Computer Networks 0605933 Revision Lecture

1
Computer Networks(06-05933)Revision Lecture

• Rob Minson, rm. 134
• R.Minson_at_cs.bham.ac.uk

2
The Exam

• Length 1 ½ hours
• Structure
• 4 Questions on the paper
• You must answer 3 of them
• All of the questions contain problems from a
  range of the topics and layers we covered. DO NOT
  just revise 3 of the 4 layers.
• Pass Marks
• Standard Module 40
• Extended Module 50 (including your essay)

3
Revision Lectures

• Lecture 1 (Monday 27th)
• Physical Layer
• Data-Link Layer
• Lecture 2 (Tuesday 28th)
• Network Layer
• Transport Layer
• Lecture 3 (Thursday 30th)
• QA and exercise recap

4
The Physical Layer
5
Layering

• Layers allow abstraction and an open model
  crucial for Inter-Networking
• Layers communicate across hosts via protocols
• Layers communicate with each other via services
• Passing a message through layers involves
  encapsulation
• OSI and TCP/IP are both models
• TCP/IP is the accepted global standard

6
Digital Communications

• Physical properties of media can be used to
  represent 1s and 0s
• V/-V voltage constants in electrical wire
• Synchronous encoding is needed to ensure sender
  knows boundary between bit periods
• Each encoding scheme has pros and cons
• Manchester
• Originally used in Ethernet but now considered
  wasteful
• Non-Return to Zero with RLL coding
• Widely used in Ethernet and others (PCI-e, SATA,
  etc)

7
Analog Signals and Modulation

• Analog signals carry data by modulating a carrier
  wave (frequency, amplitude, phase)
• Digital data is encoded by a particular
  modulation signature (can be several bits per
  period)
• Most modern schemes modulate with Amplitude and
  Phase and multiplex with Frequency
• Maximum frequency (range) limits rate of
  communication (same is true in digital comms)

8
Theoretical Limits

• Information theorists place absolute limits on
  communication rate in any medium
• Data Rate is fundamental form of equation
• max bps sample rate bits per sample
• Nyquists data rate shows us the maximum based on
  frequency bandwidth
• max bps 2f log2V
• Shannons data rate adds the effect of noisy
  media on the achievable symbol rate
• max bps bandwidth log2(S/N 1)

9
The Data-Link Layer(The Logical Link Control
Sub-Layer)
10
Error Control

• The physical layer is error-prone
• Noise, attenuation, distortion
• Data Link Layer must deal with this
• Split outgoing binary stream in to frames
• Each frame contains m data bits and r check bits
• Error Detection can detect a corrupted frame
• Parity bit, CRC
• Error Correction can correct the flipped bits
• Hamming codes

11
Error Detection with Parity Bits

• A Parity Bit makes the number of 1s in a word
  even, e.g. if m 3 and r 1

12
Error Correction with Hamming Codes

• Hamming codes correct single bit errors in a
  given data word
• Embed r parity bits at positions 20, 21, 22,
• Each parity bit checks a unique subset of the
  other bit positions (and itself)
• If a single bit error occurs a unique combination
  of the parity bits will be incorrect
• This unique combination is used to locate and
  correct the flipped bit

13
Error Detection with CRC

• CRC Algorithm the sending end
• Agree a generator divisor G of length r
• Append r zeroes to the M data bits producing xrM
• Calculate xrM G R
• Calculate xrM R T
• Equivalent to just appending R to M

1 0 1
1 1 1 0 0 0
1 1 1 0 0 0 1 0 1 0 1 1
1 1 1 0 0 0 0 1 1 1 1 1 0 1 1
14
Error Detection with CRC

• CRC Algorithm the receiving end
• Divide the received bits by G
• If the remainder 0, remove first r bits and
  continue
• If the remainder is ? 0 an error has occurred

1 1 1 0 1 1 / 0 1 1 0 0 0 0
M 1 1 1
15
Bursty Error Control

• Block Sending
• If a burst error of length n wipes out a frame
  this will be felt as a single bit error in n
  consecutive frames (each easy to detect/correct)

16
Flow Control

• Serves two functional purposes
• Prevent buffer overruns
• Recover from lost frames
• Stop and Wait
• prone to deadlock
• Automatic Repeat Request
• uses timeouts, no deadlock but inefficient
• Sliding Window
• pipelines multiple frames
• two flavours Go-Back-N and Selective-Repeat

17
The Data-Link Layer(The Medium-Access Control
Sub-Layer)
18
MAC Layer Summary

• ALOHA protocols
• Randomisation of re-send time (slotted/pure)
• CSMA protocols (with persistence level)
• Collision Detection (Ethernet)
• Collision Avoidance (WLAN)
• Both use Exponential Back-off
• MAC collision domains
• Can be separated by Bridges
• Can be eradicated by Switches

19
ALOHA Collision Detection

• Central station bounces ACK back to sender
• ACKs sent on separate frequency so never collide
• If ACK not received then sender assumes collision

20
Slotted vs. Pure ALOHA

• Both use randomisation after collision
• Slotted reduces probability of collision
• If frame is ready in middle of period
• In Pure ALOHA I will begin sending and destroy
  any currently sending frames
• In Slotted ALOHA I will wait and only destroy
  other waiting frames

21
Carrier Sense Protocols (CSMA)

• Medium can be sensed for traffic
• 1-persistent
• channel free send immediately
• channel busy send once it is free
• non-persistent
• channel free send immediately
• channel busy send after random amount of time
• (slotted) p-persistent CSMA
• channel free transmit with prob p
• channel busy repeat algorithm next slot

22
CSMA/CD in Ethernet

• Propagation delay means collision are not
  detected immediately.
• Can be thought of as a frame bouncing off
  another frame and back to the sender

23
CSMA Adaptive Backoff

• CSMA with Exponential Backoff
• Random range 0,2c-1 (where c collisions)
• Slotted time (in Ethernet 1 slot length RTT)
• t0s A and B collide
• A and B choose randomly from 21 slots 0,1
• A chooses 1, B chooses 1
• t2s A and B collide
• A and B choose randomly from 22 slots 0,3
• A chooses 2, B chooses 0
• t3s B transmits successfully
• t5s A transmits successfully

24
CSMA/CA in WiFi
25
The Network Layer
26
Link-State vs. Distance-Vector

• Link-State
• Global Coordination
• Fast Convergence
• Guaranteed Cycle Free
• High Message overhead
• Distance-Vector
• Localised Gossiping
• Slow Convergence
• Cycles (Count-to-Infinity)
• Can be solved with split-horizon/hold-down but at
  a cost
• Low overhead

27
Distance-Vector Routing
28
Distance-Vector Routing

• Count-to-Infinity Problem
• Nodes advertise others routes back to them
• When a node/link fails, nodes incorrectly
  advertise routes back to the failed node/link

29
Link-State Routing
30
Link-State Routing

• Link-State Advertisements contain local info
• combined they construct the entire graph
• LSAs are buffered and periodically forwarded
• Seq and Age numbers allow flooding to be
  constrained and for old data to be removed from
  the network

31
Interior/Exterior Gateway Protocols

• Internet organised in to Autonomous Systems
  (ASes)
• The routing graph of an AS is maintained by its
  routers with an IGP
• Distance Vector RIP and EIGRP
• Link-State IS-IS and OSPF
• The graph between ASes is maintained by gateway
  routers with an EGP
• Distance Vector BGP

32
Routing via IP

• IP Class System (Classful Addressing)
• Originally AS addresses were from 3 classes
• Classes have different number of bits for network
  portion
• Class indicated by most-significant bits

33
Routing via IP

• Classless-Inter-Domain Routing (CIDR)
• By choosing a mask length explicitly, AS size can
  be matched precisely to projected need
• Network Address includes the net mask
• IPAddress/mask
• (e.g. 152.64.0.0/14 first 14 bits are network)
• Before CIDR
• 23.0.0.0 unambiguously indicated a class A
  network
• After CIDR
• Same network must advertise itself as 23.0.0.0/8

34
NAT (Network Address Translation)

• NAT using packet manipulation

35
IP Acquisition and Resolution

• Translating between MAC and IP addresses
• ARP what is the MAC for this IP?
• RARP, BOOTP what is the IP for this MAC?
• Acquiring IP addresses
• Hierarchy of organisations manage the whole
  address space
• At single-host level, lease a single address
  via a DHCP server somewhere on your network

36
The Transport Layer
37
TCP vs. UDP

• What is UDP useful for?
• Fire-and-Forget data
• informational (time server updates, keep alives)
• wide-area search (service discovery, P2P search)
• Realtime data
• Any application where we dont care about data
  from the past, only current data
• voice over IP
• streaming video/audio
• multiplayer games

38
Fundamentals of TCP

• Basically a Sliding Window protocol
• Data sent in segments in IP packets
• Very large (32-bit) sequence number
• represents byte offset of segment

39
TCP Error/Flow Control

• Sliding window (byte-oriented)
• ACK for byte n means I have received every byte
  from 0 to n-1 correctly
• Selective Repeat
• Means sender receiver have shared window
• Receiver buffers out-of-order byte segments
• Requests retransmission via NACK
• (NACK ACK for any segment other than latest
  received)
• 3 x NACK prompts a resend
• RTO timeout value used as a failsafe

40
TCP and RTO-Estimation

• RTT segment ? receiver ACK ? sender
• What happens if
• RTO lt RTT?
• RTO gtgt RTT?
• In TCP/IP the RTT is very hard to predict
• Over time (traffic patterns, routing element
  problems)
• Between networks (bandwidth, diameter,
  population)
• TCPs various strategies for piggybacking and
  delaying ACKs dont help either
• TCP must therefore have a runtime-estimated RTO
• Sampling (RTT SRTT)
• Smoothing (aSRTT (1-a)RTT a smoothing level)

41
TCP and Dynamic Flow Control

• Application Layer drives window changes
• TCP Buffer
• used to store and re-order segments
• usually fairly small
• determines max window size
• Application Buffer
• used to store fully-ordered segments
• usually fairly large
• can be sampled by TCP process to determine best
  current window size

42
TCP and Congestion Control

• Multiple TCP flows share same routing elements
  (cant predict flows)

• Congestion Control
• Add a second window (congestion window)
• Permitted segments must lie in both windows

43
TCP and Congestion Control

• Reacting to congestion
• Determine congestion state based on ACKs
• ACKs are being received before RTO is reached
• there is low/no congestion
• ACKs are not being received or are arriving late
• router overload or router buffer overflow is
  ocurring
• Slow Start Phase
• Send 1 segment, wait...
• receive 1 ACK, increase cwnd to 2
• Congestion Avoidance Phase
• Send 8 segments, wait...
• receive 8 ACKs, increase cwnd by 8/8 MSS

44
TCP and Congestion Control

• Additive Increase/Multiplicative Decrease (AIMD)