Routing

1Recap5 min

Learning outcomes

• Explain what a route is and what role the metric plays
• State, for any route, which source it came from and what administrative distance it has
• Read a routing table and predict exactly what happens to a given packet
• Configure static routes, with a next hop or with an exit interface, and know when each works
• Write a default route and a floating backup route
• Compute a summary route and use the Null0 interface

2The role and functions of a router9 min

Architecturally, a router is a specialised computer: it has a processor, memory and input/output devices. The differences from an ordinary computer are considerable, however, and all of them serve the same purpose - to make forwarding decisions as fast as possible.

Besides the basic function, modern routers also do: traffic filtering with ACLs (lecture 7), address translation with NAT and PAT (lecture 9), setting up tunnels, handing out addresses as a DHCP server, and proxy ARP.

3The route and the metric9 min

In networks, "distance" in kilometres says nothing useful. The quantity attached to a route is called the metric and may be any of the following, or a combination:

4Where routes come from8 min

When an unconfigured router starts up, it knows no routes at all. Routes appear from three directions:

5Administrative distance11 min

The metric allows two routes from the same source to be compared. But how do we compare an OSPF route with metric 20 against a RIP route with metric 3?

A second criterion is needed, applied before the metric: the administrative distance (AD), which expresses how much the router trusts the source of the route. The lower it is, the greater the trust.

The complete selection rule

1. The longest prefix always wins. A /26 route beats a /8 route, whatever the source, whatever the metric. This is the supreme criterion and it is not negotiable.
2. At the same prefix, the lower administrative distance wins.
3. From the same source, the lower metric wins.
4. If the metric is equal too, the router keeps both paths and distributes traffic between them - a mechanism called equal-cost load balancing.

The easiest way to remember this table is through the logic behind it: the more first-hand information the router has about a route, the lower the number. An interface of its own is absolute truth (0). A route written by a person is a verifiable statement (1). A link-state protocol, which has the map, is more credible (110) than one that repeats rumours (120).

6The routing table and the decision process13 min

The routing table is the router's map of the network around it. It is organised from the most specific routes (long mask) to the most general, and the process of forwarding a packet has exactly two steps.

1. Receiving the packet. The layer 2 header is decapsulated, the destination IP address is read from the layer 3 header, the TTL is decremented and the checksum recomputed.
2. Finding the route. The table is scanned; for each route, an AND is taken between its mask and the destination address, and the result is compared with the network address of the route. The first match wins. If nothing is found, the packet is discarded and an ICMP destination unreachable is sent.

7The default route7 min

The route 0.0.0.0/0, also called quad-zero, has a mask made up entirely of zeros. An AND with it always gives 0.0.0.0, which always coincides with the network address of the route - so it matches any destination.

Being the most general, it occupies the last position in the table and applies only if nothing else matched. It is the route every small network has towards its Internet provider.

8Static routes11 min

A static route is written by hand by the administrator. The direction can be expressed in two ways, and the choice between them is not a matter of indifference.

! via the next-hop address - always works
R1(config)# ip route 192.168.10.0 255.255.255.0 10.0.0.3

! via the exit interface - only on point-to-point links
R1(config)# ip route 192.168.10.0 255.255.255.0 serial 0/0/0

! both, the most explicit and the fastest form
R1(config)# ip route 192.168.10.0 255.255.255.0 serial 0/0/0 10.0.0.3

! the default route
R1(config)# ip route 0.0.0.0 0.0.0.0 10.0.0.3

A(config)# ip route 140.20.0.0 255.255.240.0 10.0.0.6
C(config)# ip route 172.16.0.0 255.255.0.0 10.0.0.5

The floating route

Sometimes we want a backup path that activates only if the main one disappears. This is achieved by giving the backup route a higher administrative distance.

! the main one - default AD of 1
R1(config)# ip route 172.20.19.0 255.255.255.0 10.0.0.6

! the backup - AD 200, so it does not enter the table while the first is valid
R1(config)# ip route 172.20.19.0 255.255.255.0 10.0.0.2 200

Check with show ip route: the route with AD 200 does not appear in the table as long as the one with AD 1 is valid. Shut down the main interface and look again - there it is. Bring it back up and check that it withdraws.

9The complete journey of a packet11 min

Here we put together everything we have learnt: ARP from lecture 5, switching from lecture 3, and today's routing. It is the exercise that appears, in one form or another, in every examination.

10Route summarisation and the null interface11 min

A router with four routes to four consecutive subnets can replace them all with a single one, if they share a common prefix. The process is called summarisation (or aggregation).

How it is calculated

You write the addresses in binary and see how many bits from the left are common to all of them. That number is the prefix of the summary route.

Choose /26 summarised to /24: four routes become one. Choose /30 summarised to /22 - the case of the point-to-point links in a large network - and you will see why summarisation is not a luxury: 256 lines replaced by a single one, in the table of every router in the network.

The side effect and the null interface

Summarisation has a price. If one of the four subnets - say 192.168.1.128/26 - no longer exists, the summary route 192.168.1.0/24 goes on attracting traffic for it and forwarding it onward, where it will be discarded only at the end of the chain. In the case of a loop, the traffic may even oscillate between routers until the TTL expires.

The solution is a static route to Null0, the null interface: a destination that immediately discards everything it receives. Being more specific than the summary route, it wins and stops the traffic at the source.

R1(config)# ip route 192.168.1.128 255.255.255.192 null0

The equivalent in the Linux world is /dev/null; the idea is the same - a place where things disappear, on request and without error.

11Common mistakes4 min

• "I configured the route, but the ping does not work" In nine cases out of ten, the return route is missing. The packet arrives; the answer has no way back. Routing is always configured in both directions.
• "I compared the metrics of two different protocols" They are not comparable. The administrative distance applies first. Longer prefix → lower AD → lower metric. In that order.
• "I wrote the route with the exit interface, on Ethernet" The router does not know which MAC address to put in the frame, because there are several candidates on the segment. On Ethernet, always write the next hop.
• "I saved the configuration… I think" running-config is in RAM. Without copy running-config startup-config, it is lost on reboot. Save after each block of changes, not at the end.
• "I summarised and now some destinations no longer answer" The summary route also covers subnets that do not exist or are not yours. Add routes to Null0 for the unused portions of the block.

12Summary and glossary5 min

1. A route is the triple (destination, direction, distance). The distance is called the metric and depends on the protocol.
2. Selection proceeds in order: longer prefix → lower AD → lower metric.
3. The table is ordered from specific to general, and the first match wins. The default route is last.
4. On Ethernet, a static route needs a next hop, because otherwise the destination MAC address is missing.
5. IP addresses stay constant along the whole path; MAC addresses are rewritten at every hop; the TTL falls by 1 at every router.
6. Summarisation shrinks the tables; Null0 stops traffic towards the unused portions.

13Self-check questions7 min

14Further reading and bibliography2 min

Static routes are exact and predictable, but they react to no change and do not scale. Lecture 10 introduces the protocols that maintain the table by themselves. Until then, lectures 7 and 9 show what else a router can do with the packets passing through it: filter them and rewrite their addresses.

In laboratory 5 you configure exactly the topology from here, first with static routes and then with OSPF, then cut a cable and time which recovers faster.

• RFC 1812 - requirements for IPv4 routers
• RFC 4632 - CIDR and route aggregation
• Cisco - IP Routing: Protocol-Independent Configuration Guide
• Radia Perlman, Interconnections: Bridges, Routers, Switches and Internetworking Protocols