3: Transport Layer

Question 1

What is the transport layer?

Answer 1

The layer in the layered model responsible for providing an abstraction over end-end communications between applications running on different hosts.

Question 2

What is connection management?

Answer 2

The methods by which network connections are opened and maintained in such a manner to ensure they are suitable for data transmission.

Question 3

What are message segments?

Answer 3

The discrete units that messages are split into before they are passed to the network layer, which provides logical communication between hosts (but not applications on hosts - that is the job of the transport layer).

Question 4

What is the difference between the network and transport layers?

Answer 4

The network layer provides logical communication between hosts, whereas the transport layer provides logical communication between applications on hosts.

Question 5

What is reliable data transfer?

Answer 5

The concept of providing integrity and certain delivery to data transmission.

Question 6

What is the concept of fairness?

Answer 6

Every TCP connection going through a bottleneck link should have equal shares of its rate.

Question 7

When may a 2-way handshake fail?

Answer 7

If client times out and retransmits but the server didn’t timeout and the connection is established, then the retransmitted connection reaches the server after the true connection closes, and there could also be data lagging as well that could interfere with state too.

Question 8

What is pipelining?

Answer 8

When you send multiple packets between acknowledgements, rather than 1 at a time in stop-and-wait.

Question 9

What are acknowledgements?

Answer 9

Specialised messages specifying that certain other messages and packets were received successfully.

Question 10

What is connectionless demultiplexing?

Answer 10

With UDP, no connection in sockets so you may have multiple sockets at the same IP and port number when you demultiplex an Internet signal at a host.

Question 11

What is connection-oriented demultiplexing?

Answer 11

With TCP, there are connections in sockets so you have a dedicated pipe between two IP addresses and port numbers. You therefore only get datagrams from one other socket to a socket on a host when demultiplexing TCP signals.

Question 12

How are TCP sockets uniquely identified?

Answer 12

IP address and port.

Question 13

What is the structure of a TCP segment?

Answer 13

Header of source and dest port and IP address, seq no, checksum, etc. Then payload.

Question 14

Why is UDP used?

Answer 14

It removes the overhead from TCP that provides reliability so can be faster for some uses.

Question 15

What are sender events?

Answer 15

A point in the control flow of a networking app in a sender which necessitates action being taken; they include receiving data from a network app to send, timeouts for acks, and receiving acks.

Question 16

How can you implement reliable data transfer over an unreliable channel?

Answer 16

Using handshakes, acknowledgements, sequence numbers, handshakes and ECCs.

Question 17

How can you recover from channels with bit errors?

Answer 17

Use ECCs to correct them or detect with parity bets and checksums, and request retransmissions with negative acknowledgements.

Question 18

What are negative acknowledgements?

Answer 18

Specialised messages specifying that certain other messages and packets were received unsuccessfully, with errors.

Question 19

When does TCP retransmit packets?

Answer 19

If either the packets aren’t received successfully by the receiver, or if they are but the acknowledgement of such from the receiver back to the sender isn’t received successfully, due to either data loss/corruption a premature timeout waiting for an ack in the sender.

Question 20

What is rdt?

Answer 20

A specification of how communications protocols can facilitate reliable data control over unreliable links and channels.

Question 21

How does TCP implement congestion control?

Answer 21

Increase tansmisson rate (window size) additevly (1 MSS) until loss occurs, then deceaseing it multiplicatevely: halving.

Question 22

What is congestion?

Answer 22

The reduced performance of network infrastructure when they attempt to handle more data than they can.

Question 23

How can you handle duplicate transmissions?

Answer 23

Use sequence numbers to ensure you don’t mistakenly process/use the same data twice.

Question 24

What are the differences between rdt versions 1.0, 2.0, 2.1, 2.2, and 3.0?

Answer 24

rdt1.0 is the most simple, just sending and receiving packets. rdt2.0 adds ACKS to handle bit errors. rdt2.1 handles the possibility of corrupted ACKs with sequence numbers to maintain synchronisation. rdt2.2 removes negative ACKs, only using positive ones, but implements the same functionality as rdt2.1. rdt3.0 handles bit errors and packet loss by adding time limits to waiting to receive packets.

Question 25

How does Go-back-N pipelining work?

Answer 25

Sequence nos in packets, receiver sends cumulative acks for all packets successfully received up to and including a certain sequence no, sender retransmits a sequence after a certain timeout without an ack received.

Question 26

What are the 2 main forms of pipelining?

Answer 26

Go-Back-N and selective repeat

Question 27

What is Go-back-N pipelining?

Answer 27

N packets can be in a pipe at once: N more between each acknowledgement, and the receiver sends cumulative acks. The sender has a timer to resend all unacked packets (up to N more) if no ack received.

Question 28

What is Selective Repeat pipelining?

Answer 28

Limited number of packets in a pipe at once, with receiver sending individual acks for each received packet. The sender has time to retransmit each unacked packet.

Question 29

What is a cumulative acknowledgement?

Answer 29

When you ack all packets up to and including a certain sequence number rather than all individually, reducing the number of acks needing to be sent

Question 30

How does Selective Repeat pipelining work?

Answer 30

Sequence numbers in packets; receiver acknowledges all packets received successfully; resend any not received successfully.

Question 31

What is MSS?

Answer 31

The maximum segment size (MSS) is a parameter of the options field of the TCP header that specifies the largest amount of data, specified in bytes, that a computer or communications device can receive in a single TCP segment.

Question 32

How do you set the timeout value for a TCP segment?

Answer 32

Timeout interval = EstimatedRTT 4DevRTT
DevRTT is the safety margin
DevRTT = (1-β)DevRTT β|SampleRTT-EstimatedRTT|
β = weight of new sampled RTT over past safety margin
EstimatedRTT = (1- α)EstimatedRTT α*SampleRTT
α = weight of sample RTT vs past ones ~ 0.125

Question 33

How do you find the estimated RTT of a TCP segment?

Answer 33

EstimatedRTT = (1- α)*EstimatedRTT   α*SampleRTT
α = weight of sample RTT vs past ones ~ 0.125

Question 34

How does TCP acknowledgement generation vary depending upon different events at the receiever?

Answer 34

If data as normal wait for next packet, if none then ack highest currently received seq no. Receiving as normal with another ack pending, then send cumulative ack for both. If seq no received higher than expected, so out of order, send duplicate ack to ask for specific next packet. If a gap filled by a packet, send an ack that covers segment to resync, as long as packet means now all seq nos up to and including it have been received successfully.

Question 35

How does fairness apply to UDP and multimedia streaming?

Answer 35

Often not - they will seek as much throughput as possible even if this means degrading performance for other hosts and even other applications on the same host.

Question 36

How is flow control implemented in TCP?

Answer 36

rwnd value in header in packets from receiver define space remaining in buffer; packets are sent such that their data don’t overflow this limit, so buffer in receiver guaranteed to not overflow.

Question 37

What is a handshake?

Answer 37

An agreement between hosts to establish a connection that seeks to ensure both are aware it is established.

Question 38

What are some causes of congestion?

Answer 38

Many users using high-bandwidth services on a network. Sending when buffers nearly full at bottleneck routers. Some loss leading to retransmissions in a vicious circle, and similarly duplicates due to timeouts.

Question 39

What is congestion control?

Answer 39

Methods seeking to mitigate congestion on computer networks.

Question 40

What are some costs of congestion?

Answer 40

Unnecessary retransmissions, increased latency, increased network strain, more work for given goodput. Waste of capacity on retransmissions.

Question 41

How does TCP detect and react to loss?

Answer 41

Ack isn’t received for packet sent (timeout) or 3x duplicate acks

Question 42

What is the idea of slow start in TCP?

Answer 42

Initially set window to 1 MSS then double it every RTT until loss: exponential increase from initial slow rate on start.

Question 43

How can you find the throughput of a TCP connection?

Answer 43

TCP throughput = [1.22 . MSS] / [RTT . √L]
MSS = maximum segment size
RTT = round trip time
L = oribability a packet will be lost

Question 44

Why is TCP fair?

Answer 44

All connections will use additive-multiplicative adjustments to tend towards balanced and stable shares of the total rate.

Question 45

What is Explicit Congestion Notification?

Answer 45

Overt measures in Internet Protocol that are taken specifically to detect, indicate, and deal with network congestion when it occurs.

Question 46

What is network-assisted congestion control?

Answer 46

A part of Explicit Congestion Notification, in which a specialised header flag indicates if a network is congested to trigger mitigating action.