Final Review Fall 2017 Profs Peter Steenkiste & Justine Sherry

Transcription

1 Final Review Fall 2017 Profs Peter Steenkiste & Justine Sherry

2 Facts The Final is next week. It mostly covers things after the midterm, but there is significant material from before. There were two TCP questions on the original version of the midterm, but we cut one because it was too long. That TCP question is now on the final. The big things to know are Security, Wireless, Link Layer, and QoS/ Prioritization.

3 Your Piazza Questions broke down in to two camps: TCP/CWNDs and CSMA/CD, CA, etc.

4 What is the relationship between CWND and BDP?

5 What is CWND?

6 CWND: The number of unacked bytes that TCP can have outstanding.

7 CWND is another way of saying: how many bytes should TCP be able to have in the network without B having received them yet?

8 Abstract View of Link A B Sending Host Receiving Host

9 Abstract View of Link A B Sending Host Receiving Host

10 Abstract View of Link A B Sending Host Receiving Host A sends to B at a rate of r Mbps over a link with latency d millisec.

11 Abstract View of Link A B Sending Host Receiving Host A sends to B at a rate of r Mbps over a link with latency d millisec. In b milliseconds, A can stuff r*d packets into the pipe These packets have been sent by A, but have not yet been received by B

12 Abstract View of Link A B Sending Host Receiving Host r * d is our bandwidth-delay product AKA how much stuff fits in the pipe at a time.

13 Big Idea: if A maintains that there are r * d bytes in that pipe constantly, then A can keep the pipe full meaning achieve the maximum possible sending rate through that pipe.

14 Hence, perfect CWND is 2 * r * d (or r * RTT). Why * 2?

15 Need to wait for the ACKs!

16 Abstract View of Link A B Sending Host Receiving Host 2 * r * d total packets in the network.

17 From the homework: Wireshark does not show CWND directly. Give your best estimation about the initial CWND and explain how did you get the value.

18 Anyone want to explain how they did it?

19 Count the number of packets in the pipe! A B Sending Host Receiving Host After the handshake, you should have seen some number of packets leave A how many packets left A before you received the first ACK back for that first packet after the handshake?

20 Selective ACKs in TCP

21 SACKs are useful when you have multiple losses.

22 Seq = 1 Seq = 2 By default, TCP uses cumulative ACKS. Seq = 3 Seq = 4 Return one value: the next byte expected.

23 ACK 2 ACK 3 ACK 4 ACK 5 Seq = 1 Seq = 2 Seq = 3 Seq = 4 By default, TCP uses cumulative ACKS. Return one value: the next byte expected.

24 Seq = 1 ACK 2 On loss: only know one byte that was lost: Seq = 4 the next byte ACK 2 expected.

25 Seq = 1 ACK 2 On loss: only know one byte that was lost: Seq = 4 the next byte expected.

26 Seq = 1 ACK 2 On loss: only know one byte that was lost: Seq = 4 the next byte ACK 2 expected. Seq = 2 ACK 3

27 Seq = 1 ACK 2 On loss: only know one byte that was lost: Seq = 4 the next byte ACK 2 expected. Seq = 2 ACK 3 Seq = 3 ACK 5

28 Seq = 1 ACK [1, _, _, _] With SACK, we send back a scoreboard Seq = 4 of ACKs rather than a ACK [1, _, _, 4] cumulative ACK.

29 Seq = 1 ACK [1, _, _, _] With SACK, we send back a scoreboard Seq = 4 of ACKs rather than a ACK [1, _, _, 4] cumulative ACK. ACK [2, _, 4, _] ACK [5, _, _, _] Seq = 2 Seq = 3 Can retransmit more efficiently!

30 The difference between CSMA/CA and CSMA/CD

31 Quick, someone describe CSMA/CD!

32 Key to CSMA/CD: everyone who wants to transmit can detect when collision happens. (This is not the case in wireless.)

33 Hidden Terminals A B C 27

34 Hidden Terminals A B C 27

35 Hidden Terminals A B C transmit range 27

36 Hidden Terminals A B C transmit range A and C can both send to B but can t hear each other A is a hidden terminal for C and vice versa 27

37 Hidden Terminals A B C transmit range A and C can both send to B but can t hear each other A is a hidden terminal for C and vice versa Carrier Sense will be ineffective 27

38 Exposed Terminals A B C D 28

39 Exposed Terminals A B C D 28

40 Exposed Terminals A B C D 28

41 Exposed Terminals A B C D Exposed node: B sends a packet to A; C hears this and decides not to send a packet to D (despite the fact that this will not cause interference)! 28

42 Exposed Terminals A B C D Exposed node: B sends a packet to A; C hears this and decides not to send a packet to D (despite the fact that this will not cause interference)! Carrier sense would prevent a successful transmission. 28

43 Key Points 29

44 Key Points No concept of a global collision Different receivers hear different signals Different senders reach different receivers 29

45 Key Points No concept of a global collision Different receivers hear different signals Different senders reach different receivers Collisions are at receiver, not sender Only care if receiver can hear the sender clearly It does not matter if sender can hear someone else As long as that signal does not interfere with receiver 29

46 Key Points No concept of a global collision Different receivers hear different signals Different senders reach different receivers Collisions are at receiver, not sender Only care if receiver can hear the sender clearly It does not matter if sender can hear someone else As long as that signal does not interfere with receiver Goal of protocol: Detect if receiver can hear sender Tell senders who might interfere with receiver to shut up 29

47 Basic Collision Avoidance 30

48 Basic Collision Avoidance Since can t detect collisions, we try to avoid them 30

49 Basic Collision Avoidance Since can t detect collisions, we try to avoid them Carrier sense: When medium busy, choose random interval Wait that many idle timeslots to pass before sending 30

50 Basic Collision Avoidance Since can t detect collisions, we try to avoid them Carrier sense: When medium busy, choose random interval Wait that many idle timeslots to pass before sending When a collision is inferred, retransmit with binary exponential backoff (like Ethernet) Use ACK from receiver to infer no collision Use exponential backoff to adapt contention window 30

51 CSMA/CA -MA with Collision Avoidance sender receiver other node in sender s range Before every data transmission 31

52 CSMA/CA -MA with Collision Avoidance sender RTS receiver other node in sender s range Before every data transmission Sender sends a Request to Send (RTS) frame containing the length of the transmission 31

53 CSMA/CA -MA with Collision Avoidance sender RTS CTS receiver other node in sender s range Before every data transmission Sender sends a Request to Send (RTS) frame containing the length of the transmission Receiver respond with a Clear to Send (CTS) frame 31

54 CSMA/CA -MA with Collision Avoidance sender RTS CTS data receiver other node in sender s range Before every data transmission Sender sends a Request to Send (RTS) frame containing the length of the transmission Receiver respond with a Clear to Send (CTS) frame Sender sends data 31

55 CSMA/CA -MA with Collision Avoidance sender RTS CTS data ACK receiver other node in sender s range Before every data transmission Sender sends a Request to Send (RTS) frame containing the length of the transmission Receiver respond with a Clear to Send (CTS) frame Sender sends data Receiver sends an ACK; now another sender can send data 31

56 CSMA/CA -MA with Collision Avoidance sender RTS CTS data ACK receiver other node in sender s range Before every data transmission Sender sends a Request to Send (RTS) frame containing the length of the transmission Receiver respond with a Clear to Send (CTS) frame Sender sends data Receiver sends an ACK; now another sender can send data When sender doesn t get a CTS back, it assumes collision 31

57 CSMA/CA, con t receiver sender RTS other node in sender s range CTS data data 32

58 CSMA/CA, con t receiver sender RTS other node in sender s range CTS data data If other nodes hear RTS, but not CTS: send Presumably, destination for first sender is out of node s range 32

59 CSMA/CA, con t sender RTS CTS receiver other node in sender s range If other nodes hear RTS, but not CTS: send Presumably, destination for first sender is out of node s range Can cause problems when a CTS is lost 33

60 CSMA/CA, con t sender RTS CTS data ACK receiver other node in sender s range If other nodes hear RTS, but not CTS: send Presumably, destination for first sender is out of node s range Can cause problems when a CTS is lost When you hear a CTS, you keep quiet until scheduled transmission is over (hear ACK) 33

61 RTS / CTS Protocols (CSMA/CA) B sends to C B C 34

62 RTS / CTS Protocols (CSMA/CA) B sends to C A B C D Overcome hidden terminal problems with contention-free protocol 34

63 RTS / CTS Protocols (CSMA/CA) B sends to C A RTS B C D Overcome hidden terminal problems with contention-free protocol 1. B sends to C Request To Send (RTS) 34

64 RTS / CTS Protocols (CSMA/CA) B sends to C A RTS B C D Overcome hidden terminal problems with contention-free protocol 1. B sends to C Request To Send (RTS) 2. A hears RTS and defers (to allow C to answer) 34

65 RTS / CTS Protocols (CSMA/CA) B sends to C A RTS B C D CTS Overcome hidden terminal problems with contention-free protocol 1. B sends to C Request To Send (RTS) 2. A hears RTS and defers (to allow C to answer) 3. C replies to B with Clear To Send (CTS) 34

66 RTS / CTS Protocols (CSMA/CA) B sends to C A RTS B C D CTS Overcome hidden terminal problems with contention-free protocol 1. B sends to C Request To Send (RTS) 2. A hears RTS and defers (to allow C to answer) 3. C replies to B with Clear To Send (CTS) 4. D hears CTS and defers to allow the data 34

67 RTS / CTS Protocols (CSMA/CA) B sends to C A RTS B C D CTS Overcome hidden terminal problems with contention-free protocol 1. B sends to C Request To Send (RTS) 2. A hears RTS and defers (to allow C to answer) 3. C replies to B with Clear To Send (CTS) 4. D hears CTS and defers to allow the data 5. B sends to C 34

68 Alt: Preventing Collisions Altogether 35

69 Alt: Preventing Collisions Altogether Partition space into non-overlapping cells. 35

70 Alt: Preventing Collisions Altogether Partition space into non-overlapping cells. 35

71 Why doesn t Google use anycast to redirect traffic to its servers instead of DNS redirection?

72 Answer: performance

73 Answer: performance Recall: routing over the Internet follows sub-optimal paths. Why?

74 Answer: performance Recall: routing over the Internet follows sub-optimal paths. Why? Follow the money policies (choose cheap routes over short ones)

75 Answer: performance Recall: routing over the Internet follows sub-optimal paths. Why? Follow the money policies (choose cheap routes over short ones) AS path lengths don t necessarily correlate to shortest geographic paths / lowest latencies anyway

76 Answer: performance Recall: routing over the Internet follows sub-optimal paths. Why? Follow the money policies (choose cheap routes over short ones) AS path lengths don t necessarily correlate to shortest geographic paths / lowest latencies anyway And it s common for operators to just misconfigure stuff too.

77 If we use anycast, we re letting the sub-optimal routing protocols take charge: no guarantee they ll pick the best choice. google.com AS 2 AS 3 google.com ? AS 1 google.com AS 4

78 With DNS redirection, Google can pick best route themselves, based on what IP address their DNS server returns for the domain name google.com AS 2 AS 3 google.com AS 1 google.com AS 4

79 Bonus question: if anycast leads to sub-optimal routing, why does Google DNS ( ) use anycast?

80 Answer: it s a chicken and egg problem. You can t use DNS redirection to find your optimal DNS resolver, because you d need to use your DNS resolver to look up its own DNS name

81 What is a PKI?

82 Public Key Infrastructure

83 Public Key Infrastructure How do I get the public key of someone I don t know? How do I know it s them?

84 Public Key Infrastructure How do I get the public key of someone I don t know? How do I know it s them? I can go find them in the real world, check their ID card, get their public key.

85 Public Key Infrastructure How do I get the public key of someone I don t know? How do I know it s them? I can go find them in the real world, check their ID card, get their public key. Or, I can ask a central database that authoritatively stores everyone s public key.

86 Public Key Infrastructure How do I get the public key of someone I don t know? How do I know it s them? I can go find them in the real world, check their ID card, get their public key. Or, I can ask a central database that authoritatively stores everyone s public key. Everyone registers with the central database, and anyone can look up anyone else s public key in the database.

87 Public Key Infrastructure How do I get the public key of someone I don t know? How do I know it s them? I can go find them in the real world, check their ID card, get their public key. Or, I can ask a central database that authoritatively stores everyone s public key. Everyone registers with the central database, and anyone can look up anyone else s public key in the database. Implication: everyone has to trust the key provider.

88 What does the Hamming Distance figure in the lecture slides mean?

89 I don t understand the figure either.

90 Here s a review of Hamming Distances though.

91 How many bit flips do you need to turn one binary string into another? That s it. That s Hamming distance.

92 Why do Hamming distances matter? Hamming distances put a bound on how much a Forward Error Correction algorithm can fix corrupted bitstrings back to their original value. If the valid words of a code have minimum Hamming distance D, then D-1 bit errors can be detected. If the valid words of a code have minimum Hamming distance D, then [(D-1)/2] bit errors can be corrected.

93 Good luck! My office hours ARE being held next week, but Prof Steenkiste s are not. TA s OH are pending their own exam schedules.