Ethernet Switching

Bridging and Frame Switching are practically one and the same technology. Frame Switching is Bridging that has been speeded up. Bridging has always been software-based and normally a bridge would just have two ports used to connect the two LANs being bridged. Switching is hardware-based and has many ports but all the rules that apply to bridging also apply to switching and more besides. The MAC address is always left unchanged in bridging (barring the bit ordering change in Translational bridging!).

LAN Frame switches can include FDDI, Token Ring or Ethernet switches. Effectively, the switch provides single Collision Domains per switch port and each port acts as a bridge port to the rest of the network. Forwarding tables are kept per port, different media, different speeds etc. can be configured on a port by port basis. The speed enhancement to the network is achieved through the 'microsegmentation' of the large Collision Domain into many smaller ones. Each port on an Ethernet switch is effectively a very fast bridge port. The switch itself has its own MAC address e.g. 0800.a300.df00, and then each of its ports is given a MAC address, commonly in port order; port 1 has address 0800.a300.df01, port 2 has address 0800.a300.df02 etc. If this particular switch is the root bridge, then the MAC address 0800.a300.df00 is advertised as the root bridge, however the BPDUs originate from whatever MAC address is assigned to the port from which the BPDU emanates.

Some switches allow you to implement a Backpressure scheme whereby, on a particular port, jamming frames can be sent to reduce traffic coming into the switch. This stops one port hogging the backplane on a switch thereby effecting other users. Obviously, you would not wish to implement this on a server port, since this will affect many people and you would wish to keep as much of the switch processing capability for the attached servers. This is why so much play is made of the backplane capability of a particular manufacturer's switch.

A Cut-through switch first reads the Destination address of a frame and then sends the frame straight to the destination before the rest of the frame has arrived at the switch. The first 20 to 30 bytes of the frame need to be read to make sure that the frame is not a collision fragment. If the destination address remains unknown, then the switch temporarily stores the frame. Cut-through switching is fine for fixed speed networks such as all 10BaseT, and it is very fast, however if the switch has mixed speed ports such as 10/100 autosensing ports, then there is a bottle neck when packets are moving across the switch fabric from a 100BaseT segment to a 10BaseT segment. Some switches, although they forward the frame as soon as they read the destination address they still read the frame up to the CRC and if there are a certain level of errors, they can be configured to automatically change to a Store and Forward mode.

A Store and Forward switch, or 'buffered switch', stores each frame frame in a buffer before forwarding it on to the appropriate port. This gets around the underflow or overflow situation that could happen in a mixed speed environment.

This is similar to Cut Through Switching but here the frame is checked a little further than the destination address to the Length field in order to weed out collision fragments, before it is forwarded.

Latency of a network increases as the network gets busier. On a busy network, the backoffs (retransmits) that could occur with Cut-through switches increase, thereby increasing latency. A Store and Forward switch on 10Mbps LAN delays a frame by one frame time, obviously increasing latency, but there are no backoffs.

Go next to Ethernet VLANs.

Home