Internal Processing on Cisco Switches

Switches use a couple of different types of internal processing variations. Almost of the more recently released switches use store-and-forward processing, but all three types of switching are supported in at least one type of currently available Cisco switch.

Some switches, and transparent bridges in general, use store-and-forward processing. With store-and-forward, the entire frame is received by the switch before the first bit of the frame is forwarded. However, Cisco also offers two other internal processing methods for switches, called cut-through and fragment-free.

With store-and-forward processing, the switch must wait for the entire frame to be received. However, because the forwarding/filtering logic is based on the destination address, which is inside the header, the switch can make the forwarding decision before the entire frame has been received. With cut-through processing, the switch starts sending the frame out the output port before the whole frame has been received. In other words, as soon as the incoming switch port receives enough of the frame to see the destination MAC address, the forwarding decision is made and the frame is transmitted out the appropriate outgoing port to the destination device. So, each frame might experience slightly less latency.

Cut-through processing reduces latency, but it also propagates errors. Because the frame check sequence (FCS) is in the Ethernet trailer, a cut-through forwarded frame might have bit errors in it that the switch will not notice before sending most of the frame.

Fragment-free processing works similarly to cut-through, but it tries to reduce the number of errored frames that it forwards. One interesting fact about Ethernet CSMA/CD logic is that collisions should be detected within the first 64 bytes of a frame. Fragment-free processing works like cut-through logic, but it waits to receive the first 64 bytes before forwarding a frame. The frames experience less latency than with store-and-forward logic and slightly more latency than cut-through—but frames that have errors as a result of collisions are not forwarded.

With many links to the desktop running at 100 Mbps, uplinks at 1 Gbps, and faster ASICs, today's switches typically use store-and-forward processing.

The internal processing algorithms used by switches vary among models and vendors; regardless, the internal processing can be categorized as one of the methods listed in Table 9-2.

Table 9-2 Switch Internal Processing

Switching Method

Description

Store-and-forward

The switch fully receives all bits in the frame (store) before forwarding the frame (forward). This allows the switch to check the FCS before forwarding the frame. (The FCS is in the Ethernet trailer.)

Cut-through

The switch performs the address table lookup as soon as the destination address field in the header is received. The first bits in the frame can be sent out the outbound port before the final bits in the incoming frame are received. This does not allow the switch to discard frames that fail the FCS check. (FCS is in the Ethernet trailer.)

Fragment-free

This performs like cut-through switching, but the switch waits for 64 bytes to be received before forwarding the first bytes of the outgoing frame. According to Ethernet specifications, collisions should be detected during the first 64 bytes of the frame, so frames in error because of a collision will not be forwarded.

0 0

Post a comment