COMS W4995-1

Lecture 5

Dynamic routing protocols I

1. Overview of router architecture2. Overview Dynamic Routing Protocols: Distance Vector Routing 3. Intra-Domain Routing Protocols: RIP

Routing and Forwarding

Forwarding is selecting the next-hop machine for each outgoing packet. Forwarding table, FIB (Forwarding Information Base).

Routing is the process of deciding the path from a source to a destination. Routing table, RIB (Routing Information Base).

Why two tables and not just one?

Routing and Forwarding

Control plane: run routing protocols: (RIP, OSPF, BGP)

Data plane: forwarding packets from incoming to outgoing link

Routing and Forwarding

Select the next-hop router. Find the outgoing interface. Find the MAC address of the next-hop router. In Unix, you specify the IP address of the next-hop router.

Longest-prefix first.

Default routing (implied by longest-prefix rule: default has prefix of length 0).

Routing and Forwarding

Routing functions include: route calculation maintenance of the routing table execution of routing protocols

On commercial routers handled by a single general purpose processor, called route processor

IP forwarding is per-packet processing On high-end commercial routers, IP forwarding is

distributed Most work is done on the interface cards

Router Hardware Components

Hardware components of a router: Network interfaces Switching fabrics Processor with a memory

and CPU

Interface Card

Switching fabric

Interface Card Interface Card

Processor

CPUMemory

PC Router versus commercial router

On a PC router: Switching fabric is the (PCI)

bus Interface cards are NICs (e.g.,

Ethernet cards) All forwarding and routing is

done on central processor

On Commercial routers: Switching fabrics and

interface cards can be sophisticated

Central processor is the route processor (only responsible for control functions)

Interface Card

Switching fabric

Interface Card Interface Card

Processor

CPUMemory

Basic Architectural ComponentsPer-packet processing

I/O Ports I/O Ports

Switching Fabric

Evolution of Router Architectures

Early routers were essentially general purpose computers Today, high-performance routers resemble

supercomputers Exploit parallelism Special hardware components

Until 1980s (1st generation): standard computer Early 1990s (2nd generation): delegate to interfaces Late 1990s (3rd generation): Distributed architecture

Today: Distributed over multiple racks

1st Generation Routers (switching via memory)

This architecture is still used in low end routers

Arriving packets are copied to main memory via direct memory access (DMA)

Switching fabric is a backplane (shared bus)

All IP forwarding functions are performed in the central processor.

Routing cache at processor can accelerate the routing table lookup.

Memory

Shared Bus

DMA

MAC

DMA

MAC

Interface Card

DMA

MAC

Route Processor

Interface Card

Interface Card

CacheCPU

Drawbacks of 1st Generation Routers

Forwarding Performance is limited by memory and CPU

Capacity of shared bus limits the number of interface cards that can be connected

InputPort

OutputPort

Memory

System Bus

SharedBus

InterfaceCards

DMA

MAC

DMA

MAC

DMA

MAC

Route Cache

Memory

Route Cache

Memory

Route Cache

Memory

Route Processor

MemoryCacheCPU

2nd Generation Routers (switching via a shared bus)

Keeps shared bus architecture, but offloads most IP forwarding to interface cards

Interface cards have local route cache and processing elements

Fast path: If routing entry is found in local cache, forward packet directly to outgoing interface

Slow path: If routing table entry is not in cache, packet must be handled by central CPU

slow pathfast path

CPU

Cache

Memory

MAC MAC

Memory

Forwarding Bus(IP headers only)

InterfaceCards

Data Bus

Control Bus

Memory

MAC

Memory

ForwardingEngine

CPU

Cache

Memory

ForwardingEngine Route Processor

CPU

Memory

Another 2nd Generation Architecture

IP forwarding is done by separate components (Forwarding Engines)

Forwarding operations:1. Packet received on interface:

Store the packet in local memory. Extracts IP header and sent to one forwarding engine

2. Forwarding engine does lookup, updates IP header, and sends it back to incoming interface

3. Packet is reconstructed and sent to outgoing interface.

Drawbacks of 2nd Generation Routers

SharedBus

InterfaceCards

DMA

MAC

DMA

MAC

DMA

MAC

Route Cache

Memory

Route Cache

Memory

Route Cache

Memory

Route Processor

MemoryCacheCPU

Bus contentionlimits throughput

3rd Generation Architecture

Switching fabric is an interconnection network (e.g., a crossbar switch)

Distributed architecture: Interface cards operate

independent of each other No centralized processing for IP

forwarding These routers can be scaled to

many hundred interface cards and to aggregate capacity of > 1 Terabit per second

CPU

Memory

RouteProcessor

Memory

RouteProcessing

MAC

SwitchFabric

Interface

SwitchFabric

Memory

RouteProcessing

MAC

SwitchFabric

Interface

Cisco Express Forwarding (distributed mode)

Cisco Express Forwarding Benefits

Scalability & Efficiency Adjacency Tables for local hosts (same network)

Layer 2 switching is faster. The line cards perform the express forwarding between port

adapters, relieving the RSP (Route Switch Processing) of involvement in the switching operation.

Resilience No route cache: several data structures for CEF switching Line Cards maintain an identical copy of the FIB and adjacency

tables.

More at Cisco on-line documentation

Slotted Chassis

Large routers are built as a slotted chassis Interface cards are inserted in the slots Route processor is also inserted as a slot

This simplifies repairs and upgrades of components

Dynamic Routing Protocols

Distance Vector Routing

Routing Protocols

Recall: There are two parts to routing IP packets:1. How to pass a packet from an input interface to the output interface of a router (packet forwarding) ? 2. How to find and setup a route ?

We already discussed the packet forwarding part Longest prefix match

There are two approaches for calculating the routing tables: Static Routing (We modify manually the Routes) Dynamic Routing: Routes are calculated by a routing protocol

Routing protocols vs routing algorithms

Routing protocols establish routing tables at routers.

A routing protocol specifies What messages are sent between routers Under what conditions the messages are sent How messages are processed to compute routing tables

At the heart of any routing protocol is a routing algorithm that determines the path from a source to a destination

Overview Routing Protocols

Routing information protocol (RIP) Distance vectorInterior Gateway routing protocol (IGRP, Cisco proprietary)

Distance vector

Open shortest path first (OSPF) Link stateIntermediate System-to-Intermediate System (IS-IS

Link state

Border gateway protocol (BGP) Path vector

Routing protocol Routing Algorithm

Intra-domain routing versus inter-domain routing

Recall Internet is a network of networks.

Administrative autonomy internet = network of networks each network admin may want to control routing in its own

network

Scale: with 200 million destinations: can’t store all destinations’s in routing tables! routing table exchange would swamp links

Autonomous systems

aggregate routers into regions, “autonomous systems” (AS) or domain routers in the same AS run the same routing protocol

“intra-AS” or intra-domain routing protocol routers in different AS can run different intra-AS routing protocol

Ethernet

Router

Ethernet

Ethernet

RouterRouter

Ethernet

Ethernet

EthernetRouterRouter

Router

AutonomousSystem 2

AutonomousSystem 1

Autonomous Systems An autonomous system is a region of the Internet that is

administered by a single entity.

Examples of autonomous regions are: UCI’s campus network MCI’s backbone network Regional Internet Service Provider

Routing is done differently within an autonomous system (intradomain routing) and between autonomous system (interdomain routing).

RIP, OSPF, IGRP, and IS-IS are intra-domain routing protocols.

BGP is the only inter-domain routing protocol.

Distance Vector Routing Variations of Bellman-Ford algorithm.

Each router starts by knowing: Prefixes of its attached networks (“zero” distance). Its next hop routers (how to find them?)

Each router advertises only to its neighbors: All prefixes it knows about. Its distance from them.

Each router learns: All prefixes its neighbors know about. Their distance from them.

Each router figures out, for each destination prefix: The “distance” (how far away it is). The “vector” (the next hop router).

Distance Vector Routing Properties DV Computes the Shortest Path

“Routing by rumor” Each router believes what its neighbors tell it.

In steady-state, each router has the “shortest” (smallest metric) path to the destination.

Convergence time is (on the average) proportional to the diameter of the network.

Any link change affects the entire network.

Distance vector algorithm

A decentralized algorithm A router knows physically-connected neighbors and link costs to

neighbors A router does not have a global view of the network

Path computation is iterative and mutually dependent. A router sends its known distances to each destination (distance

vector) to its neighbors. A router updates the distance to a destination from all its

neighbors’ distance vectors A router sends its updated distance vector to its neighbors. The process repeats until all routers’ distance vectors do not

change (this condition is called convergence).

Bellman-Ford Algorithm

Bellman-Ford EquationDefinedx(y) := cost of the least-cost path from x to y

Then dx(y) = minv{c(x,v) + dv(y) }, where min is taken over

all neighbors of node x

Distance vector algorithm: initialization

Let Dx(y) be the estimate of least cost from x to y

Initialization: Each node x knows the cost to each neighbor: c(x,v). For each

neighbor v of x, Dx(v) = c(x,v) Dx(y) to other nodes are initialized as infinity.

Each node x maintains a distance vector (DV): Dx = [Dx(y): y 2 N ]

Distance vector algorithm: updates

Each node x sends its distance vector to its neighbors, either periodically, or triggered by a change in its DV.

When a node x receives a new DV estimate from a neighbor v, it updates its own DV using B-F equation: If c(x,v) + Dv(y) < Dx(y) then

Dx(y) = c(x,v) + Dv(y) Sets the next hop to reach the destination y to the neighbor v Notify neighbors of the change

The estimate Dx(y) will converge to the actual least cost dx(y)

Distance vector algorithm: an example

1 1

1

1 1

1

1

1

Time = 0

Distance vector algorithm: an example

Time = 1

Distance vector algorithm: an example

Time = 2 (End)

How to map the abstract graph to the physical network

Nodes (e.g., v, w, n) are routers, identified by IP addresses, e.g. 10.0.0.1 Nodes are connected by either a directed link or a broadcast link (Ethernet) Destinations are IP networks, represented by the network prefixes, e.g.,

10.0.0.0/16 Net(v,n) is the network directly connected to router v and n.

Costs (e.g. c(v,n)) are associated with network interfaces. Router1(config)# router rip Router1(config-router)# offset-list 0 out 10 Ethernet0/0 Router1(config-router)# offset-list 0 out 10 Ethernet0/1

n

vw

Net

Net(v,w)

Net(v,n)

c(v,w)

c(v,n)

Distance vector routing protocol: Routing Table

Dest

n

vw

D(v,Net)n

costvia(next hop)

Net

RoutingTable of node v

Net

Net(v,w)c(v,w)

Net(v,n)c(v,n)

Net(v,w): Network address of the network between v and w

c(v,w): cost to transmit on the interface to network Net(v,w)

D(v,net) is v’s cost to Net

Distance vector routing protocol: Messages

Dest

D (v,Net)n

costvia(next hop)

Net

RoutingTable of node v

• Nodes send messages to their neighbors which contain distance vectors• A message has the format: [Net , D(v,Net)] means“My cost to go to Net is D (v,Net)”

v n[Net , D(v,Net)]

Distance vector routing algorithm: Sending Updates

Dest

D (v,Net 2)n

costvia(next hop)

Net 2

RoutingTable of node v

D (v,Net 1)mNet 1

D(v,Net N)wNet N

Periodically, each node v sends the content of its routing table to its neighbors:

n

v wm

[Net N,D(v,Net N)]

[Net 1,D(v,Net 1)]

[Net N,D(v,Net N)]

[Net 1,D(v,Net 1)]

[Net N,D(v,Net N)]

[Net 1,D(v,Net 1)]

Initiating Routing Table I

Destc (v,w)

Net(v,w)

0m

costvia(next hop)

Net(v,m)

RoutingTablec(v,m)

Net(v,m)

c(v,n)Net(v,n) 0wNet(v,w)

0nNet(v,n)n

v wm

Suppose a new node v becomes active. The cost to access directly connected networks is zero:

D (v, Net(v,m)) = 0 D (v, Net(v,w)) = 0 D (v, Net(v,n)) = 0

Initiating Routing Table II

Dest

0m

costvia(next hop)

Net(v,m)

RoutingTable

0wNet(v,w)

0nNet(v,n)

Node v sends the routing table entry to all its neighbors:

n

v wm[w,0][n,0 ] [n,0 ]

[m,0]

[m,0][w,0]

n

v wm[Net(v,w),0][Net(v,n),0] [Net(v,n),0]

[Net(v,m),0]

[Net(v,w),0][Net(v,m),0]

n

v wm[Net(v,w),0][Net(v,n),0] [Net(v,n),0]

[Net(v,m),0]

[Net(v,w),0][Net(v,m),0]

n

v wm

[Net N,D(n,Net N)]

[Net 1,D(n,Net 1)]

[Net N,D(m,Net N)]

[Net 1,D(m,Net 1)]

[Net N,D(w,Net N)]

[Net 1,D(w,Net 1)]

Initiating Routing Table III

Node v receives the routing tables from other nodes and builds up its routing table

The Count-to-Infinity Problem

X

What happens on a link failure?

A: 1,AB:0C:1,C

A: 0B:1,BC:2,B

A: 2,BB:1,BC:0

A: 1,AB:0C:-

A: 1,AB:0C:3,A

A: 0B:1,BC:4,B

A: 1,AB:0C:5,A

A: 0B:1,BC:6,B

Count-to-Infinity

The reason for the count-to-infinity problem is that each node only has a “next-hop-view”

For example, in the first step, A did not realize that its route (with cost 2) to C went through node B

How can the Count-to-Infinity problem be solved?

Count-to-Infinity

The reason for the count-to-infinity problem is that each node only has a “next-hop-view”

For example, in the first step, A did not realize that its route (with cost 2) to C went through node B

How can the Count-to-Infinity problem be solved?

Solution 1: Always advertise the entire path in an update message to avoid loops (Path vectors) BGP uses this solution

Count-to-Infinity

The reason for the count-to-infinity problem is that each node only has a “next-hop-view”

For example, in the first step, A did not realize that its route (with cost 2) to C went through node B

How can the Count-to-Infinity problem be solved?

Solution 2: Never advertise the cost to a neighbor if this neighbor is the next hop on the current path (Split Horizon)

Example: A would not send the first routing update to B, since B is the next hop on A’s current route to C

Split Horizon does not solve count-to-infinity in all cases! You can produce the count-to-infinity problem in Lab 4.

Characteristics of D.V. Routing Protocols

Periodic Updates: Updates to the routing tables are sent at the end of a certain time period. A typical value is 30 seconds.

Triggered Updates: If a metric changes on a link, a router immediately sends out an update without waiting for the end of the update period.

Full Routing Table Update: Most distance vector routing protocol send their neighbors the entire routing table (not only entries which change).

Route invalidation timers: Routing table entries are invalid if they are not refreshed. A typical value is to invalidate an entry if no update is received after 3-6 update periods.

Inter-domain Routing Protocols: RIP

RIP - Routing Information Protocol

A simple intradomain protocol

Straightforward implementation of Distance Vector Routing

Each router advertises its distance vector every 30 seconds (or whenever its routing table changes) to all of its neighbors

RIP always uses 1 as link metric

Maximum hop count is 15, with “16” equal to “”

Routes are timeout (set to 16) after 3 minutes if they are not updated

RIP - History

Late 1960s : Distance Vector protocols were used in the ARPANET

Mid-1970s: XNS (Xerox Network system) routing protocol is the ancestor of RIP in IP (and Novell’s IPX RIP and Apple’s routing protocol)

1982 Release of routed for BSD Unix 1988 RIPv1 (RFC 1058)

- classful routing 1993 RIPv2 (RFC 1388)

- adds subnet masks with each route entry - allows classless routing

1998 Current version of RIPv2 (RFC 2453)

RIPv1 Packet Format

IP header UDP header RIP Message

Command Version Set to 00...0

32-bit address

Unused (Set to 00...0)

address family Set to 00.00

Unused (Set to 00...0)

metric (1-16)

one

rout

e en

try(2

0 by

tes)

Up to 24 more routes (each 20 bytes)

32 bits

One RIP message can have up to 25 route entries

1: request2: response

2: for IP

Address of destination

Cost (measured in hops)

1: RIPv1

RIPv2

RIPv2 is an extends RIPv1: Subnet masks are carried in the route information Authentication of routing messages Route information carries next-hop address Uses IP multicasting

Extensions of RIPv2 are carried in unused fields of RIPv1 messages

RIPv2 Packet Format

IP header UDP header RIP Message

Command Version Set to 00...0

32-bit address

Unused (Set to 00...0)

address family Set to 00.00

Unused (Set to 00...0)

metric (1-16) one

rout

e en

try(2

0 by

tes)

Up to 24 more routes (each 20 bytes)

32 bits

One RIP message can have up to 25 route entries

1: request2: response

2: for IP

Address of destination

Cost (measured in hops)

2: RIPv2

RIPv2 Packet Format

IP header UDP header RIPv2 Message

Command Version Set to 00.00

IP address

Subnet Mask

address family route tag

Next-Hop IP address

metric (1-16) one

rout

e en

try(2

0 by

tes)

Up to 24 more routes (each 20 bytes)

32 bits

Used to provide a method of separating "internal" RIP routes (routes for networks within the RIP routing domain) from "external" RIP routes

Identifies a better next-hop address on the same subnet than the advertising router, if one exists (otherwise 0….0)

2: RIPv2

Subnet mask for IP address

RIP Messages

This is the operation of RIP in routed. Dedicated port for RIP is UDP port 520.

Two types of messages: Request messages

used to ask neighboring nodes for an update Response messages

contains an update

Routing with RIP

Initialization: Send a request packet (command = 1, address family=0..0) on all interfaces:

RIPv1 uses broadcast if possible, RIPv2 uses multicast address 224.0.0.9, if possible

requesting routing tables from neighboring routers

Request received: Routers that receive above request send their entire routing table

Response received: Update the routing table

Regular routing updates: Every 30 seconds, send all or part of the routing tables to every neighbor in an response message

Triggered Updates: Whenever the metric for a route change, send entire routing table.

RIP Security

Issue: Sending bogus routing updates to a router RIPv1: No protection RIPv2: Simple authentication scheme

IP header UDP header RIPv2 Message

Command Version Set to 00.00

Password (Bytes 0 - 3)

Password (Bytes 4 - 7)

0xffff Authentication Type

Password (Bytes 8- 11)

Password (Bytes 12 - 15) Auth

etic

atio

nUp to 24 more routes (each 20 bytes)

32 bits

2: plaintext password

RIP Problems

RIP takes a long time to stabilize Even for a small network, it takes several minutes until the

routing tables have settled after a change

RIP has all the problems of distance vector algorithms, e.g., count-to-Infinity RIP uses split horizon to avoid count-to-infinity

The maximum path in RIP is 15 hops