TCP/IP basics

Let me try to give you a short (it will not be very short) and simple introduction to TCP/IP. I will try to keep it as simple as possible (it may to be too simple for some of you, but if you need all the TCP/IP details, you need to go to your local bookstore and buy on of these books with 1000+ pages on TCP/IP). I will also not bother you with the history of TCP/IP and the Internet, unless required for the understanding of the setup.

Each Ethernet board worldwide has a unique Ethernet-address, it is a 48 bit number (the first 24 bits indicate the manufacturer, the last 24 bits are a unique number for each Ethernet board/controller-chip assigned by the manufacturer).

When systems on a local area network ("LAN") are configured with NetBEUI or IPX/SPX protocol, they use these hardware-addresses to identify each other, so there is no need to define manually a network address.
But TCP/IP was designed as a Wide-area-network ("
WAN"), able to continue to function, even if part of the network was not operating ( damaged or destroyed).

TCP/IP uses IP-addresses, which are 32-bit numbers. To make it easier to memorize such IP-addresses, they are usually expressed as 4 8-bit numbers (example:, where each of the 4 numbers is within the range of '0' to '255' (there are restriction on using '0' and '255', avoid using them.).
When setting up a small private network, you are free to use ANY IP-address, however, when you are connected to a company network, you need to ask the Network-administrator to assign you an IP-address. And if you are connected to the Internet, your
ISP (Internet Service Provider) will assign an IP-address to you.
Even if a network is NOT connected to the Internet, it has become custom to use on private networks a range of IP-addresses, which are reserved for private networks (that makes it later possible to connect your private network to the Internet without having to re-configure everything). The reserved IP-address is: 192.168.x.y, where x=same number on all systems and y=different/unique number on all systems.
A small network of 3 systems would use:

You configure this IP-address in the properties of the TCP/IP-protocol:

(For now, simply enter as 'Subnet Mask", it will be explained later in this document)
That's it, if you just like to connect systems on a small network, the network should work and you can
test the Connection using the TCP/IP ping.

On a small network, you can still memorize the IP-addresses used, but if your network grows to 50+ systems, it becomes a serious management job. But TCP/IP offers some help by allowing to configure it to:
"obtain an IP address

To be able to make this automatic assignment, there needs to be now on the network a database, keeping track of possible IP-addresses and to whom these addresses have been assigned:
DHCP (Dynamic Host Configuration Protocol)
On bootup, the system sends out a call on the network to find a DHCP-server, which assigns an IP-address to such a system. The IP-addresses are usually assigned NOT permanently, but for a specific time (could be days, weeks, months or on Internet-connections just for the ONE connection). If the system contacts the DHCP-server again during this time, the 'lease' on the IP-address is extended. But if you come back from a long vacation, your 'lease' of the IP-address may have expired, that IP-address may have been assigned now to somebody else, and you/your computer get now assigned a new IP-address.
Windows95 itself does NOT include any DHCP-server, you need to connect to a Windows NT (or similar class) server , which is configured as DHCP-server.

Microsoft supplies now with Windows98 a feature for
IP-Auto-Confuguration without a DHCP-server on the network.

Looks simple until now ? Actually there is already a lot more 'hidden' actions:
The systems have IP-addresses, but Ethernet-boards ONLY know their Ethernet-address, so as soon as a TCP/IP configured system is switched on, it is advertising its presence onto the network:" Hey, I am alive, my Ethernet address is '08000b 0a0238' and my IP-address is '' ". , and each TCP/IP system on the network builds up a table with all this information, which is usually checked/verified in time-intervals of 15 min.

If your system needs now to communicate with a station, for which it does NOT have an entry in its table of IP/Ethernet-Addresses, it sends out a search-message to everybody ("
Broadcast-Message") like: " Hey, I like to communicate with the IP-address '', but I do NOT know your Ethernet-Address. Please, identify yourself". This causes the system with the requested IP-address to send out its advertising again.

These processes are called
ARP (Address Resolution Protocol) and RARP (Reversed Address Resolution Protocol).

This ARP/RARP works fine on a local-area-network (on an Ethernet network), but will NOT work for Internet communications, because:
- the Database of Ethernet-to-IP-address would need to have 10+ Million entries
- the Internet would only be busy with ARP/RARP.

To connect a TCP/IP local-area-network to another TCP/IP LAN (which could be the complete Internet), you need now a device called : Gateway or Router

You need to tell Windows95 about the Gateway in the TCP/IP-properties:

Now, also the '
Subnet-Mask', which is usually '', becomes important: if you now like to connect to (which is the Website of Microsoft), TCP/IP checks your own IP-address and the IP-address of the destination against the Subnet-mask. Lets do that comparison on a binary level:

System: IP/subnet-mask Binary
your system 11000000 10101000 00001010 00000001
Microsoft 11001111 01000100 10001001 00110101
Subnet-mask 11111111 11111111 11111111 00000000

TCP/IP compare now the part of the addresses, defined by the '1's in the subnet-mask: if your system connects to another system on the same network, that part of the address (the first 24-bits in this example) are the same, so TCP/IP looks up the Ethernet address in its ARP table and connects directly to that system.
But if there is a difference in these 24-bits, then TCP/IP connects to the Gateway (in this example:, and it is now the job of the Gateway to establish somehow the connection to the destination system (somewhere inside that Internet 'cloud'). The Gateway/Router keeps for that purpose special tables and passed on the request to the next router, which itself goes to the next, which itself goes to the next,....., until you reach the destination.

Each Router/gateway on the network (which could be the Internet) passes
on the message, until it reaches its destination, and the reply comes back
the same way (for more details on Routing:
Setup TCP/IP Routing )

The above assumes, that all systems have an IP-address, which is valid on the Internet !
If you connect via adialup-connection to the Internet, but like to use the connection on
multiple systems on a network, you need a

You can check this yourself: open on Win95 a DOS-window and run the '
tracert' command (which is installed as part of the TCP/IP protocol).
In my example, I traced the router to "":

it is difficult to remember IP-addresses, it is much easier to remember names (and having the computer lookup the name and find the IP-address). That is the purpose of the 'HOSTS'-file and 'LMHOSTS'-file: Windows95 TCP/IP installs in
C:\WINDOWS a file called 'hosts.sam' and 'lmhosts.sam', rename/copy it to 'hosts'/'lmhosts' and then use it to define the names:

The formatting of 'hosts' and 'lmhosts' is the same:
IP-address, some spaces, computername

When to use HOSTS and when to use LMHOSTS ?
That is a confusing subject: having 2 files with a very similar job.
HOSTS is read by basic TCP/IP software (ping, ftp, ......)
LMHOSTS is used by the Microsoft Networking/Client/Workgroup management. If systems are on the same cable segment, the system broadcast their presence and find each other automatically, no need to enter anything in LMHOSTS.
However, such broadcast-packets to NOT get routed. Adding then the IP-address manually in LMHOSTS makes the system aware about a system on different segments.
NOTE: LMHOSTS originates from "Lan Manager HOSTS", a name from the history of Microsoft networks.

Are you already tired to typ these IP-addresses ? Looks like another item for automation, and exactly that is DNS : Domain Name Service:
it allows to use names instead of IP-addresses, but you need to configure it as part of the TCP/IP-properties:

If you now define an address ( like: or in the picture below) TCP/IP will make:
1: a call out to a DNS-server, asking for the IP of
2: the DNS server will reply with the IP-address (in this example
3: TCP/IP makes now the connection to the requested server, using the IP-address

That's it for a basic course on TCP/IP.
For more info , see :
FTP : File Transfer Protocol
(Yes, I know, I did not touch here on 'telnet, and the other TCP/IP goodies, maybe later on another page....).