One of three transmission methods is used to move frames from source to destination: unicast, multicast, or broadcast. Unicast transmission occurs when there is a direct path from source to destination, a "one-to-one" relationship. Multicast has a one-to-many relationship in which the frame is delivered to multiple destinations that are identified as part of a multicast group. Broadcast is a one-to-all relationship in which the frame is delivered to all the hosts on the network segment, whether or not they want the traffic.
Frame size is measured in bytes and has a minimum and maximum length, depending on the implemented technology, such as Ethernet, Token Ring, or with WAN technologies (such as Frame Relay or IP VPN). The maximum frame length supported by a technology is called the maximum transmission unit, or MTU, and is measured in bytes. A frame received by the switch that is less than the minimum frame length for that technology is called a runt, and a frame greater than the maximum frame length is called a giant. Giant frames must be fragmented into smaller frames, smaller than the acceptable MTU, before these frames can be forwarded across the switch's or router's network interface.
There are two common categories of switches: store-and-forward switches and cut-through switches. Store-and-forward switching accepts the complete frame into the switch buffers for error checking before forwarding on to the network. Cut-through switching reads just the destination MAC address (the first 6 bytes of the frame following the preamble) to determine the switch port to forward the traffic. Store-and-forward switching adds some delay to the time it takes for the frame to get from source to destination; unlike cut-through switching, however, store-and-forward switching does not forward a frame with errors. The delay added by store-and-forward switching is minimal and should not be a determining factor when deciding between using cut-through and store-and-forward switching. Store-and-forward has an advantage over cut-through switching by virtue of its error-handling mechanisms.
A third switching category is fragment-free switching, which accepts the first 64 bytes of the frame and checks for errors. Fragment-free switching works on the precept that if there are any errors on the line, they are detectable within the first 64 bytes of the frame.
The fundamental difference between Layer 2 and Layer 3 switch operation is the layer at which each forwarding decision is made. Layer 2 switches make their forwarding decisions based on tables that store the mapping between MAC addresses and switch ports. Layer 3 switches build a table of network addresses and switch ports, making the forwarding decisions based on the network address information found in Layer 3, rather than just the MAC address found in Layer 2. Layer 3 switches function like routers because of the similar Layer 3 forwarding decision handling. However, Layer 3 switches tend to have better throughput because of the hardware processing of the address tables rather than the software.