Complexity in large scale distributed communication networks

1
Complexity in large scale distributed
communication networks

• David Saad
• Aston University

2
Agenda

• 1015 1100 Presentation group 1
• 1100 1115 Coffee
• 1115 1200 Presentation group 2
• 1200 1245 Presentation group 3
• 1300 1400 Lunch
• 1400 1500 Tutorial Manfred Opper
• 1500 1515 Coffee
• 1515 Onward discussions

3
Outline

• Groups involved
• Multi-user communication channels
• Data deployment in distributed, mobile systems
• Resource allocation and routing
• Work packages and resources

4
Groups involved

• Aston NCRG
• Aston EE
• BTexact
• Kings College London

5
Main challenge

• To analyse complex communication scenarios using
  methods of statistical physics to exploit
  insight gained from the analysis to develop
  effective data deployment and multi-user
  communication methods

6
Objectives

• Analyse capabilities of multi-user communication
  channels with emphasis on multi-access and
  broadcast channels
• Robust and efficient data deployment methods in
  distributed and mobile systems
• Resource allocation and routing using methods
  from game theory, analysed using SP

7
I. Multi-user communication

• Conventional information theory cannot analyse
  complex communication and computing networks
  (mobile?)
• Methods of statistical physics are particularly
  suitable in the case of large systems, providing
  typical case results
• Can be used to analyse specific scenarios to
  obtain theoretical and practical limitations

8
I. What is the problem?
9
I. Multiuser communication
10
I. Multi-user communication

• Of particular interests broadcast, relay and
  multiple access - CDMA (many receivers,
  transmitters)
• Scenarios to be studied using Low Density
  Parity Check codes, non-linear approaches
• Main methods to be used replica analysis Bethe
  approximation

11
I. Broadcasting
12
I. Multiple access - CDMA
13
II. Distributed comm. computing

• Data deployment in distributed (mobile?) networks
• Weak nodes storage space, computing capability,
  unreliable, dynamic
• Data retrieval by getting data from neighbours
• Communication can be based on P2P

14
II. Data deployment

• Data is divided to segments and deployed on a
  graph representing the comp./comm. nodes
• Data retrieval by obtaining a subset of
  segments from neighbours

15
II. Coding / decoding

• Analysing codes suitable for data deployment
  (fountain and raptor, MDS codes?)
• Fast encoding/decoding methods
• Statistical mechanics analysis typical case for
  large systems, practical and theoretical
  limitations

16
II. Data distribution

• Optimal segment deployment for minimising access
  time (graph colouring)
• Maximise the probability of having a certain
  number of neighbours with different colours
• Consider different graphs

17
II. Methods

• Code analysis graph colouring replica method,
  Bethe approximation
• Modelling and performance estimation numerical
  simulations based on real data
• Improved approximation methods for both
  distribution design and decoding
• Additional advantage regulating transmission
  times by fixed length messages

18
III. Routing and resource management

• The main problem efficient allocation of
  resources, efficient routing
• Can be regarded as a local decision problem
• Routing strategies change dynamically

19
III. Routing and resource management

• Given local information from neighbours and
  distance to destination local decision
• Can be treated using same tools as minority game
  (probably) to optimise decision making
• From a global point of view
• Selfish local optimisation
• Methods used generating functional analysis

20
I. Work-packages resources

• Develop SP framework for analysing multi-user
  communication channels
• Obtain typical case results focusing on
  broadcast, multi-user access and LDPC
• Consider SP to other network and channel types in
  both lossy and lossless scenarios
• Resource 3 year RF, PhD student(?)

21
II. Work-packages resources (NC)

• Design analyse suitable codes
• Analyse data deployment strategies
• Devise efficient data deployment algorithms
• Develop computationally efficient decoding
  techniques
• Resource 2-3 year RF, PhD student

22
II. Work-packages resources (EE)

• Design suitable codes
• Model traffic patterns, especially of the
  non-Poissonian distributions
• Establish proper communication and networking
  protocols
• Investigate the performance
• Resource 2-3 year RF, PhD student(?)

23
III. Work-packages resources (KCL)

• Set up framework of a generic traffic problem
• Derive exact solutions for network optimal and
  selfish strategies using local information
• Study the effects of network structures on the
  solutions
• Study more realistic scenarios
• Resource 2-3 year RF, PhD student(?)

24
Total resources

• 10-12 year RF
• 2-4 PhD students
• 600-700K