SMAC May 1999 Durkin - 5

cable, 100BaseT copper wire, or fiber-optic strand) that carries information from node to node. Information is transmitted in the medium as discrete framed data packets. As a baseband medium, Ethernet can only transmit one packet at a time. The job of the CSMA/CD MAC on each node is to delay sending its data packet until it senses that the medium is clear of any other transmissions. Once the MAC perceives the line is clear, it transmits its data. In the split-millisecond world of high-speed networking, however, a line can look clear to multiple nodes at the same time. Inevitably, some of the nodes will transmit simultaneously. And when that happens, data packets collide and the information is lost. When a CSMA/CD MAC detects a collision, it backs off a certain increment of time before attempting to retransmit the lost data. The MAC generates the "backoff” time using pseudo-random numbers. Thus, each of the nodes involved in a collision will wait a different length of time to retransmit so they won't collide again. Each time there is another collision, the MAC backs off a little longer until the data eventually gets through.

Putting CSMA/CD in Its Place

The CSMA/CD MAC protocol exists in most networked computers on an Ethernet NIC (network interface card). The protocol is implemented by semiconductors and software resident on the card. Its operation is so deeply embedded within the network layers that its operation is completely transparent to the user. Although installing an Ethernet NIC into a computer is quite simple, the implementation of the CSMA/CD MAC protocol on that NIC is extremely complex. It is so complex, in fact, that there have been few challengers to its supremacy over the past 25 years. IBM's "token ring" MAC has carved out a niche in the market. And FDDI uses a variation of the token ring MAC. Nevertheless, the consensus is that more than 80 percent of the world's computer networks use the Ethernet CSMA/CD MAC.