux or IE/windows. If I turn off proxy use in the browser, browsing is very fast. I thought it was my upstream proxy server having a cow, but I changed squid to always use DIRECT and the problem remains. The problem started the very first day I installed FC5. My setup ran perfectly fine under FC3 right before that. Any ideas?
Have you tried putting "net.ipv4.tcp_window_scaling = 0" in "/etc/sysctl.conf" (reboot required) or running "sysctl -w net.ipv4.tcp_window_scaling=0" (not for existing TCP connections, only for new ones from now on)? Just a wild guess, and not related to the original problem of this ticket. ;-)
I retract comment #10. The problem I was having seemed to be a multi-day but transient internet connectivity issue where much of the net was dropping packets. I was mistaken into thinking it was squid because one site I used to browse through squid now refuses to work through squid (europac.net). DIRECT works fine, through squid (with IE, FF, etc) it fails. Weird! But obviously some other sort of bug. (And that site works fine with my other network that uses FC3 squid!) Please don't keep this bug open on my account.