A number of types of NIC have been used including ARCNet, Ethernet, Token Ring, etc. In general, NIC has come to mean either 10mbps or 100mbps Ethernet, although Token Ring is widely used in mainframe installations.
NICs represent the bottom layers of the OSI data transfer model -- the physical layers that transmit data between computers. Additional, and very confusing, layers of software are required to implement a real functioning network. Several protocols are in common use -- primarily IPX/SPX (Novell default), TCP/IP (Unix/Internet) and NBF-NETBEUI (Microsoft software). Some interconnectivity exists.
The critical performance parameter for an individual NIC is the transfer speed -- usually 10Million bits per second or 100 million bits per second. However, the network requires control information so that each bunch of bytes sent is augmented by additional bytes that control routing. Ethernet operates on a Carrier Sense Multiple Access (CSMA) basis. In simple terms, this means that a card transmits data then checks to see if someone else was also transmitting. Thus, the heavier the network load, and the more nodes, the more collisions and the higher the retransmission rate. It is generally felt that Ethernet networks peak out at about 60% of rated capacity.
Typical NIC data rates for 10mbps networks as of 1997 are between 220 and 550 KBytes/sec. For networks using PCI NICs and higher speed CPUs/buses, rates may be somewhat higher. 600 KBytes/sec is a recommended figure for planning lightly loaded networks, 450 for more heavily loaded networks.
NICs almost always require an interrupt (IRQ). For some reason, this is typically set by default to IRQ3 which conflicts the COM2 on almost all PCs. NICs also require a set of I/O port addresses.
Return To Index Copyright 1994-2008 by Donald Kenney.