A survey of Routing Algorithms Ignasi Paredes Oliva [email_address] [email_address]

Outline Introduction Routing strategies Routing algorithms Shortest Path Routing Flooding Flow-based Routing Distance Vector Routing Link State Routing Deflection Routing Broadcast Routing Multicast Routing Routing for Mobile Hosts Onion Routing Hierarchical Routing Final considerations

Introduction

Routing strategies

Routing algorithms

Shortest Path Routing

Flooding

Flow-based Routing

Distance Vector Routing

Link State Routing

Deflection Routing

Broadcast Routing

Multicast Routing

Routing for Mobile Hosts

Onion Routing

Hierarchical Routing

Final considerations

Outline Introduction Routing strategies Routing algorithms Final considerations

Introduction

Routing strategies

Routing algorithms

Final considerations

Introduction Routing: Driving packets from source node to destination node in a network Desirable properties Correctness Simplicity Robustness Stability Fairness Optimality

Routing: Driving packets from source node to destination node in a network

Desirable properties

Correctness

Simplicity

Robustness

Stability

Fairness

Optimality

Robustness It’s expected that a network run continuously The algorithm should cope with changes in both topology and traffic when something crashes Fairness and Optimality They’re often contradictory  trade-off needed Global efficiency Fairness to individual connections Introduction (II)

Robustness

It’s expected that a network run continuously

The algorithm should cope with changes in both topology and traffic when something crashes

Fairness and Optimality

They’re often contradictory  trade-off needed

Global efficiency

Fairness to individual connections

Outline Introduction Routing strategies Routing algorithms Final considerations

Introduction

Routing strategies

Routing algorithms

Final considerations

Routing strategies Non-adaptative (static routing) The state of the network doesn’t matter The route to use is computed before the network is booted Adaptative (dynamic routing) Routing decisions are taken according to the topology and the state of the network Alternative Routing and Multipath Routing There is more than one static route to each destination The same route is not always taken (it changes each time) AR  random decision MR  according to network state

Non-adaptative (static routing)

The state of the network doesn’t matter

The route to use is computed before the network is booted

Adaptative (dynamic routing)

Routing decisions are taken according to the topology and the state of the network

Alternative Routing and Multipath Routing

There is more than one static route to each destination

The same route is not always taken (it changes each time)

AR  random decision

MR  according to network state

Routing strategies (II) Distributed Each node receives some information about the network from its adjacent nodes This information is used to determine the way each router forwards its traffic Centralized There is one central node for routing intelligence ( master ) A node without an associated master ( slave ) is unable to route Each slave node needs a master node Less resilient master failure  all its slaves are useless until they find a new master node

Distributed

Each node receives some information about the network from its adjacent nodes

This information is used to determine the way each router forwards its traffic

Centralized

There is one central node for routing intelligence ( master )

A node without an associated master ( slave ) is unable to route

Each slave node needs a master node

Less resilient

master failure  all its slaves are useless until they find a new master node

Outline Introduction Routing strategies Routing algorithms Final considerations

Introduction

Routing strategies

Routing algorithms

Final considerations

Shortest Path Routing Static routing Only topology information is used A graph is build. Each router is a node and each link is represented as a weighted arc. Find a route between routers ≡ Find the shortest path in the graph between two nodes  Dijkstra Algorithm

Static routing

Only topology information is used

A graph is build. Each router is a node and each link is represented as a weighted arc.

Find a route between routers ≡ Find the shortest path in the graph between two nodes  Dijkstra Algorithm

Shortest Path Routing (II) First 5 steps computing shortest path from A to D Source: [1]

Flooding Static routing Only topology information used Every incoming packet is sent out on every outgoing link except the one it arrived on. Infinite number of duplicate packets!  some control needed Hop counter When it’s equal to 0 the packet is discarded Maintain a list of already flooded packets If an incoming packet is in the list it’s not sent

Static routing

Only topology information used

Every incoming packet is sent out on every outgoing link except the one it arrived on.

Infinite number of duplicate packets!  some control needed

Hop counter

When it’s equal to 0 the packet is discarded

Maintain a list of already flooded packets

If an incoming packet is in the list it’s not sent

Flow-based Routing Static routing Both topology and load are taken into account Sometimes the mean data flow between a certain pair of nodes is stable and predictable Compute mean delay for the entire network Requirements Known topology Traffic matrix Capacity matrix A routing algorithm Goal: find the routing algorithm that produces the minimum average delay for the subnet

Static routing

Both topology and load are taken into account

Sometimes the mean data flow between a certain pair of nodes is stable and predictable

Compute mean delay for the entire network

Requirements

Known topology

Traffic matrix

Capacity matrix

A routing algorithm

Goal: find the routing algorithm that produces the minimum average delay for the subnet

Flow-based Routing (II) A network with line capacities in Kbps The traffic in packets/s and the routing matrix Source: [1] Source: [1]

Flow-based Routing (III) If the requirements are satisfied it’s easy to calculate the total traffic in each line: λ i Mean delay: Queuing theory: T i = ( µ i C i - λ i ) -1 Line AB Assume mean packet size 800b Link capacity: µ AB C AB = 1pkt/800b x 20kbps = 25pkts/s Actual traffic: λ AB = 14pkts/s  delay = T AB = 91ms Analysis of the previous slide subnet The mean delay time for the entire subnet is 86 ms Source: [1]

If the requirements are satisfied it’s easy to calculate the total traffic in each line: λ i

Mean delay:

Queuing theory: T i = ( µ i C i - λ i ) -1

Line AB

Assume mean packet size 800b

Link capacity: µ AB C AB = 1pkt/800b x 20kbps = 25pkts/s

Actual traffic: λ AB = 14pkts/s  delay = T AB = 91ms

Which is the added value of this algorithm? It uses the load to make routing decisions Sometimes it’s better to use a longer path instead of the shortest one (when this last one holds a huge amount of traffic) Compare different routing algorithms and choose the best one according to your network characteristics Flow-based Routing (IV)

Which is the added value of this algorithm?

It uses the load to make routing decisions

Sometimes it’s better to use a longer path instead of the shortest one (when this last one holds a huge amount of traffic)

Compare different routing algorithms and choose the best one according to your network characteristics

Distance Vector Routing Dynamic routing Each router has a table with the best known “distances” to every destination and which line to use to get there Different metrics are used as “distance” #hops delay in ms … This vector is updated using information messages from the neighbors

Dynamic routing

Each router has a table with the best known “distances” to every destination and which line to use to get there

Different metrics are used as “distance”

#hops

delay in ms

…

This vector is updated using information messages from the neighbors

Distance Vector Routing (II) Example (the used metric is delay but other metrics are also possible) J neighbors are A, I, H and K J receives the routing tables of its neighbors J computes its new route to G Via router A: 8 + 18 =26 ms Via router I: 10 + 31 =41 ms Via router H: 12 + 6 =18 ms Via router K: 6 + 31 =37 ms The minimum delay is 18, so J updates its routing table with: Delay=18ms, Output router: H Source: [1] Source: [1]

Example (the used metric is delay but other metrics are also possible)

J neighbors are A, I, H and K

J receives the routing tables of its neighbors

J computes its new route to G

Via router A: 8 + 18 =26 ms

Via router I: 10 + 31 =41 ms

Via router H: 12 + 6 =18 ms

Via router K: 6 + 31 =37 ms

The minimum delay is 18, so J updates its routing table with:

Delay=18ms, Output router: H

Link State Routing Dynamic routing Five parts Discover its neighbors and learn their network addresses Measure the delay or cost to each of its neighbors Construct a packet telling all it has just learned Send this packet to all other routers Compute the shortest path to every other router

Dynamic routing

Five parts

Discover its neighbors and learn their network addresses

Measure the delay or cost to each of its neighbors

Construct a packet telling all it has just learned

Send this packet to all other routers

Compute the shortest path to every other router

Link State Routing (II) Send special HELLO packets and wait for a reply of each router telling who it is (names must be globally unique) Send special ECHO packet and wait for response  delay = RTT/2 Packet=sender identifier+#seq+age+list of delays to each neighbor When? Periodically / When some significant event occurs A subnet The link state packets for this subnet Source: [1] Source: [1]

Send special HELLO packets and wait for a reply of each router telling who it is (names must be globally unique)

Send special ECHO packet and wait for response  delay = RTT/2

Packet=sender identifier+#seq+age+list of delays to each neighbor

When? Periodically / When some significant event occurs

Link State Routing (III) Fundamental idea: distribute packet using flooding There is a control for each (source, #seq) pair If new  packet is flooded If duplicated or #seq is lower than the last seen  packet is discarded Some troubles Sequence numbers wrap around  Solution: 32bits #seq Router crash, #seq corrupted Solution: age field  It’s decremented once per second If age = 0  information from that router is discarded Also decremented at each hop ( ≡ TTL) Run Dijkstra algorithm to find the shortest path to all destinations

Fundamental idea: distribute packet using flooding

There is a control for each (source, #seq) pair

If new  packet is flooded

If duplicated or #seq is lower than the last seen  packet is discarded

Some troubles

Sequence numbers wrap around  Solution: 32bits #seq

Router crash, #seq corrupted

Solution: age field  It’s decremented once per second

If age = 0  information from that router is discarded

Also decremented at each hop ( ≡ TTL)

Run Dijkstra algorithm to find the shortest path to all destinations

Deflection Routing Also known as Hot-Potato routing More than one packet want to leave along the same output link at the same time (contention) Only one packet is sent through that link The other ones are sent along other available links The network is used as a memory No memory needed in the network nodes! Packets travel through the network while they can’t be send along the correct outgoing line No QoS

Also known as Hot-Potato routing

More than one packet want to leave along the same output link at the same time (contention)

Only one packet is sent through that link

The other ones are sent along other available links

The network is used as a memory

No memory needed in the network nodes!

Packets travel through the network while they can’t be send along the correct outgoing line

No QoS

Deflection Routing (II) Cold-Potato routing Opposite to hot-potato Packets are kept (in the “net”) until the last possible minute They are delivered when they get as close as possible to the end-user Good end-to-end QoS for the ISP customers ISPs don’t want to waste bandwidth on their backbone, so they choose hot-potato

Cold-Potato routing

Opposite to hot-potato

Packets are kept (in the “net”) until the last possible minute

They are delivered when they get as close as possible to the end-user

Good end-to-end QoS for the ISP customers

ISPs don’t want to waste bandwidth on their backbone, so they choose hot-potato

Deflection Routing (II) Private network (T1) Router A Router B ISP’s backbone (T3) Router C Router D Cold-Potato is better for the PN T3 is faster than T1 Hot-Potato is better for the ISP LANB can be reached via router A  there is no need to waste bandwidth between router C and router D Source: [5] Hot Potato Routing vs. Cold Potato Routing

Private network (T1)

Router A

Router B

ISP’s backbone (T3)

Router C

Router D

Cold-Potato is better for the PN

T3 is faster than T1

Hot-Potato is better for the ISP

LANB can be reached via router A  there is no need to waste bandwidth between router C and router D

Broadcast Routing Send messages to all or many other hosts Send a packet to each destination / Flooding Wasteful of bandwidth Multi-destination routing The packet includes the list of destinations Sink tree based The root is the node initiating the broadcast Each router copies the packet only for the output lines corresponding with that tree Reverse path forwarding The router checks if the packet arrived on the line used for sending packets TO the source of the broadcast If it matches the packet is sent for all the other output links If not (the packet doesn’t arrive via the preferred path)  discard packet Routers don’t need to know nothing about the ST

Send messages to all or many other hosts

Send a packet to each destination / Flooding

Wasteful of bandwidth

Multi-destination routing

The packet includes the list of destinations

Sink tree based

The root is the node initiating the broadcast

Each router copies the packet only for the output lines corresponding with that tree

Reverse path forwarding

The router checks if the packet arrived on the line used for sending packets TO the source of the broadcast

If it matches the packet is sent for all the other output links

If not (the packet doesn’t arrive via the preferred path)  discard packet

Routers don’t need to know nothing about the ST

Broadcast Routing (II) Example – sink tree and reverse path forwarding Subnet Spanning tree rooted on “I” RPF Source: [1]

Example – sink tree and reverse path forwarding

Multicast Routing Send messages to many nodes Keeping groups information in the router Each router has to know which of their hosts belong to which groups [hosts  router] or [router  host] router learns router  router: info propagates Each router computes a spanning tree (ST) covering all routers in the subnet Prune the ST according to each multicast group It scales poorly for large networks N groups and M members in each one High memory needed to store the N x M trees

Send messages to many nodes

Keeping groups information in the router

Each router has to know which of their hosts belong to which groups

[hosts  router] or [router  host] router learns

router  router: info propagates

Each router computes a spanning tree (ST) covering all routers in the subnet

Prune the ST according to each multicast group

It scales poorly for large networks

N groups and M members in each one

High memory needed to store the N x M trees

Multicast Routing (II) Example: ST rooted on A Subnet ST for group 1 ST for group 2 A Source: [1]

Example:

Routing for Mobile Hosts Motivation: access the Internet wherever in the world we may be New complication: where is the mobile node (MN)? MIP scheme Home Agent (HA), Home Address (at home) and Foreign Agent (when you are away) Step 1: Tunneling to send packets to the correct location Step 2: routing directly from CN to MN ’s current location Example: MN moves from B to C. Then, A wants to send information to the MN Source: [1] A B C

Motivation: access the Internet wherever in the world we may be

New complication: where is the mobile node (MN)?

MIP scheme

Home Agent (HA), Home Address (at home) and Foreign Agent (when you are away)

Step 1: Tunneling to send packets to the correct location

Step 2: routing directly from CN to MN ’s current location

Onion Routing Anonymous communication over a network “ Routing onions“ concept Routing information is encoded in a set of encrypted layers Main goals Protect the privacy of the sender and recipient of a message Provide protection for message content How does it work? messages travel from source to destination via a sequence of proxies ("onion routers"), which re-route messages in an unpredictable path Source: [6]

Anonymous communication over a network

“ Routing onions“ concept

Routing information is encoded in a set of encrypted layers

Main goals

Protect the privacy of the sender and recipient of a message

Provide protection for message content

How does it work?

messages travel from source to destination via a sequence of proxies ("onion routers"), which re-route messages in an unpredictable path

Hierarchical Routing At some point is not feasible to continue adding nodes to a network Split the network into various “subnetworks” Routers are divided into regions Each router only knows how to route packets to destinations within its own region The hierarchy deepness varies It depends of the net size Regions into clusters, clusters into zones, zones into groups …

At some point is not feasible to continue adding nodes to a network

Split the network into various “subnetworks”

Routers are divided into regions

Each router only knows how to route packets to destinations within its own region

The hierarchy deepness varies

It depends of the net size

Regions into clusters, clusters into zones, zones into groups …

Hierarchical Routing (II) Example Net divided into 5 regions From 17 to 7 entries! But gains are not free… Path length is increased Source: [1] The best route for 1A - 5C traffic is not via region 3. That route is better for most destinations in region 5 Full routing table for router 1A Hierarchical routing table for router 1A

Example

From 17 to 7 entries!

But gains are not free…

Path length is increased

Outline Introduction Routing strategies Routing algorithms Final considerations

Introduction

Routing strategies

Routing algorithms

Final considerations

Final considerations Virtual Circuit vs. Datagram Connection oriented (like ATM and MPLS) Each packet contains a short VC number A route is established at the beginning: all packets for that VC follow it (the route is already computed) If some router crashes all VCs using it are terminated Each VC needs some space in each of the nodes that it uses Connectionless (like IP) Each packet contains the full source and destination @’s Each incoming packet requires a routing decision Failures don’t matter Not state information needed Hard re-routing Easy re-routing

Virtual Circuit vs. Datagram

Connection oriented (like ATM and MPLS)

Each packet contains a short VC number

A route is established at the beginning: all packets for that VC follow it (the route is already computed)

If some router crashes all VCs using it are terminated

Each VC needs some space in each of the nodes that it uses

Connectionless (like IP)

Each packet contains the full source and destination @’s

Each incoming packet requires a routing decision

Failures don’t matter

Not state information needed

Ordinary routing vs. Failure recovery routing Static algorithms Problems with failures The algorithm must be restarted to adapt new topology Dynamic algorithms Failures are not a problem Routes are automatically reconfigured taking changes into account Final considerations (II)

Ordinary routing vs. Failure recovery routing

Static algorithms

Problems with failures

The algorithm must be restarted to adapt new topology

Dynamic algorithms

Failures are not a problem

Routes are automatically reconfigured taking changes into account

Final considerations (III) Internet Hierarchical routing ISPs use OSPF and BGP OSPF (Open Shortest Path First) Used in each AS ( ≡ ISP ) Network is divided into areas Uses link-state in the individual areas that make up the hierarchy BGP (Border Gateway Protocol) Network reachability among autonomous systems (ASs)

Internet

Hierarchical routing

ISPs use OSPF and BGP

OSPF (Open Shortest Path First)

Used in each AS ( ≡ ISP )

Network is divided into areas

Uses link-state in the individual areas that make up the hierarchy

BGP (Border Gateway Protocol)

Network reachability among autonomous systems (ASs)

References [1] Andrew S. Tanenbaum. Computer Networks. 3rd ed. New Jersey: Prentice Hall International Editions; 1996. [2] http://sarwiki.informatik.hu-berlin.de/RoutingPrinciples [3] http://www.cs.rpi.edu/~bushl2/project_web/page4.html [4] http://en.wikipedia.org/wiki/Hot-potato_routing [5] http://www.bgpbook.com/archpolicyenter.html [6] http://en.wikipedia.org/wiki/Onion_routing

[1] Andrew S. Tanenbaum. Computer Networks. 3rd ed. New Jersey: Prentice Hall International Editions; 1996.

[2] http://sarwiki.informatik.hu-berlin.de/RoutingPrinciples

[3] http://www.cs.rpi.edu/~bushl2/project_web/page4.html

[4] http://en.wikipedia.org/wiki/Hot-potato_routing

[5] http://www.bgpbook.com/archpolicyenter.html

[6] http://en.wikipedia.org/wiki/Onion_routing