Chapter 5: outline. Network-layer functions. Chapter 5: outline. Routing protocols. data plane. Logically centralized control plane

Transcription

1 Chapter : otline. introdction. roting protocols distance ector. intra-as roting in the Netork-laer fnctions Recall: to netork-laer fnctions: forarding: moe packets roter s inpt to appropriate roter otpt roting: determine rote taken b packets sorce to destination data control To approaches to strctring netork control : per-roter control (traditional) logicall centralied control (softare defined netorking) - - Per-roter control Indiidal roting algorithm components in each and eer roter interact ith each other in control to compte forarding s Logicall centralied control A distinct (tpicall remote) controller interacts ith local control agents (CAs) in roters to compte forarding s Remote Controller Roting Algorithm Local forarding header otpt control data CA CA CA CA CA control data - - Chapter : otline. introdction. roting protocols distance ector. intra-as roting in the Roting protocols Roting protocol goal: determine good paths (eqialentl, rotes), sending hosts to receiing host, throgh netork of roters path: seqence of roters packets ill traerse in going gien initial sorce host to gien final destination host good : least cost, fastest, least congested roting: a top-0 netorking challenge! - -6

2 Graph abstraction of the netork Graph abstraction: costs graph: G = (N,E) N = set of roters = {,,,,, } E = set of links ={ (,), (,), (,), (,), (,), (,), (,), (,), (,) } aside: graph abstraction is sefl in other netork contets, e.g., PP, here N is set of peers and E is set of TCP connections c(, ) = cost of link (, ) e.g., c(,) = cost cold alas be, or inersel related to bandidth, or inersel related to congestion cost of path (,,,, p) = c(,) + c(,) + + c(p-,p) ke qestion: hat is the least-cost path beteen and? roting algorithm: algorithm that finds that least cost path - -8 Roting algorithm classification Q: global or decentralied information? global: all roters hae complete topolog, link cost info link state algorithms decentralied: roter knos phsicallconnected neighbors, link costs to neighbors iteratie process of comptation, echange of info ith neighbors distance ector algorithms Q: static or dnamic? static: rotes change slol oer time dnamic: rotes change more qickl periodic pdate in response to link cost changes -9 Chapter : otline. introdction. roting protocols distance ector. intra-as roting in the. -0 A link-state roting algorithm Dijsktra s algorithm Dijkstra s algorithm net topolog, link costs knon to all nodes accomplished ia link state broadcast all nodes hae same info comptes least cost paths one node ( sorce ) to all other nodes gies forarding for that node iteratie: after k iterations, kno least cost path to k dest. s notation: c(,): link cost node to ; = if not direct neighbors D(): crrent ale of cost of path sorce to dest. p(): predecessor node along path sorce to N': set of nodes hose least cost path definitiel knon - Initialiation: N' = {} for all nodes if adjacent to then D() = c(,) 6 else D() = 8 Loop 9 find not in N' sch that D() is a minimm 0 add to N' pdate D() for all adjacent to and not in N' : D() = min( D(), D() + c(,) ) /* ne is either old or knon shortest path pls cost to */ ntil all nodes in N' -

3 Dijkstra s algorithm: eample D() D() D() D() D() Step N' p() p() p() p() p() 0,,, 6,,, 6,,, 0,,, notes: constrct shortest path tree b tracing predecessor nodes ties can eist (can be broken arbitraril) Dijkstra s algorithm: another eample Step 0 N' D(),p(),,, D(),p(),,,, * Check ot the online interactie eercises for more eamples: D(),p(), D(),p(), D(),p(),,, - Dijkstra s algorithm: eample () reslting shortest-path tree : Dijkstra s algorithm, discssion algorithm compleit: n nodes each iteration: need to check all nodes,, not in N n(n+)/ comparisons: O(n ) more efficient implementations possible: O(nlogn) reslting forarding in : destination link (,) (,) (,) (,) (,) - -6 Chapter : otline. introdction. roting protocols distance ector. intra-as roting in the Distance ector algorithm Bellman-Ford eqation (dnamic programming) let d () := cost of least-cost path to then d () = min {c(,) + d () } cost neighbor to destination neighbor min taken oer all neighbors of - -8

4 Bellman-Ford eample clearl, d () =, d () =, d () = B-F eqation sas: d () = min { c(,) + d (), c(,) + d (), c(,) + d () } = min { +, +, + } = node achieing minimm is net hop in shortest path, sed in forarding Distance ector algorithm D () = estimate of least cost to maintains distance ector D = [D (): є N ] node : knos each neighbor : c(,) maintains its neighbors distance ectors. For each neighbor, maintains D = [D (): є N ] -9-0 Distance ector algorithm ke idea: time-to-time, each node sends its on distance ector estimate to neighbors hen receies ne DV estimate neighbor, it pdates its on DV sing B-F eqation: D () min {c(,) + D ()} for each node N nder minor, natral conditions, the estimate D () conerge to the actal least cost d () Distance ector algorithm iteratie, asnchronos: each local iteration cased b: local link cost change DV pdate message neighbor distribted: each node notifies neighbors onl hen its DV changes neighbors then notif their neighbors if necessar each node: ait for (change in local link cost or msg neighbor) recompte estimates if DV to an dest has changed, notif neighbors - - node 0 node 0 D() = min{c(,) + D(), c(,) + D()} = min{+0, +} = D() = min{c(,) + D(), c(,) + D()} = min{+, +0} = node 0 node 0 D() = min{c(,) + D(), c(,) + D()} = min{+0, +} = D() = min{c(,) + D(), c(,) + D()} = min{+, +0} = node 0 time - node time -

5 Distance ector: link cost changes link cost changes: node detects local link cost change pdates roting info, recalclates distance ector if DV changes, notif neighbors good nes traels fast 0 t0 : detects link-cost change, pdates its DV, informs its neighbors. t : receies pdate, pdates its, comptes ne least, sends its neighbors its DV. t : receies s pdate, pdates its distance. s least costs do not change, so does not send a message to. Distance ector: link cost changes link cost changes: node detects local link cost change bad nes traels slo - cont to infinit problem! iterations before algorithm stabilies: see tet 60 0 poisoned reerse: If Z rotes throgh Y to get to X : Z tells Y its (Z s) distance to X is infinite (so Y on t rote to X ia Z) ill this completel sole cont to infinit problem? * Check ot the online interactie eercises for more eamples: Comparison of LS and DV algorithms message compleit LS: ith n nodes, E links, O(nE) msgs sent DV: echange beteen neighbors onl conergence time aries speed of conergence LS: O(n ) algorithm reqires O(nE) msgs ma hae oscillations DV: conergence time aries ma be roting loops cont-to-infinit problem robstness: hat happens if roter malfnctions? LS: node can adertise incorrect link cost each node comptes onl its on DV: DV node can adertise incorrect path cost each node s sed b others error propagate thr netork Chapter : otline. introdction. roting protocols distance ector. intra-as roting in the - -8 Making roting scalable or roting std ths far - idealied all roters identical netork flat not tre in practice scale: ith billions of destinations: can t store all destinations in roting s! roting echange old samp links! administratie atonom internet = netork of netorks each netork admin ma ant to control roting in its on netork -9 Internet approach to scalable roting aggregate roters into regions knon as atonomos sstems (AS) (a.k.a. domains ) intra-as roting roting among hosts, roters in same AS ( netork ) all roters in AS mst rn same intra-domain protocol roters in differentas can rn different intra-domain roting protocol gatea roter: at edge of its on AS, has link(s) to inter-as roting roting among AS es gateas perform interdomain roting (as ell as intra-domain roting) roter(s) in other AS es -0

6 Interconnected ASes c a b AS a c d Intra-AS Roting b algorithm Forarding AS Inter-AS Roting algorithm c a b AS forarding configred b both intraand inter-as roting algorithm intra-as roting determine entries for destinations ithin AS inter-as & intra-as determine entries for eternal destinations - other Inter-AS tasks sppose roter in AS receies datagram destined otside of AS: roter shold forard packet to gatea roter, bt hich one? netorks c a b AS c a AS d AS mst:. learn hich dests are reachable throgh AS, hich throgh AS. propagate this reachabilit info to all roters in AS job of inter-as roting! b c a b AS other netorks - Intra-AS Roting also knon as interior gatea protocols (IGP) most common intra-as roting protocols: RIP: Roting Information Protocol OSPF: Open Shortest Path First (IS-IS protocol essentiall same as OSPF) IGRP: Interior Gatea Roting Protocol (Cisco proprietar for decades, ntil 06) OSPF (Open Shortest Path First) open : pblicl aailable ses link-state algorithm link state packet dissemination topolog map at each node rote comptation sing Dijkstra s algorithm roter floods OSPF link-state adertisements to all other roters in entire AS carried in OSPF messages directl oer IP (rather than TCP or UDP link state: for each attached link IS-IS roting protocol: nearl identical to OSPF - - OSPF adanced featres Hierarchical OSPF secrit: all OSPF messages athenticated (to preent malicios intrsion) mltiple same-cost pathsalloed (onl one path in RIP) for each link, mltiple cost metrics for different TOS (e.g., satellite link cost set lo for best effort ToS; high for real-time ToS) integrated ni- and mlti-cast spport: Mlticast OSPF (MOSPF) ses same topolog data base as OSPF hierarchical OSPF in large domains. area area border roters backbone area bondar roter internal roters backbone roter area

7 Hierarchical OSPF to-leel hierarch: local area, backbone. link-state adertisements onl in area each nodes has detailed area topolog; onl kno direction (shortest path) to nets in other areas. area border roters: smmarie distances to nets in on area, adertise to other Area Border roters. backbone roters: rn OSPF roting limited to backbone. bondar roters: connect to other AS es. Chapter : smmar e e learned a lot! approaches to netork control per-roter control (traditional) logicall centralied control (softare defined netorking) traditional roting algorithms Dijkstra Algorithm Distance ector algorithm implementation in, BGP net stop: link laer! - -8