Quality of Service
Disadvantages of packet switched networks can be solved using QoS.
ITU recommendations (G.114 specifications):
- Delay < or = 150 ms one way
- Jitter < or = 30 ms
- Packet loss < or = 1%
Bandwidth for Voice and Video RTP is subject to codec, sampling, L2 and L3 overhead. Bandwidth for signaling depends on the protocol.
Bandwidth for Video should always have 20% additional overhead for key frames where the entire picture is sent rather than the normal pixel difference.
Types of QoS Models:
Best Effort – No QoS. Used FIFO – first in first out.
Integrated Services (IntServ) – Guaranteed bandwidth using RSVP. In order for it to work, it had to be enabled on every Layer 3 device in the path of a call.
Differentiated Services (DiffServ) – Guaranteed bandwidth for voice and signaling. Classified different traffic types. Uses policies to ensure differentiated levels of priority and service. Most commonly sued in enterprise networks because carriers usually use MPLS-TE. Does not have to be enabled on every router in a VoIP call path. First we need to classify and mark the traffic. We then deal with queuing and dropping mechanisms. Voice should be queued first. Finally have link specific mechanisms like traffic shaping, fragmentation, compression then transmission.
Where do we apply QoS:
Classification and Marking – as close to the source as possible.
Policing – this is dropping the traffic. It is always applied at the ingress port.
Scheduling (Queuing, congestion management and congestion avoidance) – applied at the egress.
Link specific mechanism – always the last to be done as the packet is leaving the device.
QoS Queuing Types:
FIFO – First In, First Out
Priority Queuing (PQ) – sends all the packets in a PQ then sends from other Q’s. We first service the priority queue then service the rest of the queues. This can lead to starvation of other queues especially when the PQ is always full.
Weighted Fair Queuing (WFQ) – Schedule traffic to be transmitted based on flow.
Class- Based Weighted Fair Queuing (CBWFQ) – Schedule traffic to be transmitted based on admin defined classes to define which traffic gets how much of the available bandwidth.
Low Latency Queuing – Basically a combination of PQ and CBWFQ.
Link Specific Mechanisms : Fragment and interleave (LFI)
Serialization delay – time required to serialize the frames and put them on the wire. For slow links, this delay is a major factor affecting latency and jitter. For slow links, data packets which are large need to be fragmented and interleaved with the smaller urgent voice packets.
As we go above a link speed of 768 kbps, LFI and compression end up causing a negative value in terms of CPU etc.
Voice packets should never be fragmented because it will introduce additional delay.
Link Specific Mechanisms: Header and Payload Compression
IP /UDP / RTP headers can be reduced from 40 bytes to 2 bytes with a standard UDP or 4 bytes if UDP Check summing is used. This is only for low speed links , or = 768 kbps.
Payload compression – compresses not only the RTP header but also the RTP payload. It is extremely CPU intensive.
Link Specific Mechanisms: Traffic Shaping
Policers drop the traffic and a given line rate. Traffic shaper delays or queues the excess traffic over the line rate and releases the traffic over a given time rate. This smooths out the traffic, prevents unnecessary drops and uses the available bandwidth more efficiently.