Network Simulation COMM3F Network Planning and Evolutionary Computation

1
Network SimulationCOMM3FNetwork Planning
andEvolutionary Computation

• Dr. Sonia Tindle
• University of Sunderland
• 27th September, 2005

2
Overview

• Access networks
• Past research - BT
• Passive Optical Networks (PONs)
• Latest research - BT, Evolved Networks
• Knapsack/binpacking problems
• Particle Swarm Optimisation (PSO)
• BPON Optimisation Tool

3
Basic Telephone Network
Core blue Access yellow Exchanges
brown Customers pink
4
Copper Based Network
5
Increasing the Bandwidth

• Late 1990s new services challenging telcos to
  provide higher bandwidth quickly
• Optical fibre, Asymmetric Digital Subscriber Line
  (ADSL)
• ADSL uses existing copper wire
• Modem technology provides high-speed digital
  lines

6
ADSL

• Data is carried across a much wider range of
  frequencies through the twisted copper pairs
• Relies on advanced digital signal processing
  technology and VLSI computer technology
• Simultaneous Internet and voice/fax capabilities
  over single line
• 512 Kbps theoretically, nearer 200-500 Kbps

7
ADSL Problems

• Problems Attenuation and electrical
  interference
• PC Direct 2001 ADSL wont work at required
  rate for 25 UK consumers.
• http//www.sunderland.ac.uk/ts0jti/0library
• book1/chap04.htm

8
Copper Planning Tool - BT

• 1997 Access Network planning for greenfield sites
  was entirely paper-based
• Typically 3 days for 2 manual plans
• 57 BT offices
• Research brief To determine a least cost network
  in terms of the equipment and labour necessary to
  install the cabling infrastructure.
• GAs were employed to provide the optimisation
  mechanism.

9
Copper Planning Tool Screenshot
10
Benefits of Copper Planning Tool

• Near optimal network structures produced for
  initial design (single step approach)
• Least cost network in terms of equipment, cables,
  labour
• Rapid processing - greater efficiency
• Number of offices reduced from 57 to 20
• 20,000 plans per year for over 4 years
• compare manual method ( 2 plans for 3 days
  work)

11
Benefits of Copper Planning Tool

• Savings on capital expenditure
• 10 to 20 of 100M / year
• Strategic planning policy implemented company
  wide
• Evolved Networks
• http//www.evolvednetworks.com/
• Also produced planning tool for TPONs (Telephone
  Passive Optical Networks)

12
Passive Optical Network
Optical Network Units/ Terminals ONU/ONT
Optical Line Terminals in Local Exchange (OLT)
Secondary Splitter
Customers
Primary Splitter
Optical Distribution Network
13
Passive Optical Networks

• No active components in the network potentially
  more reliable
• Equipment is shared between customers so reducing
  costs
• Higher bandwidth less duct space required
• More secure difficult to tap into optical fibre
• Unaffected by water ingress
• Potential for even higher transmission
  frequencies gt20MHz

14
Full Service Access Networks (FSAN)

• Initially 7 telecoms companies in 1995
• Now 21 companies worldwide together with
  equipment manufacturers
• Each telco had different structures and
  deployment strategies for access networks
• Initial stage of optical access system required
  common system for effective cost and service
  spread

15
Broadband PON System
16
PON Specifications

• Telephone PON 20 Mbps
• 200 x 64 Kbps channels, up to 32 ONUs, max.
  distance 4 km.
• ATM-PON 155 or 622 Mbps
• 32/64 ONTs, max. distance 20 km.
• Ethernet PON 64 Kbps increments ? 1 Gbps
• SuperPON 2.5 Gbps downstream, 311 Mbps upstream,
  shared by 2048 ONTs, max. distance of 100 km.
  

17
Future PON Developments

• Broadcast services of APON expected to utilise
  WDM (Wave Division Multiplexing)
• Type of multiplexing developed for use on optical
  fibre
• WDM modulates each of several data streams onto a
  different part of the light spectrum
• Downstream gt 2 Gbps, will connect to customers by
  twisted pair, wireless drop or coaxial cable

18
Future PON Developments

• VDSL (Very high rate DSL)
• Supports asymmetric and symmetric applications
• Faster than ADSL (theoretically 28 Mbps)
• But has shorter reach than ADSL
• 70 of UK population are within 500m of a street
  cabinet
• Asymmetric service 10Mbps/a few Mbps
• Symmetric service 4-10 Mbps

19
Previous Research

• Optimised cabling infrastructure in terms of
  least cost solution
• Narrowband technology lt 2 Mbps
• Demand for narrowband technology grew at a steady
  2
• Greenfield sites
• Single step approach
• Cabling infrastructure is only a part of the
  overall telecommunications system

20
Latest Research

• Need to ensure full utilisation of the equipment
  necessary to operate the network
• Services are provided via electronic cards housed
  on racks in cabinets at both ends of the Access
  Network
• 6-8 exchanges 3m equipment more than was
  required to provide services

21
PON Equipment
Quantum Bridge QB5000 OLT
22
Motivation

• Need to maintain near optimal networks over time
  in addition to providing the initial design
• Demand for broadband technology is growing
  exponentially pressure from consumers and
  increased competition
• More complex networks Various types of service
  available (voice, data, video) require different
  bandwidths

23
Aim of the Project

• Reduce unnecessary capital expenditure on network
  equipment over a given period of time
• Ensure current facilities are as fully utilised
  as possible
• Consider both space (location) and time (of
  installation)
• Ensure customer demand for broadband services is
  met

24
Space (Location of ONUs)

• Equipment may be sited at a number of locations
• In general, an OLT may connect to 32 ONUs
• Each ONU houses a number of racks
• Each rack holds 10 electronic cards
• Each card may connect to 4 services
• Best combination of connections using least
  amount of equipment to meet customer demand

25
Knapsack/Bin Packing Problems

• Similar to knapsack and bin packing problems
• Knapsack
• Given items of different values and volumes,
  find the most valuable set of items that fit in a
  knapsack of fixed volume.
• Similar to attempting to find the most
  profitable way to populate the network with
  equipment.

26
Knapsack/Bin Packing Problems

• Bin Packing
• Determine how to put the most objects in the
  least number of fixed space bins.
• Similar to attempting to use the least number of
  ONT cabinets to provide telecommunications
  services.
• N.B. Also see resource allocation, scheduling
  and cutting stock problems.

27
Additional Factors

• Pre-allocate bandwidth
• Services compete for bandwidth
• Balance load between PONs
• Fill PONs in sequence
• Multi time steps
• Multi locations
• increasing
• complexity

28
Evolutionary Algorithms

• Past research - Genetic Algorithms (GAs)
• Latest research - compare GAs with Particle Swarm
  Optimisation (PSO)
• Final tests GA more robust, PSO also produced
  reasonably good results in acceptable time (under
  5 minutes)

29
Particle Swarm Optimisation

• Boids based on flocking behaviour of birds, fish
• http//www.red3d.com/cwr/boids/
• AND
• Analogy to social interaction and co-operation

30
GA and PSO Similarities

• Initial randomly generated population
• Fitness value to evaluate population
• Update population and search for optimum using
  random techniques
• Whole population moves as a group towards optimal
  area
• Success not guaranteed not exhaustive search

31
Differences between GAs and PSO

• Solution improvement
• GA Competition (survival of the fittest)
• PSO Co-operation (sharing info and experience)
• No genetic operators crossover, mutation
• Potential solutions (particles) are flown
  through the problem space following current
  optimum particles
• PSO has memory

32
Info Sharing Mechanisms

• Different info sharing mechanisms
• GA chromosomes share info with eachother
• PSO only global best or local best gives info
  to others one-way info sharing mechanism
  evolution only looks for the best solution

33
Particle Swarm Optimisation

• Imitates human social behaviour
• Individuals interact with each other
• Learn from their own experience
• Gradually the population members move into better
  regions of the problem space
• Evaluating, comparing and imitating leads to good
  solutions

34
Typical Local Search
Step A1 previous step
carry on in same direction (momentum)
A2 best particle
step to best particle for that
position A3 best neighbour
couple particles e.g. A1-3 vary
with time rand(1)

35
Typical Global Search
Step A1 previous step
carry on in same direction (momentum)
A2 best particle
step to best particle for
position A4 global best
couple particles e.g. A1-4 vary with
time rand(1)

36
Local and Global Topologies
Circle Topology Local Search
Wheel Topology Global Search

local neighbourhood
37
Time (of Equipment Installation)

• Previous tools generally provide initial design
  only
• Dont attempt to maintain a near-optimal design
  as network expands over time
• Time when equipment is to be installed during a
  network expansion period
• Cost of equipment can vary with time
• Savings if equipment purchased at right time

38
Equipment Installations

• Assumes increasing demand for broadband i.e.
  network is expanding
• Assumes equipment is not removed from system once
  installed
• Equipment costs can be set to increase/decrease
  at each time step
• System determines whether better to buy early or
  defer purchase
• BUT still meet customer demand

39
Phased Expansion of Access Network
Equipment Installations At Each Time Step
Amount of Equipment
Forecast Demand
Time Steps
Expansion Period
40
Problem Description

• Small access network two 622Mbps ATM PON systems
• 4 potential sites for locating equipment
• Planning horizon with 3 time steps
• 4 types of service measured in terms of 64Kbps
  channels

41
Service Types
42
BPON Tool

• Simultaneously generates near-optimal solutions
  at each of several time steps over a planning
  horizon.
• Ensures equipment costs minimised at each time
  step
• As well as over the network expansion period as a
  whole.

43
Data Structures (1)

• Complex programming task OMT, MVC paradigm and
  matrix computation.
• State of network representated by a set of
  communicating computer-based objects.

44
Data Structures (2)

• A planning horizon divided into a number of time
  steps
• A set of locations where network equipment may be
  sited
• A set of BPON systems servicing the access
  network
• A set of service types provided via each BPON
  system

45
Multi Time Steps and Locations
Array of Communicating Objects - ONT
46
Program Structure

• Object oriented analysis and design
• Modules do not need to be altered when a new
  object is added.
• A new object is created that inherits many of
  its distinctive features from existing objects.
• Programming in Java using Borland JBuilder
• MVC paradigm

47
Model View Controller Paradigm
48
Data Generated by the Optimisation Tool
49
Colour Coded Representation of Solutions
50
Beginning of an Optimisation Run
51
Service Allocation over Four Time Steps
52
Benefits

• Good solutions, rapidly, high precision
• Users need not be experts in network design
• Services provided with optimal allocation of
  cards
• Comprehensive computer-based records
• Schedules for ordering and installation
• Investigation of various scenarios
• Existing sites, greenfield, business case

53
References

• Swarm Intelligence by James Kennedy and Russell
  C. Eberhart, 2001, Morgan Kaufmann, ISBN
  1-55860-595-9 (Library 518.01 K25)
• Particle Swarm Optimization Homepage
• http//www.cis.syr.edu/mohan/pso/