QoS - Tail Drop

When applying various QoS mechanisms on networks, the term "tail drop" refers to a particular behavior of how newly arriving packets are handled.

  1. Basic Concept: Tail drop occurs in network devices, like routers and switches, when their data buffers become full. These buffers temporarily hold packets before they are forwarded to their next destination.
  2. Behavior: When the buffer is full and new packets arrive, the packets at the "tail" of the queue (the most recently arrived packets) are dropped. This is because there is no space left in the buffer to accommodate them.
  3. Impact on Traffic: Tail drop can lead to TCP synchronization and global synchronization issues. TCP synchronization occurs when multiple TCP flows reduce their transmission rates in response to packet loss, and then increase their rates again in unison, leading to a cyclical pattern of congestion and packet loss.
  4. In combination with QoS mechanisms: Tail drop if implemented alone is a more passive and less sophisticated method of dealing with congestion. However, the phenomenon may also occur when other QoS techniques are applied like shaping and queuing. Shaping controls the rate of traffic entering the network buffering excess traffic, while various different queuing techniques prioritize traffic to manage congestion more effectively. Tail drop may still occur in these cases if the excess traffic is extensive and available buffers are full.
  5. Alternatives: To address the limitations of tail drop, more advanced queue management algorithms like Random Early Detection (RED) or Weighted Random Early Detection (WRED) are used. These algorithms proactively manage buffer space and mitigate the issue of TCP synchronization by dropping packets before the buffer becomes completely full, based on certain criteria. Policing is another alternative which eliminates tail drop simply because any excess traffic is not buffered but immediately dropped.

Tail drop is a basic congestion management mechanism that drops the most recent packets when the buffer is full, potentially leading to issues like TCP synchronization in networks. More advanced techniques like RED and WRED are often preferred for better traffic management in scenarios where QoS is a significant concern.