An Intelligent Network Routing Algorithm by a Genetic Algorithm

1
An Intelligent Network Routing Algorithm by a
Genetic Algorithm

• Masaharu Munetomo, Yoshiaki Takai, and Yoshiharu
  Sato
• Hokkaido University, JAPAN.

2
In this paper, we propose..

• An adaptive routing algorithm which employs
  genetic operators to realize an intelligent
  routing which directly observes communication
  latency of the routes.
• Path genetic operators for the routing algorithm
  which generates alternative routes based on the
  network topology.

3
Routing Algorithms in the Internet

• Each node forwards communication packets based on
  its Routing Table.
• Routing Algorithms generate routing tables based
  on network topology.
• Two major categories of routing algorithms

• Interior Gateway Protocols (IGP)
• Exterior Gateway Protocols (EGP)

4
Interior Gateway Protocol (IGP)

• Routing protocols inside an autonomous system
  (AS) such as a Local Area Network
• We have two major protocols for the IGPs commonly
  used in the Internet

• Routing Information Protocols (RIP)
• Shortest Path First Protocols (SPF)
• or Open SPF (OSPF)

5
Exterior Gateway Protocol (EGP)

• Routing protocols outside an AS which exchanges
  routing information among ASs.
• Recently, BGP (Border Gateway Protocols) become
  popular in the Internet.
• The BGP4 employs a source routing approach which
  determines all the nodes along a route in the
  source node instead deciding only its next hop.

6
Routing Information Protocol (RIP)

• A distributed algorithm
• Each node broadcasts its routing table.
• Each node recalculates distances in the routing
  table on receiving a routing table from its
  neighbors.

Broadcast
7
Shortest Path First protocol (SPF)

• Each node broadcasts its link status.
• Each node stores network topology generated from
  the received link status information and
  calculates shortest paths by using the Dijkstras
  Shortest Path First Algorithm.
• The algorithm can reduce communication overhead
  by broadcasting only link status not all the
  routing tables.

8
Problems of the RIP and the SPF

• Not scalable they increase their communication
  overhead in larger networks.
• Not efficient when they need to collect load
  status of links repeatedly to consider delay
  along a route to be minimized.

Communication Overhead
(n of nodes in the network)
9
Genetic-Based Routing (GBR)

• Employing source routing and only maintain a set
  of alternative routes frequently used in
  communication.
• Alternative routes are generated by Path Genetic
  Operators we propose.
• Observing communication latency for the limited
  number of routes to greatly reduce communication
  overhead for the routing.

10
Overview of the GBR
11
Path Genetic Algorithm (pGA)

• Encoding paths(routes) by listing up node IDs,
  for example, (0 12 5 8 2 9).
• We have two path genetic operators
• - Path Mutation
• - Path Crossover
• Selection is performed by deleting routes not
  frequently used in the routing table.

12
Path Mutation
1. We select a node (nm) from the original
route. 2. Another node (nm) is selected from
neighbor of nm. 3. Connecting source to nm and
nm to destination.
13
Path Crossover
--- Exchanges sub-routes among a pair of routes.
14
Fitness evaluation and Selection

• Each node periodically sends delay query packets
  to observe communication latency along a route.
• Fitness value is calculated from the delay

di delay of route i

• Selection is invoked when routing table is
  overflowed.

15
Execution flow of the GBR

• 1. When we need to send a packet, we select a
  route randomly according to fitness value of
  routes (roulette wheel selection).
• 2. After sending a specified number of routes, we
  send delay query packet to evaluate fitness.
• 3. After a specified number of delay query, we
  apply path genetic operators to generate
  alternative routes in the routing table.
• 4. If the number of routes exceeds a limit, we
  perform a selection by deleting routes with
  maximum delay.

16
Simulation Experiments

• Using a network simulator written in C.
• Sample network is taken from Japanese
  geographical info.

• Simulation time is 3000s.
• Genetic operators are
• invoked at every 30
• evaluation of delay.

17
Mean arrival time of packets

• The GBR achieves much smaller mean arrival time
  of communication packets a than those of RIP, SPF
  and an adaptive SPF.
• An adaptive SPF which directly observes
  communication latency of links is not efficient
  in lightly-loaded networks.

18
Load status of Links
GBR
SPF
RIP

• Thickness of a link stands for its mean queue
  length.
• GBR achieves much less overhead of links,
  especially
• on the link 11 ltgt 13 ltgt 19.

19
Conclusions

• Path Genetic Algorithm (pGA) we propose creates
  alternative routes in routing tables.
• A genetic based routing (GBR) algorithm can
  effectively forward communication packets, which
  leads to smaller arrival time.
• Load balancing among links is realized by the GBR
  algorithm.