CS144 Review Session: Fall 2010 Behram Mistree

Transcription

1 CS144 Review Session: Fall 2010 Behram Mistree

2 Outline-ish Logistics What we're looking for/common errors Review with my questions mixed in Your questions

3 Logistics Exam on Monday, Nvidia auditorium (same room as lecture). Open book/open notes. No computers/calculators/phones/etc. Same types of questions as midterm, but more. ~50:50 old:new.

4 Logistics Exam on Monday, Nvidia auditorium (same room as lecture). Open book/open notes. No computers/calculators/phones/etc. Same types of questions as midterm, but more. ~50:50 old:new. Lots of max-ing!

5 Common Error: Explain Your Reasoning Your reasoning actually matters! You are unlikely to receive even ¼ of the problem's points if you choose the correct option, but do not have a good explanation. Explain your reasoning = Tell us what's write about your answer and what's wrong with the others We're looking for specific things.

6 Toy Question Why do we have a time_wait state in TCP? What would break without it?

7 Toy Question Why do we have a time_wait state in TCP? What would break without it? An answer: TCP connections might not shut down correctly without the time_wait state. An answer++: A node that has already sent a fin must send an ack in response to a fin received from who it is communicating with. If we did not have a time_wait state, if that ack got lost, one side of the connection would close while the other waited indefinitely for an ack.

8 Review: Lecture 1 Protocol Layering: Usually building on top of something else. Hourglass: Where is the narrow waist of the OSI model?

9 Review: Lecture 1 Protocol Layering: Usually building on top of something else. Hourglass: Where is the narrow waist of the OSI model? Network Layer (IP). There are different ways to measure performance. What are a few?

10 Lecture 2: What happened to my packet? The network can drop, duplicate, corrupt, and re-order your packets. There are a few general techniques that we use to address these challenges.

11 Lecture 2: What happened to my packet? The network can drop, duplicate, corrupt, and re-order your packets. There are a few general techniques that we use to address these challenges. Name some!

12 Lecture 2: What happened to my packet? The network can drop, duplicate, corrupt, and re-order your packets. There are a few general techniques that we use to address these challenges. Checksums: corruption Acks/timeouts/retransmissions: drops Sequence numbers: duplicates + re-ordering

13 Lecture 2 ctd: End-to-end + TCP Intro End-to-end model: (Roughly) Push all the intelligence to the ends of the network. TCP has a 3-way handshake. 7 packets must be sent in every TCP session. What does this mean for short messages?

14 Lecture 2 ctd: End-to-end + TCP Intro End-to-end model: (Roughly) Push all the intelligence to the ends of the network. TCP has a 3-way handshake. 7 packets must be sent in every TCP session. What does this mean for short messages? Short messages have lots of overhead. Let's say that you have a browser running. Every 15 seconds, you click on a box to get 1 byte of data from the website. Should you use TCP?

15 Lecture 2 ctd: End-to-end + TCP Intro End-to-end model: (Roughly) Push all the intelligence to the ends of the network. TCP has a 3-way handshake. 7 packets must be sent in every TCP session. What does this mean for short messages? Short messages have lots of overhead. Let's say that you have a browser running. Every 15 seconds, you click on a box to get 1 byte of data from the website. Should you use TCP? Sure. Just use persistent connections.

16 Lecture 3 + 4: TCP What is the difference between flow control and congestion control?

17 Lecture 3 + 4: TCP What is the difference between flow control and congestion control? How long (in ms) does a TCP packet observes before re-transmitting?

18 Lecture 3 + 4: TCP What is the difference between flow control and congestion control? How long (in ms) does a TCP packet observes before re-transmitting? Trick question: TCP dynamically measures the RTT for packets.

19 Lecture 3 + 4: TCP What is the difference between flow control and congestion control? How long (in ms) does a TCP packet observes before re-transmitting? Trick question: TCP dynamically measures the RTT for packets. TCP uses additive increase, multiplicative decrease. What shape should you expect a plot of additive increase, multiplicative decrease to have?

20 Lecture 3+4: TCP Spend a while with this diagram. Make it make sense.

21 Lecture 5: IPv4 IPv4 allows fragmentation. MF (More Fragment) bit set in fragments on all but last fragment. Offset field (in IP header) changes (in multiples of 8 bytes).

22 Lecture 5: IPv4 IPv4 allows fragmentation. MF (More Fragment) bit set in fragments on all but last fragment. Offset field (in IP header) changes (in multiples of 8 bytes). Does fragmenting a packet make it more or less likely that it reaches its destination?

23 Lecture 5 Longest Prefix Match indicates a router is closer to destination than other routers. CIDR: Allows finer resolution of IP block allocation. DHCP: Automatically assigns an IP address to a node in the network.

24 Lecture 5 Longest Prefix Match indicates a router is closer to destination than other routers. CIDR: Allows finer resolution of IP block allocation. DHCP: Automatically assigns an IP address to a node in the network. Name one way DHCP may change for Ipv6. Name one security issue associated with DHCP.

25 Lecture 6: Intra-domain routing Internet is composed of a lot of Autonomous Systems (Ases). Can manage their internal network however they want using intra-domain routing.

26 Lecture 6: Intra-domain routing Internet is composed of a lot of Autonomous Systems (Ases). Can manage their internal network however they want using intra-domain routing. Name 2 ways.

27 Lecture 6: Intra-domain routing Internet is composed of a lot of Autonomous Systems (Ases). Can manage their internal network however they want using intra-domain routing. OSPF: Link state algorithm. Has global knowledge. RIP: Distance vector algorithm. Watch out for the loops!

28 Lecture 6: Intra-domain routing Internet is composed of a lot of Autonomous Systems (Ases). Can manage their internal network however they want using intra-domain routing. OSPF: Link state algorithm. Has global knowledge. RIP: Distance vector algorithm. Watch out for the loops! Split horizon: If you're using A as your next hop for a route to B, do not advertise to A a route to B. Split horizon + Poison Reverse: If you're using A as your next hop for a route to B, advertise to A a route to B with a cost of infinity.

29 Lecture 6: Intra-domain routing Internet is composed of a lot of Autonomous Systems (Ases). Can manage their internal network however they want using intra-domain routing. OSPF: Link state algorithm. Has global knowledge. RIP: Distance vector algorithm. Watch out for the loops! Split horizon: If you're using A as your next hop for a route to B, do not advertise to A a route to B. Split horizon + Poison Reverse: If you're using A as your next hop for a route to B, advertise to A a route to B with a cost of infinity. Know Bellman-Ford and Djikstra's

30 Lecture 7: BGP Border Gateway Protocol used between Autonomous Systems. Loose notion of trust. Advertisements can be read as "I can reach /24 through ASes ". Each AS makes policy decisions about who it wants to route through.

31 Review: The Physical Layer Bottom-most layer of OSI model. What is the signal? How do we map from chips to bits?

32 Review: The Physical Layer Bottom-most layer of OSI model. What is the signal? How do we map from chips to bits? What's a chip?

33 What's a chip? A chip is essentially a physical layer bit. A bit is a quanta of information above the physical layer. Big task of physical layer is mapping chips to bits.

34 What's a chip? A chip is essentially a physical layer bit. A bit is a quanta of information above the physical layer. Big task of physical layer is mapping chips to bits.

35 What's a chip? A chip is essentially a physical layer bit. A bit is a quanta of information above the physical layer. Big task of physical layer is mapping chips to bits. Is the ratio Chips/Bits generally: > 1? = 1? < 1?

36 What's a chip? A chip is essentially a physical layer bit. A bit is a quanta of information above the physical layer. Big task of physical layer is mapping chips to bits. Is the ratio Chips/Bits generally: > 1? = 1? < 1? Why?

37 Lecture 10: TCP Friendly and NAT TCP was carefully engineered to behave a particular way under congestion other protocols could interfere with this behavior. To be TCP friendly, protocol must have a goodput proportional to 1/sqrt(p), where p is the packet loss rate.

38 Lecture 11: Wireless How is wireless different from wired?

39 Lecture 11: Wireless How is wireless different from wired? Innately broadcast. Different medium....

40 Lecture 11: Wireless How is wireless different from wired? Innately broadcast. Different medium. Signal strength decays over space faster in wireless than wired networks.

41 Ramifications of signal decay Bits more frequently lost/corrupted Coding. Link layer acks. Interacts poorly with TCP Routing: don't use hopcount metric Use ETX instead. Hidden terminal problem Full-duplex

42 Lecture 12: Fairness and RED Fairness Challenge: want to assure that each connection has its fair share of bandwidth, but costly to maintain highly detailed statistics on connection state. Practice problem about drop policy. Given example was UDP. Really encourage students to go through what would happen if both senders were TCP.

43 Lecture 12: Fairness and RED Use RED

44 Lecture 13: Coding In your labs, what did you do when a CRC failed?

45 Lecture 13: Coding In your labs, what did you do when a CRC failed? Discarded packet. Okay solution when rarely get corrupted packet. Not so great when using a noisy channel.

46 Lecture 13: Coding A number of different algorithms: Reed-Solomon, LT, etc. Good to know about: not specific to networking (cds, etc. use them).

47 Lecture 13: Coding A number of different algorithms: Reed-Solomon, LT, etc. Good to know about: not specific to networking (cds, etc. use them). Use coding to add redundancy to messages/packets. Benefit: more robust to errors. Cost: frequently unclear how much redundancy to add. Cost: increases number of bits transmitted to bits used by higher layers.

48 Example problem

49

50 Problem Hints Hint 1: Can corrupt up to 16 bytes of data and still recover (according to Reed-Solomon).

51 Problem Hints Hint 1: Can corrupt up to 16 bytes of data and still recover (according to Reed-Solomon). Hint 2: Find the smallest number of chips needed to corrupt 17 bytes of data and subtract one chip.

52 Problem Solution 983 chips. Smallest number of chips that can corrupt a 17 byte packet: 12 chips off in byte 0 64 chips in byte chips off in byte chips = * Subtract 1 chip to guarantee that only get 16 bytes off.

53 Problem Summary Be careful of boundaries between chips and bits. (Note: some packets with 984 chips will be decoded correctly. Give an example.) Protocol layering in action! Questions may ask about how layers interact.

54 Security Lectures General concern of students with the security lectures has been we understand what each attack is when it's presented, but if faced with a new protocol to evaluate what should I look for?

55 Security Lectures General concern of students with the security lectures has been we understand what each attack is when it's presented, but if faced with a new protocol to evaluate what should I look for? Multi-billion dollar question.

56 Security Lectures: Behram's Advice 1) What types of attacks are available/are you trying to engender? Availability Integrity: Pretend to be someone else. Secrecy: Read someone else's message.

57 Security and Availability An availability attack is usually only going to be effective if you can exploit some type of assymetry between clients and hosts.

58 Security and Availability An availability attack is often only going to be effective if you can exploit some type of asymmetry between clients and hosts.

59 Security and Availability What did I mean when I said asymmetry?

60 Security and Availability What did I mean when I said asymmetry? Resource asymmetry: I have 10,000 machines, you have 1.

61 Security and Availability What did I mean when I said asymmetry? Resource asymmetry: I have 10,000 machines, you have 1. Amplification-induced asymmetry: I can send one packet that turns into 40. Name a few examples.

62 Security and Availability What did I mean when I said asymmetry? Resource asymmetry: I have 10,000 machines, you have 1. Amplification-induced asymmetry: I can send one packet that turns into 40. One packet to DNS can engender many other packets Fragmentation

63 Security and Availability What did I mean when I said asymmetry? Resource asymmetry: I have 10,000 machines, you have 1. Amplification-induced asymmetry: I can send one packet that turns into 40. One packet to DNS can engender many other packets Fragmentation Computational asymmetry: I can force you to do an expensive computation that isn't expensive for me.

64 Security and Availability If I saw a question with an asymmetry like the above I'd start thinking about availability attacks. However, there are other types of availability attacks.

65 Security and Availability If I saw a question with an asymmetry like the above I'd start thinking about availability attacks. However, there are other types of availability attacks. Walk up to a computer a network's router and hit it with a baseball bat. SYN-bombs.

66 Security and Integrity If I'm a bank, I want to know that the message Withdraw $10,000 from Account X and put it in Account Y is really from Account X's owner.

67 Security and Integrity If I'm a bank, I want to know that the message Withdraw $10,000 from Account X and put it in Account Y is really from Account X's owner.

68 Security and Integrity If I'm a bank, I want to know that the message Withdraw $10,000 from Account X and put it in Account Y is really from Account X's owner. What general tool do we use to help ensure integrity of a packet?

69 Security and Integrity If I'm a bank, I want to know that the message Withdraw $10,000 from Account X and put it in Account Y is really from Account X's owner. What general tool do we use to help ensure integrity of a packet? Message Authentication Codes (MACs)

70 MAC: General idea Sender and receiver share a secret K. Some function, MAC, exists which takes in a message and K to produce v. From the message, it is very difficult to produce valid v without knowing K.

71 Security and Integrity: Behram's Advice Unless it's painfully obvious that the MAC function is bad, look for a replay attack.

72 Security and Secrecy General tool for secrecy is encryption. Look for potential misuses of encryption Ignoring man-in-the-middle attacks Using the same key with a stream-cipher multiple times Statistical anomalies from message or encryption.

73 Security: Behram thoughts Popular target is DNS Important/used in a lot of protocols Attacker has some ability to directly interfere (ie answering query to its name server) Look for replay attacks Look for asymmetries

74 Lecture 16: IPv6 Tell me two things that are different between IPv6 and IPv4.

75 Lecture 16: IPv6 Tell me two things that are different between IPv6 and IPv4. Lots of choices: Larger address space; No fragmentation; Simplified header; Etc.

76 Good luck!