Genetic Programming and Protocol Construction

1
Genetic Programming and Protocol Construction

• Paul Grace

Supervisor Prof. G. Blair
2
Introduction

• Outline of the project
• Apply genetic programming principles to the
  design of Ensemble protocols and evaluate the
  applicability of this approach.
• Protocol construction
• Use of Ensemble system to build protocols
• Genetic programming
• Machine learning technique for automatically
  generating computer programs
• Motivation
• Remove the need for a programmer/ designer to
  design and implement each protocol in Ensemble.

3
Background - Configurable Protocols

• Ensemble
• Group communication toolkit.
• Creation of protocols by layering of
  micro-protocols.
• Identical interfaces mean it is syntactically
  possible to order micro-protocols in any way.
• Why are configurable protocols needed?
• One size fits all solution creates a large and
  slow protocol stack.
• Create lightweight protocol to meet each
  application requirement e.g. QoS or network
  environment changes.
• Other configurable protocol systems
• STREAMS
• X-Kernel
• DaCapo

4
Background - Genetic Programming (1)

• Theory
• Initial population of programs
• based on Survival of the fittest theory
• Assign fitness to each program and choose the
  fittest for next generation
• Evolve the chosen into a new generation of
  programs using evolution functions such as
  crossover and mutation
• Areas applied to
• General computer program generation
• Assembly language generation
• Electronic circuit design

5
Background - Genetic Programming (2)

• GP toolkits
• Libraries containing the functions needed to
  create the GP.
• Will be used for the implementation of the
  project.
• Number of available toolkits to choose from
• GPC http//www.emk.e-technik.thdarmstadt.de/t
  homasw/gp.html
• GPJPP http//www.pcisys.net/kimk/gpjpp.htm
• .are just two of a number to select from

6
Aims

• Main aim
• To evaluate the application of genetic
  programming to the construction of Ensemble
  protocols.
• Sub goals
• Design an appropriate fitness test and selection
  method.
• Identify the required evolution algorithm. How
  the protocol stacks are represented and evolved
  by the GP.
• Implement a genetic program that automatically
  generates the protocol stack for a sub set of the
  domain of layers.
• Evaluate the performance of genetic programming
  in performing the task.

7
Methodology

• Project split into two phases
• Literature search and technical familiarisation
• Creation of Genetic Program
• Creation of Genetic Program
• incremental experimentation
• expanding the search space of GP by using a
  larger domain of Ensemble micro-protocols

8
Programme of Work
9
Conclusions

• Identified Ensemble as a search space of many
  possible protocols.
• GP can evolve protocols from this search space to
  find the one that meets the requirements of a
  needed protocol.
• Main goal is to implement the GP to do this.
• Project process of literature research
  incremental experimentation
• Time to complete - 18 weeks