Internetworking

The Internet

• 1969 - firsts hosts at UCLA
• 1982 - TCP/IP standardized
• 1991 - WWW started at CERN (Swiss Physics Laboratory)

* - In 1998, a new method was developed for more accurately determining the number of hosts.

Internet Architecture

+-----------+        +--------+        +-----------+
| Network 1 |--------| Router |--------| Network 2 |
+-----------+   ^    +--------+   ^    +-----------+
           Conection to      Connection to
	    Network 1          Network 2

Tracing the routes packets take from WVU to Berkeley in CA, we see that packets pass through 17 routers (or hops) in all. Of these, the distribution breaks down like so:

• 2 routers at WVU
• 2 routers at WVNet
• 5 routers at AlterNet
• 4 routers at SprintLink
• 4 routers at Berkeley

• Open System. The standards running the system must be public.
• Fault Tolerance. Decentralized control.
• Ability to be quickly deployed.

• Scope: solve as general a problem as possible.
• Scalability: work efficiently with large networks and small networks.
• Robustness: tolerant to failures of links and nodes. This usually includes these types of subproperties.
  • Firewalls: disruptions only affect a portion of the network.
  • Self-Stabilization: corruption should eventually "flush" out and the network return to normal operation without human intervention and in a reasonable amount of time.
  • Fault Detection: detect faults in a timely manner. Some degree should exist for fault detection in all networks.
  • Byzantine Robustness: ability to detect ongoing corrupting agents. Very difficult to do and usually not included in most networks.
• Autoconfigurability: come with reasonable default values for most environments.
• Tweakability: ability to optimize performance through variable adjustments.
• Determinism: identical conditions give identical results.
• Migration: ability to allow modifications easily.

The International Standards Organization (ISO) came up with a "complete" 7-layer network description. The system is called Open System Interconnection (OSI). TCP/IP, however, is typically thought of as having only 4 layers. The OSI layers, there responsibilities, and the corresponding TCP/IP layers are given below. Each higher layer builds on the layers below it to provide its services.

+----------------------+
|  Application Layer * |
+----------------------+
| Presentation Layer   |
+----------------------+
|   Session Layer      |
+----------------------+
|  Transport Layer   * |
+----------------------+
|   Network Layer    * |
+----------------------+
|  Data Link Layer   * |
+----------------------+
|   Physical Layer     |
+----------------------+

* - Those layers also present in TCP/IP. TCP/IP combines the Physical and Data Link layers into a single Link layer. TCP/IP also combines the Session layer into the Transport layer.

Some Handy definitions:

• Routers connect network at the Network layer
• Bridges connect networks at the Link layer(s)

Usually, protocols interact at the same layer. We will be looking at the TCP/IP protocols in various layers. First the Link layer. In the Link layer, the operaiton of Ethernet, Token Ring, and other technologies would be defined. In the Network layer, we will look at two protocols:

• Internet Protocol (IP)
  • Defines the "datagram", the basic unit of transmission in the Internet. Datagrams will be of varying sizes base don what technology it is running on (Ethernet, Token Ring, phone line, etc.). The size of a datagram on a technology is called the Maximum Transmission Unit (MTU) of the technology. Typically, the slower the media, the smaller the MTU.
  • Defines the Internet Addressing scheme.
  • Defines how datagrams are routed.
  • Defines fragmentation (spliting up a single datagram into multiple datagrams) and reassembly (putting multiple datagrams back into a single datagram).
  • IP is unreliable. It does not guarantee a datagram arrives to its eventual destination. It does its best, but it is not guaranteed.
  • IP is connectionless. It does not contain a handshake between the source and destination. It also does not exchange state information between the source and destination and successive transmissions are treated separately by routers.
• Internet Control Message Protocol (ICMP)
  • Uses IP to send and receive datagrams
  • Used to send control messages from a destination back to a source. Such control messages as "unreachability" messages for host, network, and port unreachable.
  • Used to redirect routes for datagrams
  • Used for diagnostics such as ping and traceroute.

• User Datagram Protocol (UDP)
  • Uses IP to send and receive datagrams
  • UDP is unreliable (just like IP)
  • UDP is connectionless (just like IP)
  • "lighter weight" than TCP because it is unreliable and connectionless
  • Uses 16-bit port number in datagrams for demultiplexing into applications.
• Transmission Control Protocol (TCP)
  • Uses IP to send and receive datagrams
  • Adds reliable delivery through Positive Acknowledgements. The receiver sends a message back to the source letting it know it received the datagram.
  • TCP is connection-oriented. It has a handshake/teardown for session/connection management. The source and destination continually exchange state information.
  • TCP has a byte-stream abstraction. This is a network "pipe" abstraction from the users point of view, a continuous stream of data.
  • TCP is full-duplex. Data can flow in both directions.
  • Uses 16-bit port number in datagrams for demultiplexing into applications.

Internet Addresses

 1 7 bits        24 bits
+-+------+--------------------+
|0|Net ID|       Host ID      |
+-+------+--------------------+

 1 1   14 bits      16 bits
+-+-+-----------+-------------+
|1|0|  Net ID   |   Host ID   |
+-+-+-----------+-------------+

 1 1 1     21 bits     8 bits
+-+-+-+--------------+--------+
|1|1|0|    Net ID    | Host ID|
+-+-+-+--------------+--------+

 1 1 1 1        28 bits
+-+-+-+-+---------------------+
|1|1|1|0|  Multicast Group    |
+-+-+-+-+---------------------+

 1 1 1 1 1       27 bits
+-+-+-+-+-+-------------------+
|1|1|1|1|0|     Reserved      |
+-+-+-+-+-+-------------------+

The class of an IP address specifies where in the address the network ID is and where the host ID. The Net ID specifies the network that the address is connected to. The Host ID specifies the ID of the address on the network. Addresses are allocated by a central agency called the InterNIC, the Internet Network Information Center, to entities desiring an Internet address. Internet addresses are tied to the type of transmission the address specifies. The types of transmissions are.

• Unicsat: transmit to single destination
• Multicast: transmit to a subset of all hosts
• Broadcast: transmit to all hosts

The Domain Name System (DNS) provides a distributed database that provides a mapping between text string representing IP addresses (hostnames) and the IP address value.

IP addresses specify network connections. A multihomed machine will have multiple network connections and therefore will have an IP address for every interface it has. Strictly speaking, it is therefore incorrect to say that an IP address specifies a single host. However, this is generally a true statement.

This addressing scheme allows efficient routing because the network information is encoded in the address. However, this also makes it difficult to move a host across networks. When this happens, the IP address must change. This is normally not a problem unless the host must operate while the change must occur.

Subnet Addressing (Netmasks)

It should be obvious that the class A and class B addresses have a large amount of hosts per network. Normal networks can not have or do not want to have so many hosts on them. This has lead the use of subnet addressing. The concept is to split the hostid part up into a Subnet ID and a Host ID. The Host ID part fills the same function. The Subnet ID part specifies the subnetwork (within the larger network) that the host resides. Subnet addresses are specified as a 32-bit mask value where 1's signify Net ID and Subnet ID and where 0's signify Host ID. Netmasks are not allocated and need not be unique within the Internet. Netmasks are usually specified as either a hex value, FFFFFF00, or in dotted decimal form, 255.255.255.0. As an example, use the IP address of naur.csee.wvu.edu, 157.182.194.28. The netmask used by naur and others on the same subnet is 255.255.255.0 or FFFFFF00. The class of naur's IP address is B. So, the Net ID portion is 14 bits, the Subnet ID is the 8 bits between the Net ID and the Host ID. The Host ID part is also 8 bits, the last 8 bits more specifically. An alternate way to specify the netmask and IP address together is by following the IP address by a / and the number of leading 1's in the netmask. As an example, naur would be 157.182.194.28/24. The 24 stands for 24 leading 1's in the netmask.

Given a sources IP address and netmask, it is easy to determine any other hosts relative location from its IP address. The choices are:

• The host may be on the same subnet as the source.
• The host may be on another subnet, but the same network as the source.
• The host may be on another network than the source.

To better understand subnetting, let's look at typical subnet masks and how many networks and nodes per network we have given any Class C address. (The nodes per network assumes special address given below).

The Internet address space has several addresses that have special meanings and can not be reserved. A general idium you can use is: 0 means "this" and 1 means "all".

Broadcast Addresses

Broadcasts have several different resolutions. These are below. Broadcast addresses may only be used as destination addresses.

• net-directed broadcast: An A, B, or C address with its Host ID and Subnet ID part set to all 1's. This sends to all of the hosts on the same network.
• subnet-directed broadcast: An A, B, or C address with its Host ID part set to all 1's. This sends to all of the hosts on the same subnet.
• limited broadcast: The IP address 255.255.255.255 (all 1's). This sends to all of the hosts on the attached network. It may never be forwarded by routers. This address is used if the machine does not know its IP address or netmask as is possible during bootstrapping.

Normally, the source address used in a broadcast is the address of the host sending the broadcast. However, if the host does not know its IP address (due to initialization, for example), two additional addresses are specially reserved. These are:

• 0.0.0.0 indicates the current host on the current network.
• Net ID set to 0, Host ID not 0 (and not all 1's) indicates the designated host on the current network.

Loopback Addresses

TCP/IP is designed to work transparently no matter where the source and destination are located. A whole class A network is reserved to indicate the current host. It may be used as a source or destination address. The class A network address 127.0.0.0 is reserved to indicate the current machine and may not be assigned. Packets destined for 127.0.0.0 will not go onto the network and packets coming from 127.0.0.0 have not come from the network. Most systems use a specific host on the 127.0.0.0 network to signify loopback. Typically, this is 127.0.0.1, but it may be different on some systems.

IPv6, the next version of IP, modifies addressing by making addresses 128-bits in length. There are no broadcast transmissions, it uses multicast instead. And IPv6 introduces the concept of Anycasting.

Port Numbers and Demultiplexing

• FTP (TCP port 21)
• Telnet (TCP port 23)
• TFTP (UDP port 69)
• echo (TCP port 7 and UDP port 7)
• daytime (TCP port 13 and UDP port 13)
• HTTP (TCP port 80)

Source ports are "ephemeral" (or short lived, temporary) identifiers and may be recycled over a long period of time. Ports are used internally to demultiplex packets from the protocol processing software to the applications.

Demultiplexing

Demultiplexing is how a protocol stack fits together. At the lowest level, the Link layer, a module reads in a packet. Within that packet is a field that indicates which protocol it is destined for (say IP), the module forwards that packet into the IP module for processing. The IP module looks further in the packet and examines a field, the IP protocol field, and sends the packet out to the ICMP, UDP, or TCP modules based on the value in the field. The TCP and UDP modules then examine the port numbers and sends the packet out to various applications based on the port numbers.