unsure how to set up direct link between 2 servers

[PPettit]

We've got 2 servers: one for applications (Win2k Advanced Server)and one for Citrix connections (Win2k Server with Citrix Metaframe XP). They each have 2 network cards installed: one for the network and one for a crossover connection between the two. They're all working but I just want to make sure I have the direct link set up right because there doesn't seem to be as much traffic on it as I expected. My goal was to see if I could speed things up a little bit on my end for the remote users. I haven't noticed a change yet but they may just be doing very little that requires server-to-server communication.

App server:
NIC1: 192.0.0.250 (primary address)
255.255.255.0
192.0.0.1
NIC2: 192.168.2.2 (secondary address)
255.255.255.0
192.168.2.3 (to NIC2 on the Citrix machine)

Citrix server:
NIC1: 192.0.0.50 (primary address)
255.255.255.0
192.0.0.1
NIC2: 192.168.2.3 (secondary address)
255.255.255.0
192.168.2.2 (to NIC2 on the App server)

I suppose this isn't necessary but in the HOSTS file on each machine, I have the other server's name and address:
App server:
192.168.2.3 Citrix server
Citrix server:
192.168.2.2 App server

Is this the best way to configure the direct link?

Thanks in advance for any information.

 

[bcastner]

Why are you using public IPs on the two servers? Why is your HOSTS file in illegal syntax? Why are you using two different network subnets?

 

[jimbopalmer]

And why bother with two NICs when one will do? why try to make a server into a router?

I tried to remain child-like, all I acheived was childish.

 

[Serbtastic]

How are users connecting to your App and Citrix server?

Are they using TS Client to connect to the App server? Then they'd be using the public IP (192.0.0.250) not the private.

 

[PPettit]

Allow me to clear a few things up:

The question is only about the direct link and the best way to configure it.

The actual HOSTS file on each server is correct. What I posted here was for example only. I guess that wasn't as obvious as I thought it was.

The reason for the direct link is to see if there is any gain in communication speed and reliablility when the remote users access the financial app which resides on the main server.

The people who set up the network and related devices are accountants, not techs. They've been getting a little support from a software consultant over the phone and have just had to figure out the rest on their own. They hired me about 2 months ago to take over the IT related issues so that they could get back to doing their actual jobs. I fix their mistakes when I have the time or when those mistakes create problems. Right now, the unusual addressing scheme is not high on my priority list.

Both of the servers are behind a router. The router is set to forward the remote users directly to the Citrix server via port forwarding. The 192.0.0.1 address is the internal address of the router. The router connects to two unmanaged switches. The PC's and printers are connected to one of those switches. Everything internal uses the 192.0.0.XXX scheme except the direct link between the servers. The connection was only partially set up and I decided to mess with it since I had a few spare minutes.

The Citrix server currently has no need to access anything but the main server. The main server contains our financial app and related print services. You don't even want to know about the bizarre way the app has to have printers configured. We're currently using a mix of TCP/IP, NetBEUI, and NetBIOS. We even had IPX running but I've been disabling it as I find it. When the financial app changes, we'll be able to use TCP/IP only.

Need anything else clarified?

 

[bcastner]

"Right now, the unusual addressing scheme is not high on my priority list."

The block of addresses you are using are owned by:

NS1.AGROSERVICES.NET
NUL1.JSTELECOM.COM
NUL0.JSTELECOM.COM

 

[jimbopalmer]

"The people who set up the network and related devices are accountants, not techs. "

Thanks, that explains why they think two NICs in a Computer is a good idea.

I tried to remain child-like, all I acheived was childish.

 

[PPettit]

bcastner, what is so difficult to understand?

The addresses are used internally. Not externally. You can use whatever addresses you want to, internally. The addresses we're using are not accessible from the internet so they won't be interfering with the identical public addresses at all. If it's a problem with our end users, it still isn't high priority. The only thing they need to access on the internet is their email. Except for a few individuals, if they're browsing the internet, they aren't doing their work.

jimbopalmer, two or more NIC's in a server is a good idea sometimes. Haven't you ever heard of load balancing? Also, at a wireless ISP I worked for, some of us used 2 NICs in our PC's. One was for internal communication and one was for troubleshooting. The other card allowed us to access the wireless network like our customers did. In my current situation, having 2 NIC's per server is not all that useful, but that's the way it is for the time being.

Anyone want to try answering my original question for a change?

 

[jimbopalmer]

No, Hate to be blamed for a answer to a bad question. Use your routers to route and your servers to serve, do not make your server a router.

I HAVE used two NICs with the same IP address for redundancy, that is not what you are proposing. My network gear calls it multi link trunking. I use it mostly to provide seperate paths between buildings.

Any user of your network will go to the internal IP address even when they WISH to go to a external IP address in the same subnet. I have no idea if they are actually trying for web sites with that IP address, but if so they can't get there. (note: if your company IS agroservices.net then I apologise, I am way off base)

I tried to remain child-like, all I acheived was childish.

 

[bcastner]

It was not my intention to give you a hard time, but surely you must realize that to "load balance", multi-home (which is closer to your intention, I think), or bridge (unecessary, but perhaps your intention), and to do it correctly in a TCP/IP setting than every aspect of the connection is fair game as far as I am concerned.

I think you can have issues with any client connected to the internet, and there is no reason not to correct the addressing errors. This is particularly important as the MS client has evolved to essentially a pure DNS system for resolving even local names. To exclude, as you would have to, and entire Class A network because of your internal IP addressing schema strikes me at least as an odd choice.

Do a proper Class C IP addressing schema. You cannot do "load balancing" otherwise, and multi-homing the server would be a serious mistake under your schema; cf. :
Symptoms of Multihomed Browsers

http://support.microsoft.com/default.aspx?scid=kb;EN-US;191611

Slow Computer Browsing

http://support.microsoft.com/?kbid=288801

It is possible to have the OS to balance the network load for the computer when having several Network Adapters.

This is controlled by these DWORD values :

[HKEY_LOCAL_MACHINE \SYSTEM \CurrentControlSet \Services \NetBT \Parameters]
RandomAdapter = 1 (Disabled = 0, Enabled = 1, Default = 0)
SingleResponse = 1 (Send All = 0, Send One = 1, Default = 0)

The RandomAdapter specifies whether it should respond back with a random IP-Address (One for each adapter), or if it should respond back with the IP Address for the adapter the request was received from.

The SingleResponse says that it should only send one IP address when WINS does a name query request.

http://support.microsoft.com/?id=131736

http://support.microsoft.com/?id=175767

Note there is also a technology called Windows NT Load Balancing Service (WLBS - NT4) or Network Load Balacing (NBL - Win2k+), which handles clustering of several machines to act like a single unit. The cluster of several machines can provide redundancy for critical applications and higher load handling. The above registry settings does not have anything to do with this kind of service.

Note not all applications supports that the underlying TCP/IP layer responds with random IP-addresses. Instead one can try to setup Manual load balancing using the metric of the adapters:

If you are able to connect to several gateways/routers to other networks, it is possible to prioritize in which order the gateways should be searched for a certain address.

This is done by the Metric setting which is assigned to each gateway. The gateway with lowest Metric is the first to be searched(Default Gateway). Metric is the cost of using that gateway, and it will always use the one with the lowest cost first.

The metric can also be applied when having more than one network adapter, which connects to the same Network(Subnet). Example if using a wireless NIC and a standard NIC, then one can decide if it should use go through the Wireless NIC before using the Standard NIC. This is done by lowering the interface metric of the Wireless NIC.

The metric can also be used for manual load balancing between multiple network adapters:
Install TCP/IP for all adapters and assign them a different IP-address on the same subnet
If a certain adapter should be dedicated for outgoing traffic then it should be given a lower interface metric
If a certain adapter should be used for certain incomming traffic then make sure the clients connects to its IP-Address.

Note Netbios traffic (outgoing/incoming) can only be handled by a single adapter or else a name conflict will arise (Though it doesn't have to be the one dedicated for outgoing IP traffic)

Note it is possible for clients to connect to a service (FTP/HTTP,etc.) through all network adapters (To force clients to use a certain adapter one can use DNS or port blocking)

To configure the metric in Win2k/WinXP (WinNT4/Win9x can use the command "route&quot:
Open Control Panel
Double click Network and Dial-up Connections(Win2k) or Network Connections(WinXP)
Righ click Local Area Connection and select Properties
Double click Internet Protocol (TCP/IP)
Press Advanced-button and it is possible to configure gateways and interface metric

Note Automatic Metric is a new feature in WinXP, which automaticly assign a metric value according to interface bandwidth (Not latency), the higher bandwidth the lower the metric. More information:

http://support.microsoft.com/?id=299540

http://support.microsoft.com/?id=140859

http://support.microsoft.com/?id=164214

 

[bcastner]

This may also help:

Enable Routing in Windows 2000

Note: applies to Windows XP Pro as well.

http://support.microsoft.com/default.aspx?scid=kb;en-us;230082