Design Patterns from Biology for Distributed Computing

1
Design Patterns from Biology for Distributed
Computing

• Review

2
Overview

• Explore applications inspired from emergence and
  self-organization in biological systems
• Study Design Patterns in biological systems and
  apply in distributed environment
• Inspired Applications Data aggregation, Search
  mechanisms, MANETs

3
Why get inspired from Biological systems

• Similar environment
• Dynamic
• Unreliable
• Large-scale
• Central control is not possible
• Biological systems have nice properties
• Robustness to individual failure
• Adaptivity to change
• Lack of reliance on explicit control
• Emergence of local interactions to global
  functions/properties

4
Design Patterns

• Plain Diffusion
• Inspired from equalization of concentration
• Receive/emit diffusive material
• Replication
• Inspired from epidemic spreading/proliferation
  processes
• Create new material. Eg Virus spreading
• Reaction Diffusion
• Spontaneous diffusion with addition/removal of
  material
• Used in pattern formation / developmental
  processes
• Stigmergy, learning by reinforcements
• Inspired from social systems (insects, ants,
  birds)
• Global co-ordination from properly designed
  local interactions

5
Evaluation variables

• FOM Figure of Merit (human norms to evaluate
  goodness of a system)
• EV Environment variables
• Nice properties -gt INSENSITIVITY
• Insensitivity ((?FOM)/(?EV))-1
• All systems properties are functions of
  insensitivity
• Scalability EVnumber of nodes
• Robustness EVmeasure of damage to system
• Adaptivity EVall other variables

6
Data Aggregation

• What is Aggregation
• Set of functions to provide some global system
  property
• Eg finding extremes, computing sums averages
• Aggregation through diffusion Average, Count,
  Min , Max

7
Evaluation

• Convergence Factor ?i
• ?i 1/(2ve)
• Study
• Scalabiliy Insenstivity
• By varying the size of network
• Robustness to crash failures

8
Study

• Scalabiliy Insenstivity
• By varying the size of network
• Robustness to crash failures
• Effect of sudden death (of 50 of nodes) at
  different cycles
• Removing and substituting certain of nodes
• Robustness to communication failures
• Effect of link breaks
• Effect of lost messages
• Increasing robustness
• Effect of averaging over different iterations

9
Search in overlay networks

• Well known problem
• Diffusion vs k-random walk
• Inspired by Human Immune System
• Proliferation of Antibodies to effectively track
  down antigens
• Algorithms
• Random Walk
• Proliferation
• Restricted Random Walk
• Restricted Proliferation
• Performance
• Network Coverage
• Search efficiency

10
Routing in Mobile Ad-hoc Networks

• AntHocNet inspired from
• Stigmergy
• Learning from re-inforcements
• Characteristics
• Stigmergic routing tables
• Matrix Table
• Goodness function of number of hops and
  end-to-end delay
• Reactive Path Setup
• Reactive Forward Ants (broadcast) to find route
  to destination
• Destination sends Backward Ants to source,
  updates pheromone tables en route
• Choosing a path
• Proactive Path Maintainance Exploration
• Proactive forward ants sent regularly
• Pheromone diffusion messages sent to neighbors
  (similar to DSDV)

11
Discussion

• Scalability of Evaluation Methodology to complex
  systems with many EVs
• Need for a common evaluation methodology to study
  overall performance
• Challenges and differences between biological
  systems and distributed networks
• Gossip based algorithm
• Scalability to different scenarios
• AntHocNET
• Effectiveness of a hybrid strategy?
• Congestion measured through end-to-end delay?