3. Planning Your Network Topology

While there are arguments which can be made for many different network layouts, the requirements of many organizations can be met by putting the desktop machines and private servers on a private masqueraded subnet, and the publicly accessible machines on valid external IPs. The machines on valid external IPs will be referred to in this document as ``exposed hosts''. This leads to the following (example) topology:

+--------------+               
|              |               +---------------+
| ISP-supplied |---------------| FTP server    |
| router       |        |      +---------------+
|              |        |
+--------------+        |      +---------------+
                        |------| WWW server #1 |
                        |      +---------------+
                        |
                        |      +---------------+
                        |------| WWW server #2 |
                        |      +---------------+
                        |
                        ~
                        ~
                        |      
                        |      +---------------+
                        |------| Private       |
                               | Network       |
                               | Gateway       |
                               +---------------+
                                      |
                                      |
                                      |
                                      |
     +------------+                   |      +-------------------+
     | Desktop #1 |-------------------|------| Private server #1 |
     +------------+                   |      +-------------------+
                                      |
            .      -------------------|--------        .
            .                         |                .
            .      -------------------|--------        .
                                      |
     +------------+                   |      +-------------------+
     | Desktop #N |-------------------|------| Private server #N |
     +------------+                          +-------------------+

In this example, the router provided by the ISP (Internet Service Provider), FTP server, WWW servers, and the machine labelled ``private network gateway'' all have externally visible IP numbers, while the desktop and private server machines have IP numbers allocated from RFC 1918, reserved for private use. The IP numbers you choose for use within the private network (everything below the private network gateway machine) should be chosen to be unique, not only among the hosts under your control, but should also not conflict with numbers assigned on similar private subnets at other sites or partner companies with whom you might, at some time, want to implement a virtual private network, in order to reduce confusion and reconfiguration when the networks are merged in that way. As outlined in the RFC, you can choose from any class C network from 192.168.0.* to 192.168.255.*, or any class B network from 172.16.*.* to 172.31.*.*, or the class A network 10.*.*.*. In the rest of this document I will assume that your private network (if you've chosen to create one) is on the class C network 192.168.1.*, and your private network gateway machine is at IP number 10.1.1.9, one of the IP numbers provided to you by your provider (note that this is not a valid external IP, I use it as an example only). I will also assume that there is a machine, betty.example.com, at 10.1.1.10, which will handle both www and FTP services.

Take note of the number of external IP numbers which you need for your own machines. You will need one IP number for each machine which lies outside the private network gateway, plus one for the gateway itself. This count does not include any IP numbers which may be taken by routers, broadcast addresses, and so on. You should ask your provider for a block of addresses large enough to mount the given number of machines. For example, in my office network, of the 8 IP numbers allocated from the ISP, three were not usable by my computers, leaving enough IP numbers for four machines outside the gateway, plus the gateway itself.

This network topology is not correct for everybody, but it is a reasonable starting point for many configurations which don't have special needs. The advantages of this configuration include:

• Easy expandability. If you suddenly double your number of private nodes, you don't have to worry about getting a new IP block from your provider and reconfiguring all of the interfaces on your machines.
• Local network control. Adding a new workstation to your private network requires no communication with your provider, unlike exposed nodes, which need both forward and reverse DNS (domain name service) mappings if they are to perform certain tasks (ssh and ftpd may complain if they can't perform reverse and forward DNS on incoming connections). A reverse DNS query is an attempt to obtain the host name from the IP number.
• Centralized security. The private network gateway can enforce security over the whole private network, filtering packets and logging attacks, rather than having to install such measures on each desktop and server on the private network. This can be enforced not only on incoming packets, but also on outgoing packets, so that a misconfigured desktop machine doesn't inadvertently broadcast data to the outside world which ought to remain internal.
• Easy transplantability. Because the IP numbers within the private network are yours for as long as you want them, you can move the entire network to a new range of IP numbers without having to make any changes to the network configuration on the private network. The publicly exposed hosts still have to be reconfigured, of course.
• Transparent Internet access. The machines on your private network can still use FTP, telnet, WWW, and other services with minimal obstruction, assuming a Linux masquerading router. The users may not even be aware that their machines are not on externally visible IP numbers.

Some of the potential disadvantages of such a configuration are:

• Some services will not be available directly to the machines on the internal network. NTP synchronization against an outside host, certain obscure services which may not have masquerading rules in the kernel, and .shosts authentication for logging in to external nodes are all difficult or impossible, but simple workarounds are almost always available.
• More network hardware costs. The private network gateway machine needs two network cards, and you need at least two hubs / switches, one on the visible network and one on the private network.
• Machines outside the private network cannot easily make direct connections to machines within the private network. They may have to open a session first on the private network gateway machine, then log through to the internal host. It is possible to route packets transparently through the firewall, but this is not recommended for security reasons which will be discussed in a later section.

You should consider these points in planning your network topology, and decide if a fully visible network is more appropriate for your situation. In the rest of this document I will assume that you have configured your network as shown above. If you have chosen to have a fully visible network, some details will differ, and I will try to point out such differences in this document.

As a special case, if you do not need any external servers, the ISP-supplied router can be attached directly to your external interface on the private network gateway machine, rather than with a hub.