Slow communication with Windows systems after upgrade from 7.1 to 7.2

Hello ALL,

We had a MailServer (base 7.1) with postfix, currier (IMAP/POP3), MySQL, postfix admin and MAIA and as webmail SquirrelMail and HORDE (IMP).
Everything worked perfectly until we did upgrade from 7.1 to 7.2.
After upgrade, the communication with Windows XP and also Vista clients slow down horribly. All protocols POP3/IMAP and their SSL variants. The strange thing is that also webmails are veeeerrry slow (SquirrelMail and also HORDE). It take 30 minutes to display messages in webmails. The same behavior is with Outlook (Outlook Express) communicating through POP3 or IMAP protocols. The same situation is on accessing phpMyAdmin from Windows.

From LINUX, BSD, MAC OS ... it works fine ... as it was before update... These systems doesn't have any problems.

I don's understand what changed with upgrade form 7.1 to 7.2 and why just WIndows OSes have these problems.

Do anybody have the same problem? Or do anybody have an idea what's wrong?
Where to look for the solutions?

Marek
 
If all protocols are behaving oddly to Windows there must be something wrong in your network stack but you'd think other clients would see the problem then. Still, what does ping, tracert report from a Windows client to the mail server vs a BSD client to the mail server?

Perhaps check for bizzare nameserver or DHCP issues.

Does the mailserver machine also run a firewall?
 
[SOLVED] MTU to 1200

The problem was in new FXP driver (Intel Ethernet).
IN 7.1 MTU 1500 was fine but in 7.2 MTU 1200 must be set up :)

After set up MTU to 1200 for FXP all systems are working correctly :)

Thanks anyway.

Marek
 
Interesting... where did you uncover this?

I have plenty of fxp nic's but only moved one server to 7.2 so far so I am curious about the origins of this issue and the fix you've located, especially since the default MTU is 1500.

While I'm glad that my gut sense was "network stack" and that seems born out, I'm still left wondering. Are you sure of this? Such a change in a popular device you would think would have been picked up in testing.
 
Yes, that's right ... MTU must be 1200

Hi,

I have changed MTU to 1500 and Windows colleagues were slowed down ...
when I change MTU to 1200 it's OK.

I have found the error about 35 bytes missing in tcpdump ;) This was the reason why I have tried to change MTU size ;)

This is my ethernet card: fxp0: <Intel 82551 Pro/100 Ethernet> which is doing problems with MTU 1500 :) It's funny that till 7.2 it works with default MTU 1500 ... just 7.2 something change within this driver.

Thanks anyway.

Marek
 
There's a bug in the fxp driver. Somehow it made it through to 7.2-RELEASE. You need to turn off TCP Segment Offloading by doing the following:

sysctl net.inet.tcp.tso=0

The problem seems to affect certain DSL/PPPoE type connections.
 
i have the same problem.
i found a Errata Notice about this and i followet the instructions there. Still i have the same problem. It only works if i manualy set:
sysctl net.inet.tcp.tso=0

I even tried to add the line above in /etc/sysct1.conf but after restart i still have to manualy type the command in console.
Can anybody help me?
 
Hello Mr. Daemon;

I'm very new to freebsd FreeBSD ...in my 6 months playing with it (networking stack), you have made a cameo appearance in my life THREE TIMES. I wasted a whole work day on this little beauty before I came across your note 2 YEARS AGO. If you ever get to Austin, Texas ... I'm gonna buy you a beer!
 
Back
Top