5.5
Collisions and Collision Domains in Shared Layer Environments
5.5.9 The four repeater rule
The four repeater rule in Ethernet states, that no more than four repeaters or repeating hubs can be between any two computers on the network. To assure that a repeated 10BASE-T network will function properly, the following condition must be true: (repeater delays + cable delays + NIC delays) x 2 < maximum round-trip delay. Repeater delays for 10BASE-T are usually less than 2 microseconds per repeater; cable delays are near 0.55 microseconds per 100 m trip; NIC delays are about 1 microsecond per NIC; and the maximum round-trip delay (the 10BASE-T bit time of 0.1 microseconds times the minimum frame size of 512 bits) is 51.2 microseconds. For a 500 m length of UTP connected by 4 repeaters (hubs) and 2 NIC delays the total delay would be well below the maximum round-trip delay. Repeater latency, propagation delay, and NIC latency all contribute to the 4-repeater rule. Exceeding the four repeater rule can lead to violating the maximum delay limit.

When this delay limit is exceeded, the number of late collisions dramatically increase. A late collision, is when a collision happens after the first 64 bytes of the frame are transmitted. The chipsets in NICs are not required to retransmit automatically when a late collision occurs. These late collision frames add delay referred to as consumption delay. As consumption delay and latency increase, network performance decreases. This Ethernet rule of thumb is also known as the 5-4-3-2-1 rule. Five sections of the network, four repeaters or hubs, three sections of the network are "mixing" sections (with hosts), two sections are link sections (for link purposes), and one large collision domain.