Networking Connectivity
What are forwarding modes and how do they work?

Forwarding modes
You can specify one of four possible forwarding modes for each port:

  • Cut-through
  • Fragment-free
  • Store-and-forward
  • Adaptive
Forwarding modes provide added reliability and flexibility. For example, if you are concerned about the generation of errors on a network, you can configure the ports to store-and-forward mode to ensure safe transfer of data. You can guarantee store-and-forward security on some networks, but still obtain high-speed switching with lower latency on the others.

Cut-through switching mode
Cut-through switching sends the packet to the destination as soon as the first 14 bytes of the packet are read--an approximate latency of 25 microseconds for 10Mbps devices and 7 microseconds for 100Mbps devices. The delay is minimal and the packets reach their destination in the shortest possible time.

The packets are sent through the switch as a continuous flow of data--the transmit and receive rates are always the same. Because of this, cut-through switching cannot pass packets to higher speed networks, for example, to forward packets from a 10Mbps to a 100Mbps Ethernet network.

Since the switch has forwarded most of the packet when the CRC is read, the switch cannot discard packets with CRC errors. However, the CRC check is still made and, if errors are found, the error count is updated.

Cut-through switching is recommended for networks intended to provide one switch port per user, or for lightly loaded networks. It is essential for multimedia applications and ideal for workgroup environments where minimum delays are required.

Fragment-free switching mode
Fragment-free switching is suitable for backbone applications in a congested network, or when connections are allocated to a number of users. Fragment-free switching checks that there are no collisions within the first 64 bytes of the packet--the minimum valid message size required by the IEEE 802.3 specification. This guarantees that message fragments less than 64 bytes (runts) are not forwarded to other network segments. Runts are typically the result of collision fragments.

The packets are sent through the switch as a continuous flow of data--the transmit and receive rates are always the same. Because of this, fragment-free switching cannot pass packets to higher speed networks, for example, to forward packets from a 10Mbps to a 100Mbps Ethernet network. Therefore, if you opt for fragment-free switching, you cannot make direct connections to higher speed networks from that port.

Fragment-free switching offers a compromise between cut-through (which offers the fastest possible forwarding at the expense of error checking) and store-and-forward (which offers maximum error checking at the expense of forwarding speed), to provide a latency of approximately 60 microseconds and sufficient error checking to eliminate the most common errors.

Store-and-forward switching mode
Store-and-forward switching temporarily stores a packet and checks it against the CRC field. If the packet is error free, it is forwarded; otherwise, it is discarded. Store-and-forward switching is therefore the best forwarding mode to prevent errors being forwarded throughout the network. The buffering used by store-and-forward also allows the switch to dispatch packets at a different rate than it receives them--for example, to forward packets from a 10Mbps network to higher speed networks such as a 100Mbps Ethernet.

Adaptive switching
Adaptive switching mode is a user-defined facility to maximize the efficiency of the switch. Adaptive switching starts in the default switch forwarding mode you have selected (cut-through if you selected adaptive mode as the default switching mode). Depending on the number of runts and CRC errors at that port, the mode changes to the "best" of the other two switching modes. As the numbers of runts and CRC errors change, so does the forwarding mode. This is best illustrated by the chart below:


Note: While CRC errors and runts are the most likely parameters to cause the switching mode to change, they are not the only ones.

Latency is the delay measured from the time the packet first enters a network device until it leaves it. The closer a device is to zero latency, the better. Switch latency depend on the switching mode:


The type of network can affect latency. Over wide area networks, latency is negligible in comparison to the time it takes the signals to travel over long distance lines. On local area networks, latency can influence performance.

Forwarding policy
If two communicating ports (receive port and transmit port) have different forwarding modes, then they use the "safest" mode. For example, if one port is configured for fragment-free and the other port is configured for store-and-forward, then traffic between the two ports in either direction is always switched using store-and-forward.

CRC errors
Cyclic Redundancy Check (CRC) errors are the sum of FCS, longs, very longs, alignment errors and jabbers.

A fragment is a frame consisting of only part of a packet; these can be caused by collisions on the network and are normal occurrences. Fragments have a header in the user data field to identify them.

This applies to:

Intel® InBusiness™ Hubs and Switches
Xircom® NetStation 4 Port Switch (NSE4100)
Xircom® NetStation 8 Port Switch (NSE8100)

Solution ID: CS-014410
Last Modified: 13-Jul-2010
Date Created: 09-Aug-2004
Back to Top