Presentation is loading. Please wait.

Presentation is loading. Please wait.

Routing Concepts Warren Toomey GCIT. Introduction Switches need to know the link address and location of every station. Doesn't scale well, e.g. to several.

Similar presentations


Presentation on theme: "Routing Concepts Warren Toomey GCIT. Introduction Switches need to know the link address and location of every station. Doesn't scale well, e.g. to several."— Presentation transcript:

1 Routing Concepts Warren Toomey GCIT

2 Introduction Switches need to know the link address and location of every station. Doesn't scale well, e.g. to several hundred million stations on the Internet. Routers operate at OSI Layer 3. In particular, IP routers use link-independent addresses. Routers don't route packets to individual stations, but to networks.

3 Introduction Routers must do path determination: on which output interface is it best to send a packet? How to measure best? Simple hop count? Do we care about bandwidth, $$$ cost, link load etc.

4 Routing Tables Routers have a routing table. Each entry has identity of destination network/mask, cost to reach it, details of next hop to send the packet How is a router's routing table constructed? Network admin could set up several static routes, but when links fail these static routes will prevent packet forwarding across other links We want the routers to learn the network connectivity: dynamic routes

5 Static & Default Routes Static routes are fine where there is only one possible way to send traffic A default gateway is an example of a static route where a router or station can see exactly one router which is Internet-connected: send all non-local traffic to this default gateway

6 Example Routing Table If routes overlap, use the one with the biggest mask Default route has mask /0. If specified, catches all packets which are not matched by the other routes No default route in the above table, so if packet destined for 8.8.8.8 arrives, it is dropped NetworkMaskCostVia 145.67.8.128/251118.5.5.6 145.67.0.0/16896.7.1.2 220.40.1.224/2812Serial 0/0/0 17.0.0.0/8318.5.5.6

7 Equal Cost Routes R1 has two routes to bottom network, same metric R1 can send packets to R2 and R4 Packet load is shared across two links: equal cost load balancing

8 Dynamic Routes Dynamic routes are ones which change due to information learned from other routers. A routing protocol is used to exchange information about routes between routers. Dynamic routes and enough alternate paths allow routers to work around breaks or problems in the network.

9 Goals of a Routing Protocol 1.To dynamically learn and fill the routing table with a route to all networks in the overall internet. 2. If more than one route to a network is available, to put the best route in the routing table. The other routes may also be added as fallback entries. 3. To determine when routes in the table are no longer valid, and to remove them from the routing table. 4. To find and use an alternate route if an existing route is removed from the table. 5. To do steps 1, 2, 3 and 4 as quickly as possible. The time it takes to obtain the best set of routes when something changes is known as the convergence time. 6. To prevent routing loops, i.e. to prevent routers having inconsistent views of the network.

10 Routing Metrics Each static or dynamic route has a cost, or metric, associated with it. The lower the cost/metric, the better the route. Dynamic metrics are often calculated from one or more factors e.g hop count, delay, bandwidth, link load, link reliability, link cost etc.

11 Routing Metrics Different routing protocols use different metrics RIP: hop count EIGRP: bandwidth, delay, load and reliability OSPF: bandwidth

12 Routing Protocol Types Distance Vector: Routers periodically tell their neighbours about the routes they know Router learn by “word of mouth” about each network and the cumulative cost to reach it No router has a complete map of the network Link State: Routers flood the network with details of their immediate link connectivity Each router build a complete network map with link costs Each router uses their map to find the best routes

13 Routing Protocol Types Aim of any routing protocol is convergence:  Every router has a consistent routing table  Packets take their optimal route, with no loops But, links change (new ones, failed ones) Time to reach consistency is time to convergence Link State protocols converge faster than Distance Vector protocols RIP and EIGRP: Distance Vector protocols OSPF: Link State protocol

14 Distance Vector Routing Routing table entry: network, mask, cost, next router  e.g. 176.5.0.0/16, cost 12, via 11.5.6.7 DV routers advertise their routing table to neighbour routers: remove the next router, add on the cost to cross the link

15 Distance Vector Routing Routers can keep multiple entries for a route, but they always use the one with the lowest metric Routing table exchange is done periodically, so details of a link change takes time to propagate from the closest router to the furthest router RIP sends updates every 30 seconds EIGRP sends updates every 90 seconds

16 Routing Loops Routing loops occur when time to convergence is slow, and some routers are using outdated data. For example: Router C advertises a route to Network 1 to B and D

17 Count to Infinity Say C advertises “N1 cost 7” to B B sends “N1 cost 8” to D  C stops hearing advertisements from B & D  C throws out the route it has to network N1 D advertises “N1 cost 9” to C. C adds this new route to its routing table C advertises “N1 cost 10” to B This repeats forever. To stop it, treat a certain metric value and over as infinity (impossible) RIP uses 16 as its infinity metric The discarding of a route when it reaches infinity is known as "counting to infinity" Counting to infinity will eventually break the routing loop, but it takes a long time to do so

18 Split Horizon Link to N1 breaks, but R1 has routing entry to N1 R1 sends “N1 cost 2” to R2, who adds it to table Any packet going to N1 bounces between R1 & R2 Solution is Split Horizon: information learned from one router is never sent back to that same router, just in case it is now incorrect However, can pass on information learned from another router

19 Split Horizon with Poison Reverse A variation on Split Horizon is called Split Horizon with Poison Reverse Here, information learned from one router is always sent back to that same router, but it is adverstised with an infinite cost The effect is the same in both cases: a routing loop between two routers will be broken

20 Hold Down Timers Split Horizon break loops between 2 routers, not a circle of >2 routers. Still have count to infinity: slow To improve convergence, use timers Hold-down timer: if do not receive advertisement after the hold-down time, treat it as suspicious. Use route, but do no re-advertise it. Ignore advertisements with higher metrics Flush time: if still no advertisement after flush time, remove route from routing table RIP: advertisements every 30s, hold down 180s, flush 240s

21 Hold Down Timers B has “N1 cost 2 via A” in its table. N1 link breaks E stops advertising route to A: route held down A stops advertising to B: route held down C sends “N1 cost 4” to B. Ignored as higher metric and the route is marked as suspicious

22 Route Poisoning Note: different to Poison Reverse Timers OK, but take time to expire when a route is not advertised Instead, when route is lost, advertise it to neigbours with the infinity value. In previous diagram, E sends “N1 cost infinity” to A A updates its table immediately, passes the infinity advertisement on to its neighbours Much faster than waiting for the timers to expire

23 Flash Updates Distance Vector advertises periodically: RIP is 30s Why wait? Send advertisement as soon a a new route appears or an old one changes Improves convergence time However, causes more network traffic, as new route floods out from router to router Is a problem with links that “flap”: come up, go down, come up

24 Drawbacks of Distance Vector Convergence of Distance Vector is slow for several reasons:  delays in advertisements  inability to tell if information is true or false Workaround like count to infinity, split-horizon Techniques like flash updates help to reduce advertisement delays Techniques like poison reverse, route poisoning help to distinguish false information from correct information Still, no router has full map of the network

25 Link State Routing Protocols Link State routing protocols find the best (lowest metric) path by building a full topology map of the internetwork in each router Link-state routing uses:  link-state packets (LSPs),  a topological database,  Dijkstra's algorithm (SPF) to find the best paths to all networks,  a routing table of paths and interfaces to each network.

26 Link State Packets R1 sends out this to all of its neighbours Connected Link/NetworkLink Cost and State Link to R223, up Link to R318, up Link to R49, up Link to 207.8.8.0/251, up

27 Link State Packets As each router receives this LSP, it passes it on to other neighbour routers LSP floods out to all routers Each router updates their topological database with the information

28 Dijkstra's (SPF) Algorithm When each router's topological database changes, it runs Dijkstra's algorithm to find the best routes to every network Which is the best path from 1 to 6? From 3 to 1?

29 Link State Problems LSP are flooded whenever a link changes. Can cause a lot of extra network traffic. Will cause a lot of CPU work on every router Routers need more RAM to hold the topological database Solutions: break the internetwork up into areas, LSPs are only flooded within an area Elect certain routers to be the only routers to send out LSPs

30 Link State Areas

31 Link State Designated Routers In multiaccess links, one Link State router is elected to send out LSPs: the designated router –A second router is elected the backup designated router Helps to minimise LSP traffic


Download ppt "Routing Concepts Warren Toomey GCIT. Introduction Switches need to know the link address and location of every station. Doesn't scale well, e.g. to several."

Similar presentations


Ads by Google