Fault Tolerant Network Architecture

The Internet, in its early inception, was the result of research funded by the United States Department of Defense (DoD). Its primary goal was to have a communications medium that could withstand the destruction of numerous sites and transmission facilities without disruption of service. It only follows that fault tolerance was the focus of the effort of the initial internetwork design work. Early network researchers looked at the existing communication networks, which were primarily for the transmission of voice traffic, to determine what could be done to improve the fault tolerance level.

Circuit Switched Connection-oriented Networks

To understand the challenge that the DoD researchers were faced with, it is necessary to look at how early telephone systems work. When a person makes a call using a traditional telephone set, the call first goes through a setup process, where all of the telephone switching locations between the person and the phone set that they are calling are identified. A temporary path, or circuit, is created through the various switching locations to use for the duration of the telephone call. If any link or device participating in the circuit fails, the call is dropped. To reconnect, a new call must be made, and a new circuit created between the source telephone set and the destination. This type of connection-oriented network is called a circuit-switched network. Early circuit switched networks did not dynamically recreate dropped circuits. In order to recover from failure, new calls had to be initiated and new circuits built end-to-end.

Many circuit switched networks give priority to maintaining existing circuit connections, at the expense of new circuit requests. In this type of connection-oriented network, once a circuit is established, even if no communication is occurring between the persons on either end of the call, the circuit remains connected and resources reserved until one of the parties disconnects the call. Since there is a finite capacity to create new circuits, it is possible to occasionally get a message that all circuits are busy and a call cannot be placed. The cost to create many alternate paths with enough capacity to support a large number of simultaneous circuits, and the technologies necessary to dynamically recreate dropped circuits in the event of a failure, led the DoD to consider other types of networks.

Packet Switched Connectionless Networks

In the search for a network that could withstand the loss of a significant amount of its transmission and switching facilities, the early Internet designers reevaluated early research regarding packet switched networks. The premise for this type of networks is that a single message can be broken into multiple message blocks. Individual blocks containing addressing information indicates both their origination point and their final destination. Using this embedded information, these message blocks, called packets, can be sent through the network along various paths, and can be reassembled into the original message upon reaching their destination.

Utilizing Packets

The devices within the network itself are unaware of the content of the individual packets, only visible is the address of the final destination and the next device in the path to that destination. No reserved circuit is built between sender and receiver. Each packet is sent independently from one switching location to another. At each location, a routing decision is made as to which path to use to forward the packet towards its final destination. If a previously used path is no longer available, the routing function can dynamically choose the next best available path. Because the messages are sent in pieces, rather than as a single complete message, the few packets that may be lost in the advent of a failure can be retransmitted to the destination along a different path. In many cases, the destination device is unaware that any failure or rerouting has occurred.

Packet-switched Connectionless Networks

The DoD researchers realized that a packet switched connectionless network had the features necessary to support a resilient, fault tolerant network architecture. The need for a single, reserved circuit from end-to-end does not exist in a packet switched network. Any piece of a message can be sent through the network using any available path. Packets containing pieces of messages from different sources can travel the network at the same time. The problem of underutilized or idle circuits is eliminated -- all available resources can be used at any time to deliver packets to their final destination. By providing a method to dynamically use redundant paths, without intervention by the user, the Internet has become a fault tolerant, scalable method of communications.

Connection-oriented Networks

Although packet-switched connectionless networks met the needs of the DoD, and continue to be the primary infrastructure for today's Internet, there are some benefits to a connection-oriented system like the circuit-switched telephone system. Because resources at the various switching locations are dedicated to providing a finite number of circuits, the quality and consistency of messages transmitted across a connection-oriented network can be guaranteed. Another benefit is that the provider of the service can charge the users of the network for the period of time that the connection is active. The ability to charge users for active connections through the network is a fundamental premise of the telecommunication service industry.