Rule no 1. QoS does not help in situations where there is no enough bandwidth but helps optimize performance by prioritization of the traffic which helps to mitigate issues that occur periodically on the network.
Most common problems to solve:
1. Jitter: control the difference in delay between packets
2. Packet loss: Ensure that traffic capable of handling packet loss is dropped before traffic that cannot
3. TCP global synchronisation - refers to a network performance issue that can occur in congested or heavily loaded networks. It occurs when multiple TCP connections simultaneously reduce their sending rates due to congestion, resulting in periods of low network utilisation followed by sudden bursts of increased traffic
1. Packet classification. Process of identifying different types of the network traffic based on its characteristic/flow (IP information, port, protocol, markings).
2. Traffic marking involves setting specific values (bits) in packet headers to indicate its priority following mechanisms can be used:
- at layer 3 by setting DSCP or IP precedence
- at layer 2 by setting CoS bits (Ethernet 802.1q defines three bits priority)
- when using MPLS by setting EXP bits
3. Traffic classification can be also achieved with ACLs, NBAR (Network Based Application Recognition)
4. Congestion management/avoidance - RED and WRED two mechanisms to keep the traffic going by dropping randomly or per threshold traffic ahead of expected congestion. main difference between the two is that WRED provide more granular control over what is dropped.
- marking is used to determine queue for each packet
- queues are used to make sure delay sensitive flows are not dropped i.e.: VoIP traffic is transmitted and not dropped.
- non-delay sensitive packets/lower priority queues may randomly drop traffic
- avoids TCP global synchronisation streams, drops or marks packets randomly before queues are full
WRED - Weighted Random Early Detection (RED with more granular approach in terms of traffic classification, different queues and probability profiles)
LLQ - Low Latency Queueing (extension of CBWFQ) adds strict-priority capability provides both bandwidth and latency guarantee. Recommended for real time traffic.
both use to manage and schedule packets during congestion times.
6. Policing and shaping - both are traffic managment techniques used to enforce bandwidth limitation. Shaping rate/delay traffic to configured requirement. Policing monitors traffic and drops or remarks it if the configured limit is reached.
Traffic characteristics:
- Voice traffic:
- smooth
- benign
- drop sensitive - less that 1%
- delay sensitive - less 150ms one way delay / 30ms jitter
- UDP
- voice/video conference traffic:
- bursty and greedy:
- dependent on codec and video quality
- drop sensitive (less that 1%)
- delay sensitive (150ms one way delay / 30ms jitter)
- UDP
- Data traffic:
- smooth or bursty
- bening or greedy
- drop insensitive
- delay insensitive
- TCP or UDP (TCP prevelent)
Traffic is often split into different traffic classes such us:
- premium (or platinium)
- VoIP
- Video conferencing
- gold
- critical
- silver
- Transactional
- Best Effort
- Web traffic
- etc...