A Distributed Approach for
Planning Radio Communications
1
2
3
David Kidner , Ian Fitzell , Phillip Rallings ,
2
3
Miqdad Al Nuaimi & Andrew Ware
University of Glamorgan
1
2
School of Computing School of Electronics
3
Division of Mathematics & Computing
Pontypridd, Rhondda Cynon Taff
WALES, U.K. CF37 1DL
e-mail: dbkidner@glam.ac.uk
Scope
•
•
•
•
•
•
Spatial Data Problems & Site Selection
From Sight: Visibility Analysis
To Sound: Radio Field Planning
Topographic Modelling
Parallel Solutions
Results & Conclusions
Geocomputation’99 July 25th - 28th
Optimal Site Selection & Planning
• Site selection (or location allocation) dates
back to the 1950s and 60s
• However, the availability of spatial data and GIS
(including spatial modelling and analysis)
opens up greater challenges
– More efficient and effective analysis
– Environmentally-acceptable solutions
– Optimal solutions
Geocomputation’99 July 25th - 28th
From Sight: Visibility Analysis ...
• GIS applications may require visibility
functions for
– minimising visual intrusion
• e.g. contentious developments such as wind farms
– maximising the field-of-view
• e.g. radar or missile sites
• Massive workloads, compounded by very high
resolution datasets
Geocomputation’99 July 25th - 28th
To Sound: Radio Field Planning
• Path loss models describe the signal
attenuation between the transmitter and
receiver as a function of the propagation
distance and other parameters related to the
terrain profile and its surface features.
• Role of radio planning engineer is critical
– increased deregulation & network providers
– limited radio spectrum
Geocomputation’99 July 25th - 28th
Radio Field Planning
• Point-to-point links are generally
straightforward
– milliseconds to seconds
• Broadcast Coverages (to a field-of-view)
– minutes to hours
• Optimal Transmitter Locations
– hours to days to weeks
Geocomputation’99 July 25th - 28th
Radio Path Planning
Geocomputation’99 July 25th - 28th
Radio Communications Planning
Geocomputation’99 July 25th - 28th
Topographic Modelling
• Topographic Data Quality and Accuracy
– will greatly improve application performance
– Satellite Imagery
• Clutter categories (dense urban, suburban, vegetation,
water features)
– Aerial Photography (including heights)
– Existing Mapping
– LiDAR
• Data Structures?
Geocomputation’99 July 25th - 28th
Airborne Laser Scanning - LiDAR
(Cardiff)
Geocomputation’99 July 25th - 28th
LiDAR
•
•
•
•
Very high resolution (1 or 2m as a DEM)
With or Without Clutter
Accurate
Cheap
Geocomputation’99 July 25th - 28th
Managing Complex Data
Geocomputation’99 July 25th - 28th
Complex
Features
• Elevated Features
• Roof
Ridges
• Vegetation
Geocomputation’99 July 25th - 28th
Proposed 3D Standard for
Topographic Data (for Radio Planning)
Geocomputation’99 July 25th - 28th
Design Issues for Parallel Algorithms
• Sometimes difficult to recognise parallel
aspects of a task
– If it takes 1 woman 9 months to produce a baby, how long
will it take 2 women ?
– Some things are inherently sequential
• How do we split up the tasks ?
– Data or Task Parallelism?
• How do we store the data ?
– Shared or Distributed Memory Architecture?
Geocomputation’99 July 25th - 28th
The Need for a
Parallel Radio Broadcast Algorithm
• Determining an optimal transmitter location
Geocomputation’99 July 25th - 28th
Previous Work
• Based on TRANSPUTERS (a distributed memory
architecture, specifically designed for parallel
processing)
• Very good at transferring information between
processors, but little processing power and
limited memory
• Transputers failed to capture the share of the
processor market that they should have!
Geocomputation’99 July 25th - 28th
Parallel Workstation Cluster
• Advances in the field of networks & operating systems
have provided organisations with a valuable nonspecialised, general purpose parallel processing
resource.
• Cluster computing can scale to provide a very large
parallel machine and specialised hardware can be
made available to all machines.
• Each individual machine would also have total and
independent control of its own resources (e.g. memory,
disk, etc.)
Geocomputation’99 July 25th - 28th
Current State of Play
• We have looked at a Data Parallel approach on
what is essentially a Distributed Memory
architecture.
• Looked at numerous STATIC & DYNAMIC
approaches to the allocation of data.
– Blocks, Quadrants, Octants, Rows/Columns,
Individual Points, etc.
Geocomputation’99 July 25th - 28th
Parallel Implementation
Comparison Indicators
Speed-up = elapsed time of a uniprocessor
elapsed time of the multiprocessors
Efficiency
Geocomputation’99 July 25th - 28th
=
speed-up * 100
number of processors
Test Data
Geocomputation’99 July 25th - 28th
(520 Possible Transmitter Locations)
Speed-Up Performance
24
Speed-Up Performance
22
20
Sight
Sound
18
Speed-Up
16
14
12
10
8
6
4
2
0
2
4
Geocomputation’99 July 25th - 28th
6
8
10
12
14
16
Number of Processors
18
20
22
24
Relative Efficiency
100
Relative Efficiency
Sight
99
Sound
98
%
97
96
95
94
93
92
2
4
Geocomputation’99 July 25th - 28th
6
8
10
12
14
16
Number of Processors
18
20
22
24
Summary
• Phenomenal Results!
• Distributed cluster architecture is ideally
suited for spatial data processing
• Dynamic partitioning is consistently superior
to static partitioning
– the variability of terrain can seriously affect
load-balancing
– small workloads are superior, provided
communication overheads can be minimised.
Geocomputation’99 July 25th - 28th