6.2. Introduction to TCP/IP (TCP/IP suite)

6.2.1. Services on TCP/IP

The most important traditional TCP/IP services are [Gar98]:

• File transfer: the file transfer protocol (FTP) allows the user of a computer to obtain files or send them from one computer to another. In order to do this, the user must have an account in the remote computer and identify themselves through their login name and password or the user must connect to computers containing an information repository (software, documentation etc.) under an anonymous account to read those computers on their computer. This is not the same as the more recent Network File Systems (NFS) (or netbios protocols over TCP/IP, a completely insecure "invention" in Windows, which should be replaced with an older but more secure version called netbeui) that make it possible to virtualise the file system in a machine so that it can be accessed interactively from another computer.

• Remote connection (login): the terminal network protocol (telnet) allows a user to remotely connect to a computer. The local computer is used as the remote computer's terminal and everything is executed over it, whilst the local computer remains invisible from the perspective of the user that started the session. This service has now been replaced by the SSH (secure shell), for security reasons. This can use a remote connection through telnet and the messages are sent as plain text; in other words, if someone "examines" the messages on the network, it is equivalent to looking at the user's screen. SSH encrypts the information (which is an added-value to the communication) so that the packages on the network cannot be read by any foreign node.

• Email: this service makes it possible to send messages to users of other computers. This form of communication has become an essential element for users and allows email messages to be sent to a central server, so that they can then be recovered using specific programs (clients) or read through an internet connection.

The progress in the technology and the increasingly lower cost of computers has meant that determined services have specialised and are now configured on determined computers working in a client-server model. A server is a system that performs specific services for the rest of the network or connected clients. A client is another computer that uses this service. All of these services are generally offered within TCP/IP:

• File systems in network file systems: allows a system to access the files through a remote system in a manner that is more integrated than FTP. The storage devices (or part of them) are exported to the system that wishes to access the files and this system can "see" them as if they were local devices. This protocol permits in the server side to establish the rules and ways of accessing the files, which (if properly configured) makes the place where the information physically resides independent from the place where the information is "accessed".

• Remote printing: permits users to access printers connected to other computers.

• Remote execution: permits a user to execute a program on another computer. There are various ways of executing a program in this way: either through a command (rsh, ssh, rexec) or through systems with RPC (remote procedure call), which allows a program on a local computer to execute a function in a program on another computer. The RPC processes have been studied in-depth and there are various implementations, but the most common are Xerox's Courier and Sun's RPC (the latter has been adopted in most UNIX systems).

• Name servers: in large-scale networks of computers, there are data that have to be centralised so that they can be easily used; for example, user names, passwords, internet addresses etc. All of this makes it easier for a user to have an account for all the machines in an organisation. For example, Sun's Yellow Pages (NIS in the current Sun versions) is designed to handle all these types of data and it is available for most UNIX systems. The DNS (domain name system) is another domain-name service but one that keeps a direct relationship between the hostname and the logical identification name of this machine (IP address).

• Terminal Servers: connect terminals to a server that executes telnet so as to connect to the central computer. These types of setup are basically useful for reducing costs and improving the connections to the central computer (in some cases).

• Graphical terminal servers (network-oriented window systems): these permit a computer to visualise graphic information on a display that is connected to another computer. The most common of these systems is X Window.

6.2.2. What is TCP/IP?

TCP/IP is in fact two communication protocols between computers that are independent to each other.

TCP is a connection-oriented protocol, in other words, it is equivalent to a telephone, and the communication is considered as a data stream.

In other words, it routes the data between two computers through the networks. It corresponds to layer 3 of the OSI/ISO model and it is a connectionless protocol (see following table). [Com01, Rid00, Dra99]

An alternative to TCP is the UDP protocol (user datagram protocol), which treats the data as a message (datagram) and sends packets. It is a connectionless protocol (the recipient computer does not necessarily have to be listening when the other computer establishes communication with it) and it has the advantage of creating less overload on the network than a TCP connection, but it is less reliable (the packets may not arrive or arrive duplicated).

There is another alternative protocol called ICMP (Internet control message protocol). ICMP is used for error or control messages. For example, if one tries to connect to a host computer, the local computer may receive an ICMP message indicating "host unreachable". ICMP may also be used to extract information on a network. ICMP is similar to UDP in that it handles messages (datagrams), but it is simpler than UPD, because it does not have port identification (the ports are mailboxes where the data packets are left and where the server applications read the packets) in the message header.

In the OSI/ISO communications model (OSI, open systems interconnection reference model, ISO, International Standards Organization), is a theoretical model applied by many networks.There are seven communication layers where each one has an interface for communicating with the preceding and following one.

To summarise, TCP/IP is a set of protocols including IP, TCP, UDP that provide a set of low-level functions used by most of the applications. [KD00, Dra99].

Some of the protocols that use the abovementioned services were designed by Berkeley, Sun or other organisations. They are not imcluded (officially) as part of the Internet protocol suite (IPS). However, they are implemented using TCP/IP and they are therefore considered as a formal part of IPS. A description of the protocols available by Internet can be found in RFC 1011 (see references on RFC [IET]). There is currently a new version of protocol IPv6, also called IPng (IP next generation) which replaces IPv4. This protocol significantly improves the previous ones in elements such as having a greater number of nodes, traffic control, security or improvements in the routing.

6.2.3. Physical network devices (hardware)

From the physical point of view (layer 1 of the OSI model), the most commonly used hardware for LAN is that known as Ethernet (or FastEthernet or GigaEthernet). Its advantages consist of a lower cost, acceptable speeds (10, 100 or 1,000 megabits per second) and its user-friendly installation.

The first two are obsolete (they used coaxial cable) whereas the last is through twisted pair cables and connectors similar to those used by telephones (known as RJ45). The twisted pair connection is known as 10baseT or 100baseT (according to the speed) and it uses repeaters known as hubs as interconnection points. Ethernet technology uses intermediate communication elements (hubs, switches, routers) to configure multiple segments of the network and divide the traffic to improve the performance of the data transfer. Normally, in large organisations, these Ethernet LAN are interconnected through fibre optic cables using FDDI (fibre distributed data interface) technology, which is more expensive and more difficult to install, but with which we can obtain transmission speeds equivalent to Ethernet whilst not having the limits on distance involved in Ethernet (FDDI allows for distances of up to 200 km). The costs are justified when they are used between buildings or other network segments that are very congested. [Rid00, KD00].

At the same time, there are other types of hardware that are less common, but no less interesting, such as ATM (asynchronous transfer mode). This hardware allows us to set up a LAN with a high level of service quality and it is a good option when we have to set up high-speed and low-latency networks, such as those that require real time video streaming.

There is other hardware supported by GNU/Linux for interconnecting computers, of which we would mention: Frame Relay or X.25 (used in computers that access or interconnect WANs and for servers with large data transfer needs), Packet Radio (interconnection via radio using protocols such as AX.25, NetRom or Rose) or dial-up devices that use serial lines, which are slow but very cheap, through analogical or digital (RDSI, DSL, ADSL etc.) modems. The latter are the ones commonly used domestically or in small and medium-sized businesses, and they require another protocol for the transmission of packets, such as SLIP or PPP. In order to virtualise the diverse hardware on a network, TCP/IP defines an abstract interface through which all the packets that will be sent by a physical device (which includes a network or network segment) are concentrated. Consequently, for each communication device in the machine, we will have a corresponding interface in the operating system's kernel.

In GNU/Linux, this may mean that we have to include the appropriate modules for the appropriate device (NIC network interface card) in the kernel or as modules, and this means compiling the kernel after choosing, the appropriate NIC, with, for example, make menuconfig, indicating it as internal or as a module (in the latter case, the appropriate module must also be compiled).

The network devices can be seen in the /dev directory, where there is a file (a special file, which may be a block file or a character file, according to the transfer) that represents each hardware device.[KD00, Dra99].