5.3QoS & bandwidth control

When file transfers and VoIP calls share the same network link, large volumes of file-transfer traffic can delay VoIP packets, risking dropped or garbled calls. QoS (Quality of Service) is a set of technologies for preferentially processing the traffic that should be prioritized when traffic of differing characteristics coexists like this. This section focuses on the design decisions behind guaranteeing voice/video quality within limited bandwidth.

5.3.1Why QoS is needed

• A best-effort IP network is inherently designed to make no guarantee about the order or timing of delivery, so during congestion all traffic is affected equally by delay and loss. However, real-time traffic such as voice and video is sensitive to latency, jitter, and packet loss (delays or losses of a few hundred milliseconds severely degrade perceived quality), whereas file transfer tolerates some delay (it only needs to complete eventually). Ignoring this difference in characteristics and treating all traffic equally sacrifices the quality of real-time communication.

5.3.2Priority control (priority queueing)

• Priority control (priority queueing) sorts traffic into multiple queues by priority and processes the highest-priority queue first. Strict priority processes a lower-priority queue only once every higher-priority queue is empty, maximally protecting the quality of high-priority traffic—but if high-priority traffic volume is heavy, the lower-priority queues risk starvation.
• To avoid starvation, mechanisms such as WFQ (Weighted Fair Queuing) assign a bandwidth weight to each queue, giving high priority preferential treatment while still guaranteeing lower-priority queues some processing opportunity. The choice depends on traffic characteristics and tolerance: strict priority suits latency-critical needs like VoIP, while WFQ suits environments where multiple levels of importance coexist among business systems.

5.3.3Bandwidth control (shaping and policing)

• Shaping is a control mechanism that temporarily buffers traffic exceeding the specified rate and smooths out the send timing. Because it does not discard packets, it preserves the source traffic, but it trades off increased latency due to buffering. It is used to comply with a contracted bandwidth or to smooth out bursty transmissions.
• Policing is a control mechanism that immediately discards (or re-marks) traffic exceeding the specified rate. Because it does not buffer, latency does not increase, but excess packets are lost. This suits scenarios like "unconditionally drop anything over the contracted rate" at the ingress of latency-sensitive real-time traffic, though for TCP traffic, discarding can trigger retransmissions and induce further congestion.

Suppose an office's inter-site VPN link (contracted at 10 Mbps) needs to newly run a video conference continuously (requiring 2 Mbps, latency under 200 ms, packet loss under 1%) alongside regular business traffic. First, use DiffServ (DSCP) to mark the video-conference packets with a high-priority marking (e.g., EF = Expedited Forwarding), so the router can identify them as subject to priority queueing. Next, when choosing a queueing scheme, because the video conference has a strict latency requirement (under 200 ms) and must avoid being delayed behind other lower-priority traffic, adopt strict priority. However, strict priority alone risks starving the lower-priority business traffic, so design it so the strict-priority queue is capped at a portion of the contracted bandwidth (e.g., the equivalent of 3 Mbps), reserving the rest for lower-priority traffic. To enforce this cap, apply policing at the ingress of the video-conference traffic, immediately discarding video traffic exceeding 3 Mbps (e.g., from an unexpected number of simultaneous conferences) to prevent it from affecting other traffic. Meanwhile, for latency-tolerant bulk transfers such as file backups, apply shaping to smooth the send timing via buffering so it stays within the remaining contracted-bandwidth allotment, avoiding a cascade of retransmissions caused by packet loss. In this way, the practical design decision combines multiple QoS technologies by traffic character: "priority control plus policing to immediately cut off excess" for traffic requiring real-time behavior, and "shaping to smooth out" for latency-tolerant bulk transfers.

5.3.4Section summary

• QoS is preferential processing to protect real-time traffic such as voice/video from latency, jitter, and loss. DiffServ (DSCP) marks packets for identification
• Strict priority maximally protects high priority but can starve lower priority. WFQ guarantees lower priority some opportunity via weighting
• Shaping = smooths via buffering (more latency, no loss); policing = immediately discards excess (no added latency, but loss occurs)