TCP/IP Quick Reference
TCP/IP was developed during the late 1960s and early 1970s sponsored by
the Advanced Research Project Agency (ARPA) of the american Department
of Defense (DoD). The goal was to develop a family of network protocols
reliable enough to allow communication even if some network nodes are
no longer working. Although the accurate name for this family of
protocols is "DARPA Internet protocol suite" (ARPA was later renamed to
DARPA, the "D" standing for "Defense"), it is commonly just referenced
as "TCP/IP". The protocol started to be used as 'The Internet' protocol
when an implementation of TCP/IP was included with BSD Unix 1982. The
complete documentation for TCP/IP and related issues is available as so
called Requests For Comments (RFCs).
TCP/IP consists of the following Layers:
- The lowest layer in this diagram is the Hardware Layer.
This layer consists of the physical hardware (cables and network
adapters) being used to form the network.
- The second layer in the is the Network Layer. The Network
Interfaces with the Local Network Protocol is located in this layer
which is dependent on the network hardware being used. Some common Local
Network Protocols protocols are DIX (this standard was created by
Digital, Intel and Xerox) for Ethernet, SLIP (Serial Line Internet
Protocol) or PPP (Point to Point Protocol) for serial lines.
- The third layer is the Gateway Layer. It mainly consists
of IP (Internet Protocol) and (depending on the network interface) of
some auxiliary protocols to map addresses between the network interface
and IP. IP provides a universal internet address called IP
Address to the layers above. This IP Address is a 32bit number which is
unique for each network interface in the internet. Internet addresses
are normally written in 4 x 8bit decimals seperated by 3 dots like i.e.
130.96.15.38. The IP manages all necessary routing so
higher layers need not care about how a certain package reaches its
destination.
- The forth layer is the Host Layer. UDP (User Datagram Protocol)
and TCP (Transmission Control Protocol) are two alternative protocol that
build this layer of TCP/IP and are directly used by user-written
applications. Both protocols add a second address to each network
connection, so that multiple connections between the same machines are
possible. This second address is a 16bit number known as the port number.
TCP provides a connection oriented, reliable, full-duplex stream protocol
where as UDP is still connectionless, unreliable but keeps the size of
the packets (datagram protocol). As the most commonly used protocols are
TCP on top of IP, the network protocol suite got its nickname
TCP/IP.
- The last layer is the Application Layer which is formed by the
various applications which make use of TCP/IP. Some examples for this layer
are FTP (File Transfer Protocol), TELNET and NFS (Network File System).
The communication between different applications takes place using
sockets. A socket is an end point of a network connection used
to communicate over the network. To get an unambiguous relation between
the network traffic and the applications, each socket in the network
must be different in either local IP address, remote IP address, local port
number or remote port number. Unconnected sockets don't have a remote IP
address and remote port number, so they must differ in either the local
IP address or local port number.
One problem of the TCP/IP protocol is that if a host changes its IP
address or if a service changes its port number, all references have
to be changed all over the network. For this reason, host and service are
normally not directly referenced by the IP address and port number, but by
a host name and service name which are mapped to the corresponding IP
address and port number.
The service names are mapped to the corresponding port numbers using the
services file. This services file is found at
/etc/services under Unix. On other platforms, the location of
this file is dependent on the software being used and sometimes even on
the software setup. This file has the following syntax:
service_name port_number/protocol aliases # comments
An example services may look like this:
tcpmux 1/tcp # TCP Port Service Multiplexer
discard 9/tcp sink null # Discard
discard 9/udp sink null #
ftp-data 20/tcp # File Transfer Protocol (Data)
ftp 21/tcp # File Transfer Protocol (Control)
telnet 23/tcp # Virtual Terminal Protocol
smtp 25/tcp # Simple Mail Transfer Protocol
shr1 5010/tcp # SAS/Share server 1
shr2 5011/tcp # SAS/Share server 2
shr3 5012/tcp # SAS/Share server 3
shr4 5013/tcp # SAS/Share server 4
One note for choosing port numbers: On Unix and most other platforms there
are certain restrictions for the port numbers. Port numbers 1-1023 are
normally reserved for privilegded processes. These ports cannot be assigned
to a local end of an TCP/IP connection by a normal process. Port numbers
1024-4999 are used as so called ephemeral port numbers. These port
numbers are assigned automatically by TCP and UDP if no specific port number
was requested. Therefore, any normal user service should be mapped to port
numbers greater or equal than 5000.
Sometimes, this services file can also be distributed or referenced using
the Network Information Service (NIS).
Another mapping exists between host names and IP addresses. This mapping
can either be done using a host file or using the Domain Name
System (DNS). The hosts file is mainly used in smaller networks
that can easily be administered. The hosts file can be found in /etc/hosts under Unix. On other platforms, the location of this
file is dependent on the software being used and sometimes even on the
software setup. This file has the following syntax:
ipaddress hostname aliases # comments
An example hosts may look like this:
# This is a sample hosts file
127.0.0.1 localhost # Local loopback
192.168.99.13 jupiter.scram.de jupiter
A hosts file can also be distributed or referenced using the Network Information Service (NIS).
As an alternatice to the hosts file, a DNS lookup may be done. In Unix,
the DNS is set up using the /etc/resolv.conf file. On other
platforms, the setup is again dependent on the software and setup.
Normally, up to 3 DNS servers can be specified in this configuration file.
An example may look like this:
domain scram.de
nameserver 194.162.91.71
nameserver 194.162.162.194
- ARP
- Address Resolution Protocol. Used to map IP addresses to network
specific addresses like i.e. Ethernet addresses. Defined for Ethernet in
RFC826.
- DNS
- Domain Name System. This is an application used to map host names to
their IP addresses. Defined in RFC1034 and RFC1035.
- FTP
- File Transfer Protocol. This is a protocol being used to transfer files
over the network using TCP/IP. Defined in RFC959.
- ICMP
- Internet Control Message Protocol. Used internally by IP. All ICMP
messages are handled by the TCP/IP software internally. Used i.e. by PING.
Defined in RFC792.
- ICMPv6
- Internet Control Message Protocol for IPv6. Defined in
RFC1885.
- IMAP4
- Interactive Mail Access Protocol - Version 4. Allows access to remote email
and news storage. Defined in RFC1730.
There are multiple models supported as explained in
RFC1733
- IP
- Internet Protocol. Non-reliable, connectionless protocol with basic
internetworking features as routing and fragmentation. Defined in RFC791.
- IPv6
- Internet Protocol Version 6. Proposed Standard to replace IP. Defined in
RFC1883.
- IPL
- Initial Program Load. The boot process.
- MIME
- Multipurpose Internet Mail Extensions. Defined in RFC2045,
RFC2046, RFC2047,
RFC2048 and RFC2049.
- NAT
- Network Address Translation (RFC1631)
- NIS (also YP)
- Network Information Service. A protocol defined by sun to allow a central
store of system files like /etc/hosts, /etc/services, /etc/passwd, etc.
- NFS
- Network File System. Application that allows sharing filesystems
over a TCP/IP network. Latest version defined in RFC1813.
- OSPF
- "Open Shortest Path First". Advanced Routing algorithm and distribution
protocol. Defined in RFC2178.
- PING
- Packet InterNet Groper. This program directly accesses the IP layer for
sending ICMP ECHO_REQUEST messages to a remote machine to verify the machine
is reachable in the net.
- POP3
- Post Office Protocol - Version 3. Protocol to access email on a remote
machine. Defined in RFC1939.
- PPP
- Point to Point Protocol. Advanced Network Interface protocol used for
serial connections between LANs. Defined in RFC1661.
- RADIUS
- Remote Authentication Dial In User Service. Usually used for dial-in
authentification and accounting. Defined in RFC2138
and RFC2139.
- RARP
- Reverse Address Resolution Protocol. Used to map a network specific
address into an IP address. I.e. discless workstations and X-Terminals may
use this to get their IP address during IPL. Defined for Ethernet in RFC903.
- RIP
- Routing Information Protocol. Dynamic routing protocol. Defined in RFC1058
- SLIP
- Serial Line Internet Protocol. Network Interface Protocol used for
serial connections between LANs. Defined in RFC1055.
- TCP
- Transmission Control Protocol. Connection oriented, reliable, full
duplex, stream protocol on top of IP. Most commonly used. Defined in RFC793.
- UDP
- User Datagram Protocol. Connectionless, unreliable datagram protocol.
Used by services that use their own optimized flow control like i.e. NFS.
Defined in RFC768.
Please send your comments on this page to Jochen Friedrich (jochen@scram.de).
Jochen Friedrich
jochen@scram.de
Last Modified and Validated: Sun Jan 18 1998.