1 / 19

Chapter 4 Distributed Bellman-Ford Routing

Chapter 4 Distributed Bellman-Ford Routing. Professor Rick Han University of Colorado at Boulder rhan@cs.colorado.edu. Announcements. Reminder: Programming assignment #1 is due Feb. 19 Homework #2 available on Web site, due Feb. 26 Hand back HW #1 next week

tamma
Download Presentation

Chapter 4 Distributed Bellman-Ford Routing

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Chapter 4Distributed Bellman-Ford Routing Professor Rick Han University of Colorado at Boulder rhan@cs.colorado.edu

  2. Announcements • Reminder: Programming assignment #1 is due Feb. 19 • Homework #2 available on Web site, due Feb. 26 • Hand back HW #1 next week • OH cancelled yesterday, send me email to meet • Next, more on IP routing, … Prof. Rick Han, University of Colorado at Boulder

  3. Recap of Previous Lecture • ARP • IP Forwarding Tables • Destination and Output Port • IP Routing • Distributed algorithm to create Forwarding Tables • Calculate shortest path to each node • Distance Vector (RIP) • Presentation should have been better by me, textbook, etc. Prof. Rick Han, University of Colorado at Boulder

  4. Bellman-Ford Equation • Distance vector & RIP based on distributed implementation of Bellman-Ford algorithm • Bellman-Ford equation: • Label routers i=A, B, C, … • Let D(i,j) = distance for best route from i to remote j • Let d(i,j) = distance from router i to neighbor j • Set to infinity if i=j or i and j not immediate neighbors Prof. Rick Han, University of Colorado at Boulder

  5. Bellman-Ford Equation (2) • Bellman-Ford equation: • D(i,j) = min {d(i,k) + D(k,j)} for all i<>j • k • neighbors • Ex. D(B,F) = min {d(B,k) + D(k,F)} • k=A,C,E Prof. Rick Han, University of Colorado at Boulder

  6. Bellman-Ford Algorithm • Bellman-Ford equation: • D(i,j) = min {d(i,k) + D(k,j)} for all i<>j • k neighbors • Bellman-Ford Algorithm solves B-F Equation: • To calculate D(i,j), node i only needs d(i,k)’s and D(k,j)’s from neighbors • Problem: don’t know D(k,j)’s • Solution: • For each node i, first find shortest distance path from i to j using one link, D(i,j)[1] • Shortest distance path using two or fewer links, D(i,j)[2], must depend on the shortest distance path using one link, namely D(i,j)[2] = min {d(i,j) + D(i,j)[1]} Prof. Rick Han, University of Colorado at Boulder

  7. Bellman-Ford Algorithm (2) • Key observation: • By induction, the best (h+1 or fewer)-hop path between nodes i and j must be arise from an i-to-neighbor link connected with a (h or fewer)-hop path from neighbor to j : • D(i,j)[h+1] = min {d(i,k) + D(k,j)[h]} • Bellman-Ford Algorithm: • D(i,j)[h+1] = min {d(i,k) + D(k,j)[h]} for all i<>j, h=0,1, … • k neighbors • Iterate h=0,1,2, … until reach diameter DM of graph • D(i,j)[DM] is the originally desired B-F solution D(i,j) ! • At each h, calculate D(i,j)[h+1] for all i<>j • At h=0, D(i,j)[0] = {0 for i=j, infinity otherwise} • D(i,i)[h] = link cost on which dist. vector is sent - 1 Prof. Rick Han, University of Colorado at Boulder

  8. Bellman-Ford Algorithm Example • Suppose C wants to find shortest path to each destination • First, calculate shortest one-link paths from each node: easy, D(i,j)[1]=d(i,j) • D(C,B)[1], D(C,D)[1], and • D(B,A)[1], D(B,E)[1], D(B,C)[1], and • D(D,E)[1], D(D,C)[1], and • D(A,B)[1], D(A,E)[1], D(A,F)[1], and • D(E,A)[1], D(E,B)[1], D(E,D)[1], D(E,F)[1], and • D(F,A)[1], D(F,E)[1] Prof. Rick Han, University of Colorado at Boulder

  9. Bellman-Ford Algorithm Example (2) • Second, calculate shortest 2-or-fewer hop paths from each node: • Example: for node C to F D(C,F)[2] = min (d(C,k) + D(k,F)[1]) for all j k neighbors = min {d(C,B) + D(B,F)[1], d(C,D) + D(D,F)[1]} • No one-link path from B to F, so D(B,F)[1] is infinity, same for D(D,F)[1] • Calculate D(i,j)[2] for all other combinations of i<>j Prof. Rick Han, University of Colorado at Boulder

  10. Bellman-Ford Algorithm Example (3) • Third, calculate shortest 3-or-fewer hop paths from each node: • Example: for node C to F D(C,F)[3] = min {d(C,B) + D(B,F)[2], d(C,D) + D(D,F)[2]} • No more unknowns: • D(B,F)[2] is known by now and was calculated in the last iteration, = min{d(B,k) + D(k,F)[1]} • D(D,F)[2] is also known • Since diameter = 3, we’re done and have found all shortest distance paths D(i,j) Prof. Rick Han, University of Colorado at Boulder

  11. Distributed Bellman-Ford Algorithm • Bellman-Ford Algorithm: • D(i,j)[h+1] = min {d(i,k) + D(k,j)[h]} for all i<>j, h=0,1, … • k neighbors • One way to implement in a real network: • Flood d(i,j) first to every router in the network • Calculate B-F Algorithm in each router • Drawbacks: • Generates lots of overhead • Requires much computation on each router • Duplication of many of calculations on each router • Consider an alternative to distribute calculations Prof. Rick Han, University of Colorado at Boulder

  12. Distributed Bellman-Ford Algorithm (2) • Bellman-Ford Algorithm: • D(i,j)[h+1] = min {d(i,k) + D(k,j)[h]} for all i<>j, h=0,1, … • k neighbors • Key observations: • We had to calculate D(i,j)[h] for each node i in the graph, at each step h in the iteration • At every iteration h, we only needed information about the h-1 or fewer hop paths to calculate D(i,j)[h] Prof. Rick Han, University of Colorado at Boulder

  13. Distributed Bellman-Ford Algorithm (3) • Therefore, in a real network, • Physically distribute the calculation of D(i,j)[h] to router i only, and • No duplication • Less calculation • Exchange the results of your D(i,j)[h] with neighboring routers at each iteration h • Less overhead • Satisfies condition that D(i,j)[h] only needs info on h-1 or less hop paths. • At iteration h, d(i,j) within radius h-1 will be propagated to all routers within radius h-1 Prof. Rick Han, University of Colorado at Boulder

  14. Distributed Bellman-Ford Algorithm (4) • In practice, convergence will eventually occur even if different routers are slow to propagate or calculate their D(i,j)[h] and/or d(i,j) • Bertsekas and Gallagher proved this, in the absence of topology changes • Distributed routing algorithm where each router only performs a small but sufficient part of the overall B-F algorithm • Node i calculates and sends D(i,j)[h] to its neighbors – this is a distancevector • Distributed Bellman-Ford Algorithm = Distance Vector Algorithm Prof. Rick Han, University of Colorado at Boulder

  15. Distance Table • Bellman-Ford Algorithm: • D(i,j)[h+1] = min {d(i,k) + D(k,j)[h]} for all i<>j, h=0,1, … • k neighbors • Each router i must maintain a “distance table”: • Must store d(i,k), D(k,j)[h] for each neighbor k and destination j Prof. Rick Han, University of Colorado at Boulder

  16. Distance Table (2) • In reality, each cell in distance table stores d(i,k) + D(k,j)[h], not just D(k,j)[h] • Must store d(i,k) or receive it within a neighbor’s distance vector advertisement • If d(i,k) is a hop, then d(i,j)=1 always, so no need to store Prof. Rick Han, University of Colorado at Boulder

  17. Routing Table At Router i Routing Table • Easy to derive a Routing Table from a distance table: choose the minimum distance in the row Prof. Rick Han, University of Colorado at Boulder

  18. Routing Information Protocol (RIP) • RIP is a specific realization of the distance vector or distributed Bellman-Ford routing algorithm • Distance vectors are carried over UDP over IP • RIP uses hop count as its shortest path metric, so d(i,j)=1 • Distance vectors are sent every 30 seconds • When a routing table changes, a router can send triggered updates to neighbors before 30 sec • Can lead to network storms, so limit rate: wait 5 seconds between sending new routing update and the update that caused routing table to change Prof. Rick Han, University of Colorado at Boulder

  19. Alternative Shortest Path Calc. • Compute a shortest path tree • Observation: • shortest path to nodes further from the root must go through a branch of the shortest path tree closer to the root • Strategy: expand outwards, calculating the shortest path tree from the root Prof. Rick Han, University of Colorado at Boulder

More Related