I have an internet server which operates in multiple capacities: Apache 1.x, Postfix/Cyrus, ftp, ssh, Squid, and MySQL.

For some reason the transfer rates incoming connections to Postfix/Cyrus IMAP are horrendously slow. The HTTP requests from the Internet side of our network are incredibly slow as well, averaging around 2-3Kbps. Access outbound from the server is fine. We use squid to proxy outgoing internet requests and there is no apparent slowness to it while the Postfix/Cyrus connections from the internal network are incredibly slow. I'm guessing that the problem is that the firewall is not controlling the connections properly.

Incoming from our ISP (T1 via wireless lan) we have their wireless router followed by our d-link DFL80 router of which the DMZ goes to the server and the Internal goes to some client suites we run. The server in the DMZ has a software firewall (shorewall) to protect the server as well as the internal users who are attached by a seperate ethernet card in the server.

I've tried using the TCPMSS clamping option in the firewall, but it does not seem not effect it at all, even though the reasons for using the option fit what was going wrong.

So a quick summary:
- From the internet, http (and probably IMAP) connections are very slow (2-3 Kbps)
- From the internal network, squid and mysql connections are fine, but IMAP are not.

Please Help!

any error logs on the firewalls? I had this kind of problem with a d-link... it kept on dropping returning, valid packets. A firmware upgrade did the job.

Nope, the firewall didn't drop any packets. It apparently has to do with having 3 D-Links in a SuSE box. Also, there is an onboard vii-rhine which may have been causing the bottleneck. I found yesterday that the communications between the network connected on a realtek card worked just fine and dandy. I pulled the 3 Dlinks and forego using the onboard. I then put in 4 3coms that I had sitting on a shelf (new). After some hassle getting the new tulip driver compiled, it seems to work just fine now.

My diagnosis as it stands is that it may have just been the onboard rhine causing the problem, as that was the conduit to the router. All the internal traffic to our server which was going slow was routed via DNS to our public IP. The onboard NIC was the one connected to our router and therefore it was the problem causing the network issues. Apparently it had some sort of driver misconfiguration or something. I don't really care now that I'm using 3coms. I think I 'll just let it be and not spend anymore time on the NIC issue.

Thanks

3coms are the best Nics to me, with Intels... I wouldn't use a rhine -via nic for a server.

Glad to hear all is fixed now.