Previous | Table of Contents | Next |
The principles of operation for switched virtual networks are concretely founded in the success of the global communications systems. Without doubt, the phone system is the worlds largest and most reliable network. Built using advanced digital switches controlled by software, extensive accounting and management tools ensure the success of this highly effective means of communication. The connection-oriented switch is the key. End-to-end connections across multiple switches and various transmission types ranging from copper to fiber optics to microwave to satellites allow millions of calls per day to be successfully completed, regardless of the type of phone or where the user is calling from. The telephony model is used throughout this chapter to help illustrate the workings of a virtual network.
One of the more confusing terms in the networking industry today is the word switch. For the purpose of this chapter, switching can be broken down into three fundamental areas:
The earliest form of switching enabled the network manager to assign an individual port or an entire group of ports to a particular backplane segment within an intelligent hub device. This port configuration switching allowed the logical grouping of users onto a particular segment without the need to physically travel to the wiring closet to move cables or connectors. In a sense, this offers an electronic patch panel function. Although the benefit is a reduction of moves, adds, and change costs, this advantage can only be realized within the confines of a single hub. The application of this type of switching is limited because it cannot extend beyond one intelligent concentrator. Although beneficial in the work group, the enterprise needs cannot be met.
Phone system operators in the 1940s manually patched user connections through to destinations and recorded call time and duration. Using configuration switching is similar to patching phone lines together. Just as the phone network grew at a pace that required the switching to be performed automatically without operator intervention, so too have data networks outgrown the limitations of configuration switching.
Packet switching isolates each port to deliver dedicated bandwidth to each user in the network. Fundamentally, a packet switch is any device that accepts an incoming packet on one port and then makes a decision whether to filter or forward the packet out another interface. There are two types of packet switch transports: connectionless and connection-oriented.
Connectionless devices are probably more familiar to network professionals when described as bridges or routers. A bridge is a layer 2 (of the OSI reference model) switch that bases its decisions on the MAC address of attached workstations. What many vendors describe as a switch is actually a wire-speed MAC layer bridge. Three methods of decision making in these types of devices are cut-through, modified cut-through, and store-and-forward.
The Cut-through Switch. This switch reads a packet to the destination address before it starts forwarding to the outbound interface. The benefit is an extremely low latency or delay in the forwarding of packets. The penalty is the propagation of errors, because the frame is being forwarded before it can be verified as valid, and the inability to support interfaces of different speeds that prevents high-bandwidth uplink of FDDI or ATM on these type of devices.
The Modified Cut-through Switch. This switch reads the first 64 bytes of a frame and then starts forwarding to the outbound interface, which greatly reduces the chances of propagating errored frames throughout the network. However, this method still requires all ports to be of the same type and speed.
Store-and-Forward Switch. The most flexible switch design uses a store-and-forward methodology that reads the entire frame before any filtering or forwarding decisions are made, thus ensuring that only packets that are not errored are forwarded on the network. This method also allows packets to be buffered when transferring data between networks of different types, such as Ethernet to FDDI or ATM.
Bridges and Routers. A router is a layer 3 switch that bases its decisions on the network protocol address of attached workstations. Bridges and routers are considered connectionless because they forward and forget, requiring a decision to be made on every single inbound packet. The performance implications are that even though two communicating nodes on opposite sides of a bridge or router may be the only devices on their respective networks, the bridge or router must continuously make filter or forward decisions on every packet sent between the two nodes. If the phone network were built using bridges or routers, users would have to hang up and redial their destination after every word, which is not a very practical proposition.
A connectionless transport is not capable of defining which path its payload will take, cannot guarantee delivery, and is generally slower than a connection-oriented system. When a node sends a packet through a bridged or routed network, it is analogous to dropping a letter into a mailbox. It is not apparent how the letter got to its destination. The arrival of a letter cannot be guaranteed (protocol prioritization techniques are comparable to sending a letter by express mail). If a letter is lost (or a packet dropped), determining where it was lost is often difficult. The only way the sender knows that the letter was received is if the recipient sends another letter back to the sender (i.e., frame acknowledgment).
In a sense, todays shared-access networks are like the party lines of the early telephone network. But just as the phone network evolved from party lines to dedicated lines as usage and deployment grew, so too must the data networks offer this same level of service guarantee and broad adoption.
The connection-oriented switch that the phone systems use offers immediate acknowledgment of communications when the person picks up at the other end. The exact path the call took as well as its time and duration can be logged. The destination only needs to be dialed only once and information is exchanged until both parties hang up.
Previous | Table of Contents | Next |