Tutorial 5: 7th November 2019

Question 1

What is QUIC?

Answer 1

Quick UDP Internet Connections = a new transport layer Internet protocol to improve upon and replace TCP. It’s faster and encrypted-by-default.

Question 2

What is BBR?

Answer 2

Bottleneck Bandwidth and Round-trip propagation time (BBR) is a TCP congestion control algorithm developed at Google in 2016 that uses the maximum bandwidth and round-trip time of the last window to build an explicit model of the network. Higher capacity NICs mean latency/model-based congestion control algorithms, such as BBR, provide higher throughput and lower latency as a more reliable alternative to more loss-based algorithms like CUBIC.

Question 3

What are the impacts of QUIC?

Answer 3

• Biggest change to the transport layer in >30 years
• connection id abstraction: flows by connection ID, not IP and port; if any change new UDP socket moves connection and connection still open through connection ID - don’t need to manually reconnect in app
• middleboxes not supported, inc NAT
• encrypted by default: better privacy and security
• transport moved to user-space: rapid prototyping & experiments for new transport protocols and customisation to better suit needs
• separates protocol from congestion control
• smaller rtt but same throughput with same congestion control
• enables multistreaming which TCP blocks: get data ASAP without waiting for correct ordering

Question 4

Are any properties of QUIC new or unique?

Answer 4

No. It is just the first technology that brought them together.

Question 5

How does the performance of QUIC compare to TCP?

Answer 5

QUIC has a reduced RTT (time to handshake) but not throughput: same with same congestion control mechanisms.

Question 6

What his HOL blocking?

Answer 6

Head-of-Line blocking = a performance-limiting phenomenon that occurs when a line of packets is held up by the first packet. If a packet must be sent before any others (e.g. first in FCFS/FIFO queue), then delays in processing it will hold all others up. These may include having congestion in output destination, or even from input to output via the switching fabric of a router when output buffers are full. Head of line = front of queue. Blocking as head is stopping all others.

Question 7

What was the main change of HTTP/2?

Answer 7

To solve HOL blocking with request multiplexing, which eliminates HOL blocking at the application layer (i.e. in HTTP) but HOL not at the transport layer (i.e. in TCP).

Question 8

What are the consequences of QUIC being at the transport layer?

Answer 8

Eliminates HOL blocking from TCP and HTTP and allows multistreaming. This means an app gets data bound for it as soon as it can be sent rather than TCP’s approach of blocking out-of-order data. This will be better for real-time applications like livestreams or Skype calls.

Question 9

Can you use middleboxes with QUIC? Why?

Answer 9

No, because it encrypts packets by default so middleboxes can’t inspect data in packets to perform their additional functionality on them.

Question 10

Do ISPs like QUIC? Why?

Answer 10

No. It makes it harder for them to control and throttle traffic. All packets are encrypted by default, including port numbers. They also can’t make performance measurements to check the network is working correctly, such as looking at sequence and ack numbers to find RTTs.

Question 11

What is DPI?

Answer 11

DPI = Deep Packet Inspection = a type of data processing that inspects packets being sent over a computer network, and takes according action. This may include blocking, re-routing, or logging it. DPI is used to ensure correct data formatting, security filtering, eavesdropping, censorship, performance measurement, etc.

Question 12

What is a SPIN bit?

Answer 12

A QUIC header that can be used to measure RTT.

Question 13

How does the SPIN bit work?

Answer 13

Send packets with it set to 1 until an ack with it set to 1 is received. Then send with 0 until ack with 0, then back to 1. Measure time between first outbound packet with one value to first inbound ack with it to get the RTT.

Question 14

Does QUIC work with NAT? Why?

Answer 14

No. Port numbers are encrypted. NAT sees it as invalid UDP packets. It also presents a problem using both: the destination of a QUIC connection may be remapped on NAT which makes it unreachable.

Question 15

What is BGP?

Answer 15

BGP = Border Gateway Protocol = a standardised exterior gateway protocol designed to exchange routing and reachability information among autonomous systems (AS) on the Internet. It makes routing decisions based on paths, network policies, or rule-sets configured by a network administrator and is involved in making core routing decisions. BGP is responsible for inter-AS routing.

Question 16

What is DNS over HTTPS?

Answer 16

A protocol for performing remote Domain Name System resolution via the HTTPS protocol.

Question 17

What is the purpose of DNS over HTTPS?

Answer 17

To not get blocked by firewalls and to increase user privacy and security by preventing eavesdropping and manipulation of DNS data by man-in-the-middle attacks with HTTPS encryption as well as improving performance.

Question 18

How does DNS over HTTP work on Chrome and Firefox?

Answer 18

Chrome redirects to Google’s DNS servers, Firefox to Cloudflare’s (chosen for privacy).

Question 19

What is the purpose of BBR?

Answer 19

To not fill sending queues and move away from loss-based congestion control.

Question 20

How does BBR interact with Cubic TCP?

Answer 20

It massively outcompetes Cubic flows and prevents them from getting almost any throughput at all. It’s very unfair and too aggressive. Cubic flows never have a chance to fill their queues and get any throughput.

Question 21

What does the way BBR interacts with Cubic TCP tell us about the nature of BBR?

Answer 21

Despite it specifically being designed not to do so, in some circumstances BBR must be filling its sending queue.

Question 22

How can BBR lead to loss?

Answer 22

Its aggression causes network congestion which will cause packet loss.

Question 23

When does BBR work as intended?

Answer 23

With a small number of users in a small number of scenarios. In these scenarios, it establishes a low-throughput, consistent send rate: loss becomes 0. This is it working as intended.

Question 24

Is BBR scalable if everyone were to use it? Why?

Answer 24

No as it is too aggressive. It would cause congestion, packet loss, and decreased throughput.

Question 25

What changes were made between BBR and BBR 2.0 that may help fix some of its problems?

Answer 25

Data rate of flows adjusted by +/- 5% rather than 25% to increase stability. It no longer has a persistent internal BPP to exceed the congestion BDP. (?)

Question 26

Do BBR flows use the total capacity of a link?

Answer 26

No. The sum of the mean data rates of all BBR flows on a link < its total capacity.

Question 27

Are BBR flows consistent in throughput?

Answer 27

No. They are extremely inconsistent.

Question 28

How much time elapses before BBR flows establish even (stable) flows?

Answer 28

~30 seconds.

Question 29

What happens to old BBR flows when new ones are made?

Answer 29

They are destroyed.

Question 30

What is BDP?

Answer 30

Bandwidth-Delay Product = a measurement of how many bits can fill up a network link. It gives the maximum amount of data that can be transmitted by the sender at a given time before waiting for an acknowledgment. Thus it is the maximum amount of unacknowledged data.

BDP = data rate x RTT

Question 31

What is BPP with BBR?

Answer 31

Bits Per Pixel = quality of video frames in packets (?)