Missing Pieces of the Puzzle

Transcription

1 Missing Pieces of the Puzzle EE122 Fall 2011 Scott Shenker Materials with thanks to Jennifer Rexford, Ion Stoica, Vern Paxson and other colleagues at Princeton and UC Berkeley 1

2 Announcements Everyone should be signed up and have account If you aren t see me at end of class We lost our grader, so delaying HW#2 Gives us a chance to cover more material Beanbags. Really? 2

3 Agenda for Today Finish up routing: not as simple as you might think Discuss missing pieces: preview of rest of class 3

4 Last Time Link-State Routing Spread state everywhere Nodes do local computation over entire graph Local computation, global state Doesn t scale well, involves broadcasts 4

5 This Time Global computation, local state We want to limit the distribution of state So computation must be distributed The most common example is distance-vector 5

6 Distance-Vector Details in Section 6

7 Distributed Computation of Routes More scalable than Link-State No global flooding Each node computing the outgoing port based on: Local information (who it is connected to) Paths advertised by neighbors Algorithms differ in what these exchanges contain Distance-vector: just the distance to each destination Path-vector: the entire path to each destination We will focus on distance-vector for now 7

8 Example of Distributed Computation I am three hops away I am two hops away I am two hops away I am one hop away I am three hops away I am two hops away I am one hop away I am three hops away I am one hop away Destination I am two hops away 8

9 Very similar to Monday s Class Destination stands up Announces neighbors They stand up They announce their neighbors They stand up..and so on, until source stands On Monday you started with the source, but paths are reversible so it doesn t matter. Key point: don t stand up twice! 9

10 Step 1 Destination stands up 10

11 Step 1 11

12 Step 2 Destination stands up Announces neighbors They stand up 12

13 Step 2 I am one hop away I am one hop away I am one hop away 13

14 Step 3 Destination stands up Announces neighbors They stand up They announce their neighbors They stand up 14

15 Step 3 I am two hops away I am two hops away I am two hops away I am two hops away 15

16 Why Not Stand Up Twice? Being called a second time means that there is a second (and longer) path to you You already contacted your neighbors the first time Your distance to destination is based on shorter path 16

17 Basics of Distributed Routing Nodes advertise their best paths to neighbors Nodes select among paths offered by neighbors Ignore all but the best path Iterative process eventually leads to convergence If cost is hopcounts, then nodes find out about shortest paths first (ignore later ones) If costs are general, later paths might be better Remember: nodes select among offered paths But what selection criteria are allowed? 17

18 No agreement on metrics? If the nodes choose their paths according to different criteria, then bad things might happen Example Node A is minimizing latency Node B is minimizing loss rate Node C is minimizing price Any of those goals are fine, if globally adopted Only a problem when nodes use different criteria 18

19 What Happens Here? Cares about price, then loss Low price link Cares about delay, then price Low loss link Low delay link Low loss link Low delay link Low price link Cares about loss, then delay 19

20 Can You Use Any Metric? Are there any metrics that won t work, even if everyone agrees on it? What about maximizing capacity? 20

21 What Happens Here? A high All capacity nodes link want gets to maximize reduced to capacity low capacity 21

22 Must agree on loop-avoiding metric When all nodes minimize same metric And that metric increases around loops Then process is guaranteed to converge On to the details of the algorithm. Nothing interesting, but will be on test. 22

23 Distance Vector Routing Each router knows the links to its neighbors Does not flood this information to the whole network Each router has provisional shortest path E.g.: Router A: I can get to router B with cost 11 via next hop router D Routers exchange this Distance-Vector information with their neighboring routers Vector because one entry per destination Routers update their idea of the best path using info from neighbors Iterative process converges to set of shortest paths 23

24 Information Flow in Distance Vector Host C Host A Host D N1 N2 N3 N5 Host B N4 Host E N6 N7 24

25 Information Flow in Distance Vector Host C Host A Host D N1 N2 N3 N5 Host B N4 Host E N6 N7 25

26 Information Flow in Distance Vector Host A Host C Why is this different from flooding? Host D N1 N2 N3 N5 Host B N4 Host E N6 N7 26

27 Bellman-Ford Algorithm INPUT: Link costs to each neighbor Not full topology OUTPUT: Next hop to each destination and the corresponding cost Does not give the complete path to the destination My neighbors tell me how far they are from dest n Compute: (cost to nhbr) plus (nhbr s cost to destination) Pick minimum as my choice Advertise that cost to my neighbors Next few slides show the corrosive power of ppt 27

28 Bellman-Ford - Overview Each router maintains a table Best known distance from X to Y, via Z as next hop = D Z (X,Y) Each local iteration caused by: Local link cost change Message from neighbor Notify neighbors only if least cost path to any destination changes Neighbors then notify their neighbors if necessary Each node: wait for (change in local link cost or msg from neighbor) recompute distance table if least cost path to any dest has changed, notify neighbors 28

29 Bellman-Ford - Overview Each router maintains a table Row for each possible destination Column for each directly-attached neighbor to node Entry in row Y and column Z of node X best known distance from X to Y, via Z as next hop = D Z (X,Y) Node A Neighbor (next-hop) A 2 B C 1 D B C B 2 8 C 3 7 D 4 8 Destinations D C (A, D)

30 Bellman-Ford - Overview Each router maintains a table Row for each possible destination Column for each directly-attached neighbor to node Entry in row Y and column Z of node X best known distance from X to Y, via Z as next hop = D Z (X,Y) Node A A 2 B C 1 D B C B 2 8 C 3 7 D 4 8 Smallest distance in row Y = shortest Distance of A to Y, D(A, Y)

31 Distance Vector Algorithm (cont d) 1 Initialization: 2 for all neighbors V do c(i,j): link cost from node i to j 3 if V adjacent to A D 4 D(A, V) = c(a,v); Z (A,V): cost from A to V via Z 5 else 6 D(A, V) = ; 7 send D(A, Y) to all neighbors loop: 8 wait (until A sees a link cost change to neighbor V /* case 1 */ 9 or until A receives update from neighbor V) /* case 2 */ 10 if (c(a,v) changes by ±d) /* case 1 */ 11 for all destinations Y that go through V do 12 D V (A,Y) = D V (A,Y) ± d 13 else if (update D(V, Y) received from V) /* case 2 */ /* shortest path from V to some Y has changed */ 14 D V (A,Y) = D V (A,V) + D(V, Y); /* may also change D(A,Y) */ 15 if (there is a new minimum for destination Y) 16 send D(A, Y) to all neighbors 17 forever D(A,V): cost of A s best path to V 31

32 Example:1 st Iteration (C A) A 2 B loop: 13 else if (update D(A, Y) from C) 14 D C (A,Y) = D C (A,C) + D(C, Y); 15 if (new min. for destination Y) 16 send D(A, Y) to all neighbors 17 forever 3 C 1 D Node A B C B 2 8 C 7 D 8 Node C A B D A 7 B 1 D 1 Node B A C D A 2 C 1 D 3 D C (A, B) = D C (A,C) + D(C, B) = = 8 D C (A, D) = D C (A,C) + D(C, D) = = 8 Node D B C A B 3 C 1

33 Example: 1 st Iteration (B A) A 2 B loop: 13 else if (update D(A, Y) from B) 14 D B (A,Y) = D B (A,B) + D(B, Y); 15 if (new min. for destination Y) 16 send D(A, Y) to all neighbors 17 forever 3 C 1 D Node A Node B B C A C D B 2 8 A 2 C 3 7 C 1 D 5 8 D 3 D B (A, C) = D B (A,B) + D(B, C) = = 3 D B (A, D) = D B (A,B) + D(B, D) = = 5 Node C Node D A B D B C A 7 A B 1 B 3 D 1 C 1

34 Example: End of 1 st Iteration Node A Node B B C A C D A 2 B C 1 D B 2 8 C 3 7 D 5 8 A 2 8 C D 2 3 End of 1 st Iteration All nodes knows the best two-hop paths Node C A B D A 7 3 B D 4 1 Node D B C A 5 8 B 3 2 C 4 1

35 Where What How harm does could does this we 57 this fix come this? cause? from? Example: 2 nd Iteration (A B) Node A Node B B C A C D A 2 B C 1 D B 2 8 C 3 7 D 5 8 A 2 3 C D loop: 13 else if (update D(B, Y) from A) 14 D A (B,Y) = D A (B,A) + D(A, Y); 15 if (new min. for destination Y) 16 send D(B, Y) to all neighbors 17 forever D A (B, C) = D A (B,A) + D(A, C) = = 5 D A (B, D) = D A (B,A) + D(A, D) = = 7 Node C Node D A B D B C A 7 3 A 5 8 B B 3 2 D 4 1 C 4 1

36 Example: End of 2 nd Iteration Node A Node B B C A C D A 2 B C 1 D B 2 8 C 3 7 D 4 8 A C D End of 2 nd Iteration All nodes knows the best three-hop paths Node C A B D A B D Node D B C A 5 4 B 3 2 C 4 1

37 Example: End of 3rd Iteration Node A Node B B C A C D A 2 B C 1 D B 2 8 C 3 7 D 4 8 A C D End of 2 nd Iteration: Algorithm Converges! Node C A B D A B D Node D B C A 5 4 B 3 2 C 4 1

38 Intuition Initial state: best one-hop paths One round: best two-hop paths Two rounds: best three-hop paths Kth round: best (k+1) hop paths This must eventually converge.but how does it respond to changes in cost? 38

39 Distance Vector: Link Cost Changes loop: 8 wait (until A sees a link cost change to neighbor V 9 or until A receives update from neighbor V) / 10 if (c(a,v) changes by ±d) /* case 1 */ 11 for all destinations Y that go through V do 12 D V (A,Y) = D V (A,Y) ± d 13 else if (update D(V, Y) received from V) /* case 2 */ 14 D V (A,Y) = D V (A,V) + D(V, Y); 15 if (there is a new minimum for destination Y) 16 send D(A, Y) to all neighbors 17 forever A 1 4 B 50 1 C Node B Node C A C A 4 6 C 9 1 A B A 50 5 A C A 1 6 C 9 1 A B A 50 5 A C A 1 6 C 9 1 A B A 50 2 A C A 1 3 C 3 1 A B A 50 2 good news travels fast B 54 1 B 54 1 B 51 1 B 51 1 Link cost changes here time Algorithm terminates 39

40 DV: Count to Infinity Problem loop: 8 wait (until A sees a link cost change to neighbor V 9 or until A receives update from neighbor V) / 10 if (c(a,v) changes by ±d) /* case 1 */ 11 for all destinations Y that go through V do 12 D V (A,Y) = D V (A,Y) ± d 13 else if (update D(V, Y) received from V) /* case 2 */ 14 D V (A,Y) = D V (A,V) + D(V, Y); 15 if (there is a new minimum for destination Y) 16 send D(A, Y) to all neighbors 17 forever 60 A 4 B 50 1 C Node B Node C A C A C A 4 6 A 60 6 C 9 1 C 9 1 A B A B A 50 5 A 50 5 B 54 1 Link cost changes here B 54 1 A C A C A 60 6 A 60 8 C 9 1 C 9 1 A B A B A 50 7 A 50 7 B B More like avoidance: believe anything if it hides the bad news time bad news travels slowly 40

41 Distance Vector: Poisoned Reverse If B routes through C to get to A: - B tells C its (B s) distance to A is infinite (so C won t route to A via B) 60 A 4 B 50 1 C Node B A C A 4 6 C 9 1 A C A 60 6 C 9 1 A C A 60 6 C 9 1 A C A C 9 1 A C A C 9 1 Node C A B A 50 5 B 1 A B A 50 5 B 1 A B A 50 B 1 A B A 50 B 1 A B A 50 B 1 Link cost changes here; C updates D(C, A) = 60 as B has advertised D(B, A) = time Algorithm terminates 41

42 Will PR Solve C2I Problem Completely? D A B C 42

43 Another Distributed Loop Avoidance Exchange entire paths, not just cost of paths My path to the destination is <n 1, n 2, n 3, > Path-vector routing (PV) Nodes can use arbitrary metrics to choose paths No need to agree on single metric Each node can evaluate the path independently No loops, but algorithm might not converge Previous example of multiple criteria showed this 43

44 Scary Thought The Internet s interdomain routing is based on PV Domains can use arbitrary criteria to choose paths There is no guarantee that the entire Internet routing system won t oscillate We ve just been lucky so far. 44

45 Another Scary Thought In both DV and PV, what happens when node lies UCB s router says: Sure, I m one hop away from MIT What happens to traffic destined for MIT? This has happened, major outages Could this happen with Link-State? 45

46 Routing: Just the Beginning Link state and distance-vector (and path vector) are the deployed routing paradigms But we know how to do much, much better Stay tuned for a later lecture where we: Reduce convergence time to zero Respond to failures instantly! 46

47 Missing Pieces 47

48 Where are we? We have covered the fundamentals How to deliver packets (routing) How to build reliable delivery on an unreliable network With this, we could build a decent network But couldn t actually do anything with the network Too many missing pieces Today: identify those pieces Will guide what we cover rest of semester 48

49 Scenario: Jane Wants Her Music Jane is sitting in her dorm room, with a laptop Has overwhelming urge to listen to John Cage What needs to happen to make this possible? Go one step at a time 49

50 What Are The Steps Involved? Accessing the network from laptop Wireless or ethernet Network management (someone needs to make it work) Mapping real world name to network name Mapping network name to location Download content from location Finding nearby content Addressing general security concerns Verifying that this is the right content And that no one can tell what she s downloading 50

51 Access Networks If access network is switched, we understand it Just like any other packet-switched network If the access network is shared medium, then we need to figure out how to share the medium Wireless Classical ethernet 51

52 Media Access Control (MAC) Carrier sense: (CSMA) Don t send if someone else is sending Collision detection: (CD) Stop if you detect someone else was also sending Collision avoidance: (CA) How to arrange transmissions so that they don t collide 52

53 Network Management Control how network interconnects to Internet Interdomain routing Keep unwanted traffic off network Firewalls and access control Share limited number of public addresses NAT Keep links from overloading Traffic engineering Most undeveloped part of the Internet architecture 53

54 Current Network Management No abstractions, no layers Just complicated distributed algorithms Such as routing algorithms Or manual configuration Such as Access Control Lists and Firewalls 54

55 Future Network Management Clean abstractions No complicated distributed algorithms Treat networks like systems Last lecture of class. 55

56 Real World Name to Network Name Jane knows what music she wants Doesn t know how to tell network what she wants How can we do this? Need to map real world name to network name Search engine! Maps keywords to URL 56

57 Map Network Name to Location Name resolution converts name to location We would like location to be nearby copy Speeds up download Reduce load on backbone and access networks 57

58 How is this done today? Name resolution: Domain Name System (DNS) Hand in a domain, get back an IP address Nearby copy of the data? CDNs: content distribution networks (like Akamai) P2P systems can also point you to nearby content 58

59 Download Data from Location Need a reliable transfer protocol: TCP Must share network with others: congestion control But must be able to use URL to retreive content Need higher-level protocol like HTTP to coordinate 59

60 Ensuring Security Privacy: prevent sniffers from knowing what she downloaded ( it was for EE122, I promise! ) Integrity: ensure data wasn t tampered with during its trip through network Provenance: ensure that music actually came from the music company (and not some imposter) 60

61 How do we do this today? Cryptographic measures enable us to do all three Public Key cryptography is crucial No need to share secrets beforehand 61

62 Scenario Requires Media Access Control Network management Naming and name resolution Content distribution networks And perhaps P2P Congestion control HTTP Cryptographic measures to secure content 62

63 Rest of Course Details of IP and TCP Bringing reality to general concepts Filling in pieces of name resolution and HTTP Congestion control Advanced routing Security Ethernet and Wireless Network Management 63