Towards a Science of Networks: Communication Networks I

illnurturedtownvilleMobile - Wireless

Nov 21, 2013 (4 years and 1 month ago)

75 views

Towards a Science of Networks:

Communication Networks I

Report on workshop held at the University of
Birmingham, UK, 3


4 November 2005

(including brief comments on Rome and
Budapest workshops)


Costas Constantinou

31 August 2006

2

Workshop aims


Examine communication network theory from
different perspectives


physics, engineering
and computer science


Take stock of state of the art


Challenge “conventional wisdom” assumptions
for network design/operation (strong held
opinions versus fundamental properties)


Define a set of grand challenges to enable the
creation of a coherent scientific theory of
networks

31 August 2006

3

Some background history


Workshop was preceded by a U.S.
workshop on the complex behaviour of
adaptive, network
-
centric systems, hosted
by the University of Maryland on 12


14
July 2005


Interdisciplinary aspects of complexity science


Applicability to predictable emergent
behaviour in distributed systems/networks


Very strong emphasis on wireless networks

31 August 2006

4

Workshop presentations


Report on Workshop on the Complex Behaviour of
Adaptive, Network
-
Centric Systems


Dr. Stuart D. Milner, University of Maryland


Logical Network Abridgement leading to Diversity
and Resilience Measures on Networks


Dr Costas Constantinou, University of Birmingham


Spreading Processes on Complex Networks: Theory
and Applications


Maziar Nekovee Complexity Group, Mobility
Research Centre, BT

31 August 2006

5

Workshop presentations (cont.)


Algorithms and Survivable Protocols for Scale
-
Free
and Scale
-
Free Small World Networks


András Lörincz, Eötvös Loránd University


Information flow issues in cross
-
layer models of
wireless communication networks


Professor Leandros Tassiulas, University of
Thessaly


MAC
-
Layer Selfish Behaviour in Wireless Networks:
A Repeated Game Approach


Jerzy Konorski, Gdansk University of Technology

31 August 2006

6

Workshop presentations (cont.)


Transmit beamforming strategies for PHY
-
layer
multicasting with QoS guarantees


Professor Nikos Sidiropoulos, Technical university of
Crete


Greece


Traffic theory for the Internet and its implications on
network design


Jim Roberts, France Telecom


Congestion and Centrality


Raul J. Mondragón, Queen Mary, University of
London

31 August 2006

7

Workshop presentations (cont.)


Physics of networks: state of the art


José F. F. Mendes, University of Aveiro


Large
-
Scale Behaviour of Packet
-
Switched
Networks


Sanya Stepanenko, University of Birmingham

31 August 2006

8

Workshop panel discussions


The first panel session concentrated on the

network properties that need to be predicted
reliably


Conclusion
: Nearly all observable variables are
either
data rates
, such as capacity, throughput, loss
rate, etc., or
delays
, such as packet transport delay,
start
-
up and recovery times, etc., all of which should
be predicted statistically to estimate their
distributions & correlations
, as their mean values
are not sufficient

31 August 2006

9

Workshop panel discussions (cont.)


The second panel session attempted to
list the
ingredients for a viable theory of networks


Four main components to such a theory:


Input traffic demand to a network


Network topology


Routing protocol operation and


Interaction between above three components


A clean separation of time
-
scales for the various
component processes would make the
formulation of a theory of networks easier, but
the discussion did not go into establishing
whether such a separation is applicable

31 August 2006

10

Workshop panel discussions (cont.)


Conclusion:

The individual components of a
theory of networks already exist and are well
understood in isolation. However, these partial
theories fall short of an overall theory of
networks in two ways:


The
interactions

between different components need
to be specified in a unified framework, taking
particular care to determine the relevant time
-
scales
pertinent to the problem under study


Even though
in principle

current formalisms can be
used to describe networks in all their aspects, the
number of variables becomes intractably large

even
for modest sized networks

31 August 2006

11

Rome Workshop


Biologically Inspired Information Systems
Workshop


Held in Rome, 24


26 July 2006


Organised by the Universit
à

di Roma


La
Sapienza


Funded by ONR

31 August 2006

12

Rome Workshop


Biologists largely working with neural networks (NN)


NN nodes are non
-
linear and have non
-
linear
interactions


NN are viewed as computation circuits


Unlike communication networks coarse
-
graining is
meaningless within our current framework of
understanding


Neuroplasticity, i.e. the brain's ability to reorganize itself
by forming new neural connections to compensate for
injury and disease and to adjust its activities in response
to new situations or to changes in their environment, is
an interesting unexplored mechanism that could be
exploited in self
-
healing networks

31 August 2006

13

Rome Workshop (cont.)


Analogies with some biological networks are
useful:


Routing based on social insect paradigm (ants)


Synchronisation based on coupled oscillators
(fireflies)


Topology management in overlay networks based on
differential cell adhesion


Immune system inspired security


Game theory [not really biology, but economics] for
emergence of self
-
organised cooperation

31 August 2006

14

Rome Workshop (cont.)


General problem is the very restricted
range of known/predictable emergent
behaviour from local interaction rules


Concept of time/time
-
scales and their
impact on dynamics is often based on
educated guesswork


Strict adherence to biological analogies
not always sensible


should only be used
as starting point

31 August 2006

15

Budapest workshop


Social Networks and Complexity
Workshop


Held in Budapest, 31 July


2 August 2006


Organised by the Institute for Advanced
Study at the Collegium


Co
-
funded by AFOSR and ONR

31 August 2006

16

Budapest workshop (cont.)


Social scientists are heavily concerned with the
study of social network structures: either that of
static topologies (e.g. clustering, cliques, etc.), or
the topology evolution “dynamics”


Theory (random, scale
-
free and small world graph
[growth process] models dominate); eigenvector
spectrum of connectivity matrix


Numerical simulations (node/link decimation/addition,
percolation thresholds)


Measurements

31 August 2006

17

Budapest workshop (cont.)


Simple coupled dynamical process node models
(agents as local dynamical processes


often
based on Game Theory)


Co
-
clustering of overlay processes on network
topology using bi
-
partite graphs


A bipartite
network can be used as a (general) mapping
between two interacting networked processes


Smallpox epidemic dynamics (not social
networks) has strong similarities to mobile ad
hoc networks


Many social networks tend to be “networks” in a
not very precise sense

31 August 2006

18

Common problems across
disciplines


Topology and overlay processes are both
dynamic


but often on different time scales


Layering of processes (more than one overlays)


The predictable mapping local interaction
behaviours to emergent global observable
properties is poorly understood


Not enough effort goes towards identifying
differences as opposed to similarities (not
everything is scale
-
free symptom)

31 August 2006

19

Common problems across
disciplines (cont.)


Identifying and modelling the sources of
stochasticity is often incompletely done


Too many types of topology classification
(taxonomy), but no fundamental understanding
of topology groups/types


No identified order parameters for emergent
properties


conversely we know nothing about
invariants

31 August 2006

20

Differences between disciplines


Exaggerated view:


Communications


Data flow networks


Routing process


Biology


Neural networks


Topology re
-
construction and pattern identification


Social sciences


Agent networks


Static topology analysis

31 August 2006

21

State of the art?


All three disciplines are currently posing
problems


New problems that arise could lead to a
convergence in the posing of the
fundamental problems in network science