Lesson 6 Congestion Control Mike

Question 1

What is the goal of Congestion Control?

Answer 1

The goal of Congestion Control is to “fill the internet’s pipes” without overflowing them

• Efficient throughput, fairness amongst resources & avoiding Congestion Collapse

Question 2

What is Congestion Collapse?

Answer 2

• It is where an “increase in load causes a decrease in useful work”
• Happens when two streams merge together inefficiently

Question 3

What causes Congestion Collapse and what are the solutions?

Answer 3

• Retransmissions of packets -> solved by better timers and to use TCP Congestion Control
• Undelivered packets -> solved by Congestion Control

Question 4

What are the two different approaches to Congestion Control?

Answer 4

• End-To-End (TCP Congestion Control)

* Network Assisted

Question 5

What is End-To-End (TCP) Congestion Control?

Answer 5

• all rates are inferred by packets loss & delays
• just keeps increasing rates until packets are lost then TCP interprets packet loss as congestion and backs off the rate

Question 6

What are the two ways to adjust sending rates in End To End Congestion Control?

Answer 6

• Window-Based

* Rate-Based

Question 7

What is Additive Increase?

Answer 7

It’s the part of the window based algorithm that increases the packet size by one every time the sender receives a successful ACK from the receiver

Question 8

What is Multiplicative Decrease?

Answer 8

It’s the part of the window based algorithm that reduces the window size by half because we didn’t receive a successful ACK

Question 9

What does AIMD stand for?

Answer 9

Additive Increase, Multiplicative Decrease

Question 10

What is the Window-Based way to adjust sending rates?

Answer 10

AIMD - only allowed a certain number of packets “in flight” until it receives an Acknowledgement from the receiver

Question 11

What is the Rate-Based way to adjust sending rates?

Answer 11

It monitors loss rates and uses timers to modulate rates

Question 12

What is Network-Assisted congestion control?

Answer 12

It’s where routers provide feedback on sending rates and update sender if there is congestion and even give specific rates that the sender should be sending at

Question 13

Given the below variables, what is the sending rate?

• RTT = 100 milliseconds
• Packet = 1 kB (1 Byte = 8 bits)
• Window Size = 10 packets

Answer 13

• 1 kB = 8 * 1000 = 8000 bits per packet
• 10 packets –> 8000 * 10 = 80,000 bits per window size
• So rate is 80,000 bits per 100 milliseconds or 80000/100 = 800 kB per second OR 800,000 bits per second

Question 14

What increases when Additive Increase is applied?

Answer 14

EFFICIENCY

Question 15

What increases when Multiplicative Decrease is applied?

Answer 15

FAIRNESS

Question 16

Servers are ____ bandwidth and ____ latency

Answer 16

high bandwidth (can send a lot at one time)

low latency (have a short RTT to the sender and back)

Question 17

What is Throughput Collapse occur?

Answer 17

It occurs because switches get overloaded due to small switch buffer lines

Question 18

What is Barrier Synchronization and what problem does it fix?

Answer 18

It fixes Throughput Collapse by adjusting the granularity of retransmission timers to microseconds and ACK’ing every other packet

Question 19

What are the challenges of multi-media streaming?

Answer 19

• Large volumes of data
• No constant rate due to variable volume of data
• low tolerance for delay variation or delay in general )but can tolerate some loss)

Question 20

In video compression what are the main 2 types of compression called?

Answer 20

• Spatial Redundancy

* Temporal Redundancy

Question 21

What is Spatial Redundancy?

Answer 21

It’s the video compression of a SINGLE image

Question 22

What is Temporal Redundancy?

Answer 22

It’s the video compression across MULTIPLE images

Question 23

What are Reference/Anchor frames?

Answer 23

They are the base frames that are used

Question 24

What are Derived Frames (aka P Frames)?

Answer 24

a small variation on the base Anchor frame plus a few motion vectors

Question 25

What can streaming video/audio tolerate?

Answer 25

• Some loss is fine
• Some delay at the beginning is fine
• VARIATION IN DELAY IS NOT TOLERATED

Question 26

Why can UDP be a better choice than TCP for Audio/Video streaming?

Answer 26

TCP has protocol overhead, is reliable and is slower

UDP doesn’t care about retransmissions or reliability an just needs to be fast

Question 27

What is the Incast problem and when does it occur?

Answer 27

• when collective communication (i.e., many-to-one or many-to-many patterns) occurs on high fan-in switches
• This results in many small packets arriving at the switch at the same time, thus causing some of the packets to be lost.
• In addition, low latency network leads to the timeout delay being more than the round-trip time of the network. This causes large delays to occur in which the system is simply waiting for timeouts

*Solutions:
Reducing the retransmission timeout for TCP improves system throughput

Client acknowledges every other packet rather than every packet to reduce network load