Lecture 18: Network Layer Link State and Distance Vector Routing

Transcription

1 Lecture 8: Network Laer Link State and Distance Vector Routing COMP 33, Spring 08 Victoria Manfredi Acknowledgements: materials adapted Computer Networking: A Top Down Approach 7 th edition: , J.F Kurose and K.W. Ross, All Rights Reserved as well as slides b Abraham Matta at Boston Universit, and some material Computer Networks b Tannenbaum and Wetherall.

2 . Announcements homework 6 due toda b :59p homework 7 posted but will likel make minor clarifications to programming part. Control plane link state routing distance vector routing compare link state vs. distance vector 3. Network programming raw sockets and bte packing vumanfredi@weslean.edu

3 Control Plane 3

4 Link state: i.e., network topolog, link costs known to all nodes, accomplished via link state broadcast msg sent to ever other node in network all nodes have same global info Computes least cost paths one source node to all other nodes obtain forwarding table for that node Given path, put st hop router for each dst in forwarding table Iterative after k iterations, know least cost path to k destinations if n nodes, loop n times vumanfredi@weslean.edu 4

5 c(,): link cost node to D(v): current cost source u to dst node v p(v): predecessor node along path source u to v N': set of nodes whose least cost path definitivel known Step N' u D(v),p(v),u D(w),p(w) 5,u D(),p(),u D(),p() D(),p() Source node u 5 v 3 3 w 5 Initialiation N' = {u} for all nodes v if v adjacent to u then D(v) = c(u,v) else D(v) = 5

6 c(,): link cost node to D(v): current cost source u to dst node v p(v): predecessor node along path source u to v N': set of nodes whose least cost path definitivel known Step Source node u N' D(v),p(v) D(w),p(w) D(),p() D(),p() D(),p() u,u 5,u,u u,u 4,, u,u 3, 4, uv 3, 4, uvw 4, uvw 5 v 3 3 w 5 Loop Find w N' s.t. D(w) is min Add w to N' Update D(v) for all neighbors v N' of w D(v) = min( D(v), D(w)+c(w,v) ) Until all nodes in N' 6

7 Step N' D(v),p(v) D(w),p(w) D(),p() D(),p() D(),p() u,u 5,u,u u,u 4,, u,u 3, 4, uv 3, 4, uvw 4, uvw Source node u Resulting shortest path tree 5 v 3 3 w 5 Forwarding table at u dst v w link (u,v) (u,) (u,) (u,) (u,) 7

8 Each iteration: need to check all nodes not in N n(n+)/ comparisons: O(n ), more efficient implementations possible Network is dnamic link goes down: link state broadcast router goes down: remove link and all nodes recompute Oscillations possible when congestion or dela-based link cost initiall w +e e e recompute routing w +e 0 0 +e 0, detect better path recompute w 0 +e 0 0 +e,, detect better path Need to prevent routers snchroniing computations: Have routers randomie when the send out link advertisements 8

9 Control Plane 9

10 Distance vector (DV) vector of best known costs router to each dst and link to use Each node maintains Link cost to each neighbor v c(,v) s own DV D (): estimate of least cost path to node D = [D (): є N ] DV for each nbr v D v (): estimate of least cost path neighbor v to node D v = [D v (): є N ] Each node periodicall sends its own DV to neighbors rather than link state costs vumanfredi@weslean.edu 0

11 Uses dnamic programming break problem into simpler sub-problems solve each sub-problem once and store solution Bellman-Ford equation D () := cost of least-cost path to D () = min { c(,v) + D v () } for each node N v cost neighbor v to destination neighbor v min taken over all neighbors v of When receives new DV estimate neighbor updates its own DV using B-F equation vumanfredi@weslean.edu

12 Bellman-Ford equation u 5 v 3 3 w 5 D u () = min { c(u,v) + D v (), c(u,) + D (), c(u,w) + D w () } = min { + 5, + 3, 5 + 3} = 4 Where D v () = 5, D () = 3, D w () = 3 Node achieving minimum is net hop in shortest path put in forwarding table vumanfredi@weslean.edu

13 Initialiation For all dst є N if is nbr of D () = c(, ) else D () = For each nbr w and dst є N: D w () = Send s DV to all nbrs w D = [D () : є N] Loop waits for change in local link cost or DV msg neighbor recompute estimates D () = min v { c(,v) + D v () } if s DV to an dst has changed, notif neighbors Q: when does loop terminate? When no more changes 3

14 D () = min{c(,)+d (), c(,)+d ()} = min{+0, 7+} = D () = min{c(,)+d (), c(,)+d ()} = min{+, 7+0} = 3 Node Node Node 7 0 vumanfredi@weslean.edu 4 time

15 D () = min{c(,)+d (), c(,)+d ()} = min{+0, 7+} = D () = min{c(,)+d (), c(,)+d ()} = min{+, 7+0} = 3 Node Node Node D () = min{c(,)+d (), c(,)+d ()} = min{7+0, +} = 3 7 vumanfredi@weslean.edu 5 time

16 Node Node Node No 0 change: 7 don t 0 send 7 out DV vumanfredi@weslean.edu 6 time

17 Node Node No 0 change: 7 don t 0 send 7 out DV0 No 0 change: 3 don t 0 send 3 out DV 0 No 0 change: 3 don t 0 send 3 out DV0 7 Node No 0 change: 3 don t 0 send 3 out DV 0 DONE vumanfredi@weslean.edu 7 time

18 . Updates routing info. Recalculates DV 3. If DV changes, notif neighbors 4 50 Good news travels fast t 0 : detects link-cost change, updates its DV, informs its neighbors t : receives update, updates its table, computes new least, sends its neighbors its DV t : receives s update, updates its distance table. Y s least costs do not change, so does not send a message to vumanfredi@weslean.edu 8

19 Count to infinit problem 44 iterations before algorithm stabilies Intuitivel when tells it has a path to, has no wa of knowing that is using on its path 60 Y Y D () = min{c(,)+d (), c(,) + D ()} = min{60+0, +5} = 6 Problem arises because Routing Loop still epects can get to with cost of 5 D () = min{c(,) + D (), c(,) + D ()} = min{50+0, +6} = 7 Count-to-infinit 9

20 If Z routes through Y to get to X Z tells Y its (Z s) distance to X is infinite (so Y won t route to X via Z) Y D () = min{c(,)+d (), c(,)+d ()} = min{60+0, + } = 60 Q: Will this completel solve count to infinit problem? no, onl for node loops Another proposed solution: hold time don t process route updates for period of time after route retraction ameliorates problem but does not solve vumanfredi@weslean.edu 0

21 Eas to implement likel ou will implement for hw8 :-) Distributed doesn t compute paths in isolation requires route info (path costs) computed b neighbors Iterative updates its DV whenever local link costs change DV update received nbr Asnchronous updates, echanges happen asnchronousl Self-terminating stops updating DV when no more changes received

22 Control Plane

23 Link state O(nE) messages sent ever node floods its link state message out over ever link in network to reach ever node smaller messages message sie depends on the number of neighbors a node has an link change requires a broadcast n nodes E links Distance vector # of messages depends on convergence time which varies nodes onl echange messages between neighbors larger routing update messages message sie is proportional to the number of nodes in the network if link changes don't affect shortest path, no message echange vumanfredi@weslean.edu 3

24 Link state n- Σ i= = n(n+)/ = O(n ) search through n- nodes to find min, recompute routes search through n- nodes to find min, recompute routes converges quickl but ma have oscillations route computation is centralied a node stores a complete view of the network n nodes E links Distance vector slow to converge and convergence time varies route computation is distributed ma be routing loops, count-to-infinit problem vumanfredi@weslean.edu 4

25 Link state node can advertise incorrect link cost each node computes onl its own table n nodes E links Distance vector DV node can advertise incorrect path cost each node s DV used b others: errors propagate through network Both have strengths and weaknesses. One or the other is used in almost ever network vumanfredi@weslean.edu 5

26 6

27 Network Programming 7

28 Take btes put into socket and push out of network interface no IP or transport laer headers added b operating sstem! Lets ou create our own transport and network laer headers set field values as ou choose e.g., time-to-live fields vumanfredi@weslean.edu 8

29 9

30 30