Cisco QoS FAQ: Quality of Service Overview
Q1. List the four traffic characteristics that QoS tools can affect.
Answer: Bandwidth, delay, jitter, and loss.
Q2. Describe some of the characteristics of voice traffic when no QoS is applied in a network.
Answer: Voice is hard to understand; voice breaks up, sounds choppy; calls are disconnected; large delays make it difficult to know when the other caller has finished talking.
Q3. Describe some of the characteristics of video traffic when no QoS is applied in a network.
Answer: Picture displays erratically; picture shows jerky movements; audio not in sync with video; movements slow down; video stream stops.
Q4. Describe some of the characteristics of data traffic when no QoS is applied in a network.
Answer: Data arrives too late to be useful; erratic response times cause users to stop using application; customer care agents waiting on screen refresh, so customer waits.
Q5. Interpret the meaning of the phrase, “QoS is both ‘managed fairness,’ and at the same time ‘managed unfairness’.”
Answer: QoS tools improve QoS characteristics for particular packets. However, improving one packet’s behavior typically comes at the expense of another packet. The terms “managed fairness” and “managed unfairness” just refer to the fact that QoS policies may be fair to one packet but unfair to another.
Q6. Define bandwidth. Compare and contrast bandwidth concepts over point-to-point links versus Frame Relay.
Answer: Bandwidth refers to the number of bits per second that can reasonably be expected to be successfully delivered across a network. With point-to-point networks, bandwidth is equal to the speed of the link—the clock rate. With Frame Relay, the actual bandwidth is difficult to define. Typically, the minimum bandwidth equals the CIR of a VC. However, engineers at the provider and at the customer typically expect more than CIR to get through the network. The maximum bandwidth would be bounded by the slower of the two access rates on the access links.
Q7. Compare and contrast bandwidth and clock rate in relation to usage for QoS.
Answer: Bandwidth refers to the router’s perceived bandwidth on the interface/subinterface, and is referenced by QoS tools. Clock rate defines the physical encoding rate on router interfaces that provide clocking; QoS tools ignore the clock rate setting.
Q8. List the QoS tool types that affect bandwidth, and give a brief explanation of why each tool can affect bandwidth.
Answer: Compression, CAC, and queuing affect bandwidth. Compression reduces the number of bits needed to transmit a frame, allowing more frames to be sent over the same amount of bandwidth. CAC reduces the overall load of voice and video traffic in the network by disallowing new calls. Queuing can reserve subsets of the bandwidth on a link for a particular queue, guaranteeing a minimum amount of bandwidth for that queue.
Q9. Define delay, compare/contrast one-way and round-trip delay, and characterize the types of packets for which one-way delay is important.
Answer: Delay is the time taken from when a frame/packet is sent until it is received on the other side of the network. One-way delay just measures the delay for a packet from one endpoint in the network to its destination. Round-trip delay measures the time it takes to send a packet to one destination and for a response packet to be received. Voice and video are concerned with one-way delay.
Q10. List the categories of delay that could be experienced by all three types of traffic: data, voice, and video.
Answer: Serialization, propagation, queuing, forwarding/processing, shaping, network. Note that codec, packetization, and de-jitter delays are unique to voice and video, so technically these delays should not have been part of your answer for this question.
Q11. Define, compare, and contrast serialization and propagation delay.
Figure: Serialization and Propagation Delay for Selected Packet and Link Lengths
Answer: Serialization delay defines the time it takes to encode a frame onto the physical link. For instance, on a point-to-point link of 56 kbps, a bit is encoded every 1/56,000 seconds; therefore, a frame that is 1000 bits long takes 1000/56000 seconds to encode on the link. So, serialization delay is a function of link speed and length of the frame. Propagation delay defines the time taken for a single bit to be delivered across some physical medium, and is based solely on the length of the physical link, and the speed of energy across that medium. If that same point-to-point link were 1000 km (approximately 620 miles) in length, the propagation delay would be 1,000,000m/2.1 * 10^8 ms, or 4.8 milliseconds.
Q12. Define network delay.
Answer: Network delay refers to the delay incurred by a packet inside a packet network, like ATM, Frame Relay, or MPLS networks. Because the customer does not know the details of these networks, and because many customers’ packets share the carrier network, variable delays occur.
Q13. List the QoS tool types that affect delay and give a brief explanation of why each tool can affect delay.
Queuing, link fragmentation and interleaving, compression, and traffic shaping. Queuing methods use an algorithm to choose from which queue to take the next packet for transmission, which can decrease delay for some packets and increase delay for others. LFI tools break large frames into smaller frames, so that smaller delay-sensitive frames can be sent after the first short fragment, instead of having to wait for the entire large original frame to be sent.
Compression helps delay because it reduces the overall load in the network, reducing congestion, reducing queue lengths, and reducing serialization delays. Finally, traffic shaping actually increases delay, but it can be applied for one type of traffic, allowing other traffic to be sent with less delay.
Q14. Define jitter. Give an example that shows a packet without jitter, followed by a packet with jitter.
Answer: Jitter measures the change in delay experienced by consecutive packets. If a PC sends four packets one after the other, practically at the same time, say 1 ms apart, so the departure times are T=0, T=1, T=2, and T=3, for instance, packets arrive at T=70, T=71, T=80, T=81, respectively. The second packet was sent 1 ms after the first, and arrived 1 ms after the first packet—so no jitter was experienced. However, the third packet arrived 9 ms after the second packet, after being sent 1 ms after the second packet—so 8 ms of jitter was experienced.
Q15. List the QoS tool types that affect jitter and give a brief explanation of why each tool can affect jitter.
Answer: Queuing, link fragmentation and interleaving, compression, and traffic shaping. These same QoS tools can be used for addressing delay issues. Queuing can always be used to service a jitter-sensitive queue first if packets are waiting, which greatly reduces delay and jitter. LFI decreases jitter by removing the chance that a jittersensitive packet will be waiting behind a very large packet. Compression helps by reducing overall delay, which has a net effect of reducing jitter. Traffic shaping may actually increase jitter, so it should be used with care—but if shaping is applied to jitter-insensitive traffic only, jitter-sensitive traffic will actually have lower delays and jitter.
Q16. Define packet loss and describe the primary reason for loss for which QoS tools can help.
Answer: Packet loss means that a packet, which has entered the network, does not get delivered to the endpoint—it is lost in transit. Routers and switches drop packets for many reasons. However, QoS tools can affect the behavior of loss when packets will be lost due to queues being too full. When a queue is full, and another packet needs to be added to the queue, tail drop occurs.
Q17. List the QoS tool types that affect loss and give a brief explanation of why each tool can affect loss.
Answer: Queuing and RED. Queuing tools allow definition of a longer or shorter maximum queue length; the longer the queue, the less likely that drops will occur. Also by placing traffic into different queues, more variable traffic may experience more loss, because those queues will be more likely to fill. RED tools preemptively discard packets before queues fill, hoping to get some TCP connections to slow down, which reduces the overall load in the network—which shortens queues, reducing the likelihood of packet loss.
Q18. Describe the contents of an IP packet carrying the payload for a G.729 VoIP call.
Answer: The IP packet contains an IP header, a UDP header, an RTP header, and the voice payload. With G.729, the payload uses 20 bytes, with an 8-byte UDP header, and a 12-byte RTP header. The IP header is 20 bytes long.
Q19. Describe the amount of bandwidth required for G.711 and G.729 VoIP calls, ignoring data-link header/trailer overhead.
Answer: G.711 consumes 64 kbps for payload, for an 80-kbps stream with IP, UDP, and RTP headers. G.729 consumes 8 kbps for payload, plus another 16 kbps for IP, UDP, and RTP headers, for a total of 24 kbps.
Q20. List the delay components that voice calls experience, but which data-only flows do not experience.
Answer: Codec delay, packetization delay, and de-jitter initial playout delay.
Q21. Define the meaning of the term “packetization delay” in relation to a voice call.
Answer: Voice must be converted from sound waves to analog electrical signals, and finally to digital signals, and then placed into a packet. Before 20 ms of voice digital payload can be placed into a packet, the speaker must speak for 20 ms. Packetization
delay refers to the (default) 20 ms of delay, waiting for the speaker to speak long enough to fill the packet with the correctly sized payload.
Q22. List the different one-way delay budgets as suggested by Cisco and the ITU.
Answer: The ITU in document G.114 suggests a budget of up to 150 ms for quality voice calls; Cisco suggests a delay budget of up to 200 ms one-way if you cannot meet the 150-ms goal.
Q23. Define the term “codec delay” and discuss the two components when using a G.729 codec.
Answer: Voice calls incur codec delay when the codec converts the analog signal into digital voice payload. Every codec requires some time to process the incoming signal, which adds delay. With G.729, because it is predictive, it must also wait for some additional incoming voice to arrive, because it is algorithm-processing the voice sample to be encoded, plus a part of the next sample that will be encoded. The delay waiting for the additional voice is called “look-ahead” delay.
Q24. Describe the affects of a single lost packet versus two consecutive lost packets, for a G.729 voice call.
Answer: Lost voice packets result in the receiver having a period of silence corresponding the length of voice payload inside the lost packet(s). With two consecutive G.729 packets lost, 40 ms of voice is lost; the G.729 codec cannot predict and generate replacement signals when more than 30 ms of consecutive voice is lost. A single lost G.729 packet would only cause a 20-ms break in the voice, which could be regenerated. So, a single lost packet is not perceived as loss in a G.729 call.
Q25. Describe a typical video payload flow in terms of packet sizes and packet rates.
Answer: Video payloads use variable-length packets. The packet rates are also typically variable.
Q26. Discuss the delay requirements of video traffic.
Answer: Interactive video (video conferencing, for instance) requires low delay because it is interactive. Delay budgets up to 200 ms are the norm. However, streaming video— one-way video—can tolerate long delays. When playing an e-learning video, for instance, the playout may start after 30 seconds of video has been received into a de-jitter buffer—but each packet may have experienced several seconds of delay.
Q27. List the basic differences between TCP and UDP traffic.
Answer: TCP performs error recovery, whereas UDP does not. TCP also uses dynamic windowing to perform flow control, whereas UDP does not. Both use port numbers to multiplex among various applications running on a single computer.
28 Contrast the QoS characteristics needed by interactive data applications, as compared to the QoS needs of voice payload flows.
Answer: Answering such a question requires one to understand that QoS requirements for data applications are more subjective than those for voice. Generally, interactive data wants consistent delay (low jitter), but relative to voice, more jitter is tolerable. Bandwidth demands vary greatly for data applications, whereas a single voice call uses a static amount of bandwidth.
Delay for interactive data can be relatively longer than for voice, but the key measurement for data is application response time, which includes round-trip packet delays. Finally, data applications are much more tolerant of packet loss, because either the application will resend the data, or rely on TCP to resend the data, or just not care whether some data is lost.