W2K server to W2003 Server via cross over cable
We have a Rampage workflow here with some older equipment. As network traffic is high we have been advised to have a direct connection between our file server (W2K) and our Rampage RIP (WK2003) via a cross over cable.
I have done this with Mac's before (cross over cable) but I'm looking for a little advice before we try this as it's some critical equipment that I don't want to screw up. Making a cross over cable is not a problem, setting up the static IP addresses is understood but in addition to have both servers directly connected we also need both servers available on our LAN so clients can access both of them at the same time.
So that's 4 NIC cards we need to address, 2 for the LAN and 2 to make the connection between the 2 servers.
Do we setup static IP addresses for each NIC or create some kind of alias between the two cards on each machine?
Never attempted this before so looking for advice, thanks in advance.
Are you running an AD domain or just at workgroup level? Connecting two machines over two different subnets can confuse a domain if it isn't set correctly. Also it may make more sense to upgrade your switch and bond multiple NIC's on the file and rip servers to increase the throughput. This will provide more for all uses and reduce the potential for conflicting activity. The only pieces of your network that experience contention is likely the segments from the switch to the servers. There are 4 port NIC cards out there specifically for this purpose.
It shouldn't be too big a deal. I will assume each server has at least 2 NICs. I will also assume that currently, the one NIC that is actively used in each server already has a static IP assigned. You will need to assign another set of static IPs to the NICs in each server that will be directly connected. HOWEVER, the important thing to realize is that you want to make a subnet, rather than simply a second NIC in each machine that is on the same network. That would cause you grief and not give any benefit besides.
The subnet is the portion of the IP before the node. Where the split occurs between subnet and node can vary; that's the whole purpose of a subnet mask, which for most of us in prepress has always been 255.255.255.0
This common subnet mask is telling the configuration that the first three numbers are the subnet and the last number is the node. For example, using a subnet mask of 255.255.255.0, a subnet could 192.168.1 and another subnet 192.168.2, etc. (or any number scheme; 192.168 is common with intranets, and so it 10.1) Within each of those, the last number tacked onto each is the individual machine's identification number (the node). So one machine is 192.168.1.13 and another could be 192.168.1.14.
My advice is that when you add a subnet, keep each machine's node number the same. It's not required but it sure helps keeps things straight.
I have a pair of ApogeeX servers that I have connected as you are proposing. Both boxes are visible on two of my workstation networks and also share a private subnet for raster data. All done like so:
AX1, main server has 3 NICs: 192.168.1.13 serves Mac clients. 192.168.3.13 serves windows clients. 192.168.9.13 is on my private raster data network (which also includes all imagers and proofing devices to keep their higher traffic off of workstation networks).
AX2, my satellite and failover also has 3 NICs: 192.168.1.14, 192.168.3.14 and 192.168.9.14.
All these IP addresses use the common subnet mask of 255.255.255.0, splitting the 4 IP number 3/1 for subnet/node.
If you have a WINS server running for name resolution, you will want to enter its address in the IP config for ONLY ONE of the NICs. In my case, name resolution is most valuable to my Windows clients, so I choose to enter a value for WINS in the NIC connected to the 192.168.3 subnet. Leave the others blank.
Lastly, and probably MOST IMPORTANTLY, as these multiple servers are visible in multiple subnets, you need to enforce the choice of which "route" is taken when one server is talking to another. It's a random toss up if you don't take additional steps (which I will explain). Take my example above -- when server 1 wants to send data to server 2, it might decide to use the 192.168.3 subnet, making a bunch of noise on my windows client subnet and slowing down the network performance of those users.
The solution is to edit the hosts files on both servers. On Windows the file is located at: /<window dir>/system32/drivers/etc
In the hosts file, add an entry in server 1 that tells it the address of server 2. And likewise in the hosts file of server 2, an entry with the address of server 1. Use the private subnet you created -- like my example above, it's the 192.168.9 subnet. Then all data between the servers takes that route.
However, the final step may not be necessary if you enter static IPs in the various places of your workflow config that define where data goes. But still, I would edit the hosts file. Most modern systems use names rather than IP numbers, and even if it appears to take an IP, there could be other areas that go by name. Play it safe and add the names and corresponding IPs to the hosts file.
Clear as mud? Reply with further questions if I've glossed over any details. To sum up, it's really not a big deal, but it is true, the potential performance gains are definitely worth the effort.
Last edited by William Campbell; 09-28-2012 at 06:00 AM.