So far we have assumed that using a bridge instead of a repeater to connect two network segments always increases performance.  This was because the number of nodes in each segment decreases and the percentage of bandwidth for each node accordingly increases.  This is correct if the bridge propagates the intersegment traffic without significant delays.  When using bridging algorithms, however, a bridge can delay frames and, under certain circumstances, lose them.

The delay produces by a typical bridge is at least as large as the time required to put the incoming frame into the buffer.  The greater the frame size, the greater the delay.  Additionally, time is needed to consult the bridge table, especially if the bridge table is large.  The total time required by a bridge to process a packet includes not only the packet delays but also the possibility the packet will be lost.  If the time to process a packet is greater than the time until the arrival of the next packet, the incoming packet is put into the buffer where it will wait until the bridge finishes processing the previous packet.  Obviously if the average rate of incoming traffic exceeds the bridge’s ability to process packets, the bridge’s buffers will overflow and the bridge will drop incoming frames.

Discarding (or losing) a frame is very undesirable, since it cannot be recovered at the lower layers of the OSI model through the LAN protocols.  This is because the lower layer protocols have only one method to control the consistency of packets—CRC.  CRC travels inside the same packet as the data and, if the whole packet is lost, the CRC is lost as well.  Packet loss is detected only at the higher layers of OSI model, by transport and application level protocols that enumerate packets and will sense data loss and request retransmission.  But think about the processing overhead for such recovery.  In the case of regular data link layer packet loss, network performance can decrease several times over because timeouts for upper layer protocols are much longer than the transmission time at the data link layer.  The retransmission of the lost packet may occur after dozens of seconds.

To prevent packet loss, a bridge must have an average productivity higher than the average intensity of the intersegment traffic, and a large memory buffer for storing packets during peak load.  In a local area network, an often-used rule of thumb is “80/20.”  This means that a bridge should divide the network so 80% of the traffic remains local and only 20% is intersegment.  If this rule is applied for a 10 Mbps Ethernet network, a bridge must have a capacity to process 0.2×14880=3000 frames per second.

Bridges play a considerable role in network management.  Processing all the network traffic from all connected segment bridges is an excellent way to monitor and collect statistical information.  Gathered information can be displayed on the terminal connected to a bridge through its serial port, or a bridge can have an agent—a special program that works with a remote management program to provide the desired information (e.g. SNMP).

To summarize what we’ve learned about bridges—Bridges:

• Operate at the Data Link layer of the OSI model
• Have higher processing overhead that repeaters
• Segment networks into collision domains
• Propagate broadcast messages
• Can connect segments with different media access control methods
• Make intelligent decisions based on a bridging table
• Can improve network performance by isolating intrasegment network traffic
• May overflow during excessive network traffic