Multihomed Server Connect issue

I'm the host of the "24/7 Katabatic Craze" server, and I've noticed an issue with multihomed servers. Multihomed, as in where you have multuple network adapters and IPs.

Heres the configuration:

Windows Server 2008
8 GB DDR3 RAM
Core2 Quad @ 3.2ghz
7x500GB in RAID5
1KW Power Supply
2 NICs, one for internal 10.1.1.0/24, one for external, one virtual for VPNs 10.1.1.2/24.

Externally it functions fine, but the moment somebody connects from the internal network, it drops all the external clients, and they can no longer connect.

Is there a way to forcefully bind the server to one IP, and not "default"?

Comments

  • OK, so this is what I get in the console:

    Client 2537 timed out.
    Issuing Disconnect packet.

    I've tried connecting normally, directly, and i've tried forcing it to bind to one IP and port. SAME RESULT. What the hell...?
  • Hi

    How do I set a multihomed server to run with a specific IP address?


    You can bind to a specific IP with:
    $Host::BindAddress = "";

    in ~tribes2/base/prefs/ServerPrefs.cs
  • I did that, and confirmed with console output... same result. :-[
  • You set the bind address in that variable to the 10.1.1.x address, correct?

    This behavior is peculiar, even for multi-homed servers. Do you happen to have a firewall installed on that host, beyond the standard Windows firewall?
  • The server itself is acting as a router, the one NIC connected directly into the modem.

    The server is set to bind to the external, internet-routed IP.

    Undesirable operation, yes, but my router decided to brick out on me and well, im a college student so... yeah.
  • There might not be any straightforward mitigation in this case. It sounds like Windows ICS, or whatever routing platform you are using for connection sharing is not compatible with the way T2 binds a datagram socket.

    You might have some luck by running the T2 server inside a virtual machine, where the VM contains a bridged adapter to the internal network and a NAT adapter to the external network, and using a default bind. This is a pretty unique configuration, though.
  • Would there be a significant performance disadvantage to doing so?

    (processor doesn't have extensions needed for HyperV)
  • VMware's virtualization products should be just as fast for the sort of workload you will have for a T2 server -- their ring compression software virtualizer is actually faster than some of the early VT-x/AMD-V CPUs.

    Those products are only slower for high I/O applications (like databases). As a T2 server is mostly CPU utilization, I doubt the difference will be worth noting.
  • Alright I'll give her a shot. i'm kinda worried that installing vmware might screw with RRAS on server 2008, but we'll see where it goes.
Sign In or Register to comment.