When you provide QoS in a network, one of the major elements you must consider is congestion, and the traffic management behavior during congestion. Congestion in a network is caused by many different conditions and events, including node failures, link outages, broadcast storms, and user traffic bursts.
At a high level, three main types or stages of congestion exist:
No congestion
Bursty congestion
Severe congestion
In a noncongested network, QoS actions ensure that delay-sensitive applications, such as real-time voice and video traffic, are sent before lower-priority traffic. The prioritization of delay-sensitive traffic is essential to minimize delay and reduce or eliminate jitter, which has a detrimental impact on these applications.
A network can experience momentary bursts of congestion for various reasons, such as network failures, rerouting, and broadcast storms. The switch has sufficient capacity to handle bursts of congestion in a seamless and transparent manner. If the burst is not sustained, the traffic management and buffering process on the switch allows all the traffic to pass without loss.
Severe congestion is defined as a condition where the network or certain elements of the network experience a prolonged period of sustained congestion. Under such congestion conditions, congestion thresholds are reached, buffers overflow, and a substantial amount of traffic is lost.
When you perform traffic engineering and link capacity analysis for a network, the standard design rule is to design the network links and trunks for a maximum average-peak utilization of no more than 80%. This value means that the network peaks to up to 100% capacity, but the average-peak utilization does not exceed 80%. The network is expected to handle momentary peaks above 100% capacity.