Gentoo+qmail POP3 and hardware DSL router

[matofesi]

Hi,
I'm trying to find the soultion to my problem. I have the following setup:

On one side:
1. Desktop machine with Gentoo (but it's not the problem - see below)
2. Hardware DSL router connecting to the Internet (I tried two - Microcom
Deskoprte 400 and D-Link DSL-504T)

On the other side:
3. Server running Gentoo and qmail
4. DSL connection with fixed IP connected via some SpeedStream modem/router
I can't access.

As a test subjects:
5. Server running some Mandrake version with qmail in exactly the same version
as the Gentoo server.
6. Desktop machine from point 1 connected to the net via the same type of DSL as
the router in point 2 but with Gentoo box with USB DSL modem acting as a router.
7. Another Gentoo server configured almost exactly like the one in point 3 - same
version of qmail, same type of DSL connection.

My task is to fully set-up the box from point 1. It basically works - I can do everything
I need to do with on slight and veeery annoing exclusion. When I try to connect from
this box to the POP3 server from point 3 (or to the test subject from point 7) I get the
connection (using telnet I see "Trying x.x.x.x...; Connected to abc.abc.abc.;
Escape character is '^]'.") and then it sits there doing something/nothing for some 25-30
seconds before I get the "+OK <3219.1111238338@abc.abc.abc>" messgae. After that
I can check the mailbox, download mails etc. without any problems. Only when connecting
next time the same dealy occurs and to make things even worse sometimes it just sits
there doing nothing infinitely, like there is some problem with the connection.

I did some tests:
1. I connected other machine to the 2. router - effect is the same - 30 secs delay
2. I tried to test it using Windows - no change - delay
3. I tried to connect to the 5. server - it connects immediately without delays
4. I connected the 1. machine to the other network (point 6.) and it connects to 3. server
without any problems and delays.
5. I connected the 2. router in place of the second network's usuall router+modem
setup. After that the 6. connection started to act as before - 30 secs delay. Everything
else works fine without any changes.
6. I tested a few other machines with various systems (Gentoo included) and different
network connections - all of them can connect to my 3. server without any delays.

After this my conclusion is that:
a. the 1. machine is set-up properly
b. DSL line 2. itself is working fine
c. qmail server 3. works fine as I use it for quite some time and have a few users who
didn't complained once
d. the probable cause of problem is one of the following:
- DSL router connecting server 3. to the net (with low probability - test server 5. is
connected via tha same type of hardware)
- some strange interactions between 2. router, 3. modem/router or 3. server configuration

I tried also some other weird tests (lowering MTU on every accessible device etc.) but without
any success.

Is there anything anyone could suggest to try? Any suggestion would be greatly appreciated.

Thanx in advance,

M.

[DeathAndTaxes]

Sounds to me suspiciously like a dns lookup failure. Check your reverse DNS and if you don't have one, then get that fixed...In the meantime, you could either disable the reverse DNS lookups or you could put an entry in your /etc/hosts file.

Just a thought.

[matofesi]

[quote="DeathAndTaxes"]Sounds to me suspiciously like a dns lookup failure. Check your reverse DNS and if you don't have one, then get that fixed...In the meantime, you could either disable the reverse DNS lookups or you could put an entry in your /etc/hosts file.

Just a thought. [/quote]

Well... That was one of my first thoughts. Unfortunatelly (for resolving the problem) my server does
have the correct RDNS setup and adding /etc/hosts record did not helped either.

M.

[DeathAndTaxes]

One other thing that I've come across, and it's pretty weird...A crashed out or misbehaving nscd. If you have nscd running, try restarting it.

I've had nscd bring down my imap server and a few game servers don't like it when nscd misbehaves.

Just another thought.

[matofesi]

I didn't even knew what nscd was until now I never used it as I use
"real" bind on the server. As for the workstation - if it isn't started by
some defaults then it's not there too.

Any other thoughts?

M.