When we looked at bridging in the previous chapter, I mentioned that a bridge would take the time to calculate the CRC value of every frame that it intended to forward. While this is true for a bridge, on Cisco switches you can configure the different methods in which switching will occur. Certainly a network will be more reliable if every frame has its CRC value calculated by the switch. However, this also introduces a degree of delay (or latency) to the network.
Network equipment today is more reliable than ever before. On a well-planned network using quality equipment, frame corruption shouldn’t be a huge issue. Because of this, you might consider eliminating the CRC check on frames to take advantage of higher forwarding rates. This can be done on Cisco switches (including the 1900) by configuring the method by which the switch forwards frames. The three frame forwarding methods supported by Cisco include:
Store-and-Forward switching is best described as traditional frame forwarding. When used, the switch will calculate the CRC value for each and every frame. When a frame enters a port, the switch will copy the entire frame onto its buffers and will calculate the CRC. If the CRC calculation shows the frame not to be corrupt, it will be forwarded to the destination port according to the information found in the switch’s MAC address table. In situations where the frame is corrupt, it will be dropped. Because Ethernet frames can be different sizes, the latency introduced by the Store-and-Forward method will vary according to the size of a frame. Latency is the time it takes the device to send the packet from the source to its destination. Latency coupled with bandwidth defines the network speed and the capacity of that network. Store-and-Forward switching is the slowest method of the three outlined in this section, but is the default on high-speed Catalyst switches like those in the 5000 and 6000 series.
Cut-Trough switching is the fastest switching method offered by Cisco. Cut-Through doesn’t copy the entire frame to its buffers, but instead begins the forwarding process immediately once the first six bytes of the frame have entered the switch (remember that the first six bytes provide destination MAC address information). Because CRC values aren’t calculated, Cut-Through switches may very well forward corrupted frames. This is the tradeoff – you give up the added reliability of the additional CRC check, but gain in terms of faster frame forwarding rates. Another version of this switching method is referred to as adaptive Cut-Through. This involves a switch using Cut-Through until an error threshold is encountered. Once reached, the switching method automatically changes to Store-and-Forward.
FragmentFree is a variation on the Cut-Through method described previously. Instead of initiating the forwarding process after the first six bytes of a frame, a switch configured to use FragmentFree waits until the first 64 bytes have been received, making sure that they are properly formatted. If they are, the switch will forward the frame to the port on which the destination MAC address resides.