The Networking King: Ethernet

CP and IP have made the interconnection of the world possible, but Ethernet must be given a silver medal for building up the network from the ground up. Without Ethernet we would have never evolved organizational networks and the Internet so quickly. For anyone who has used a modem from home, and had to make a dial-up connection, will know how annoying this can be. But, Ethernet, plug in a cable from a hub to the computer, and it works seamlessly. No setting dial-ups, no screeching, no flashing lights, no tying up your phone line, no telephone bills, no bringing down the network while you connect or disconnect. Nothing. You don't even have to know what the physical address of the computer is. So how does it work? Well the key is ARP, as ARP allows a computer to broadcast a message to the rest of the network, asking for the MAC address of a given network address. Thus computers can quickly determine the physical addresses of all the devices on their network segment, simply by broadcasting an ARP request. So what if the destination is outside the network segment? Well with this the computer sends the data frame to the router (as it will know its address from a previous ARP request), but with the destination network address. The router detects that the destination address is outside the network segment, and that the data frame is addressed to itself. It will then forward to another router, or a network segment.

Until recently, it seemed unlikely that Ethernet would survive as a provider of network backbones and for campus networks, and its domain would stay, in the short-term, with connections to local computers. The world seemed distended for the global domination of ATM , the true integrator of real-time and non real-time data . This was due to Ethernet's lack of support for real-time traffic and that it does not cope well with traffic rates that approach the maximum bandwidth of a segment (as the number of collisions increases with the amount of traffic on a segment). ATM seemed to be the logical choice as it analyses the type of data being transmitted and reserves a route for the given quality of service. It looked as if ATM would migrate down from large-scale networks to the connection of computers, telephones, and all types of analogue /digital communications equipment. But, remember, not always the best technological solution wins the battle for the market - a specialist is normally always trumped by a good all-rounder.

Ethernet is the best poker player in town. It knows all the tricks. It's a heavyweight prize fighter. It'll slug it out with anyone, and win. I took Token Ring on, head to head, and thrashed it. So what would you choose for your corporate network? Would it be a technology that was cheap, and could give you 1 Mbps, 10Mbps for your connections to workstations and server, and, possibly, 1Gbps for your backbone. Ethernet always makes a sensible choice, as it's cheap and it's going to be around for a lot longer, yet. Any problems within an Ethernet network can be solved by segmenting the network, and by relocating servers. And for cable , it supports twisted-pair, coaxial and fiber. Who would have believed that you could get 1Gbps down a standard Cat-5 , twisted-pair cable. Amazing.

Ethernet also does not provide for quality of service and requires other higher-level protocols, such as IEEE 802.1p. These disadvantages are often outweighed by its simplicity, its upgradeability, its reliability and its compatibility. One way to overcome the contention problem is to provide a large enough bandwidth so that the network is not swamped by sources which burst data onto the network. For this, the gigabit Ethernet standard is likely to be the best solution for most networks.

A key method of increasing the bandwidth of a network is to replace hubs with switches, as switches allow simultaneous transmission between connected ports. Thus if the bandwidth of a single port on a switch is 10 Mbps, then a multi-port switch can give a throughput of several times this. But, switches have the potential of improving the configuration of networks.

Many workers are now used to open-plan offices, where the physical environment can be changed as workgroup evolve. This is a concept which is now appearing in networking, where virtual networks are created. With this computers connect to switches. The switch then tags data frames for destination virtual networks and puts the tagged data frame onto the backbone. Other switches then read the tag, and, if the destination is connected to one of their ports, they remove the data tag, and forward the data frame to the required port. This technique is now standardized with IEEE 802.1q, an important step in getting any networking technique accepted. Imagine if whole countries were setup like this. What we would have is a programmable network, where system administrators could connect any computer to any network. Presently we are constrained by the physical location of nodes.

Virtual networks will also bring enhanced security, where it will be possible to constrain the access to sensitive data. For example a server which contains data which must be kept secret can be located in a safe physical environment and only users which a valid MAC address would be allowed access to the data.

Hats off to the IEEE who have carefully developed the basic technology, after its initial conception by DEC, Intel and the Xerox Corporation.