Parallel Cellular Environments to Enable Scientists to Solve Complex Problems

1
Parallel Cellular Environments to Enable
Scientists to Solve Complex Problems

• DOMENICO TALIA
• ISI-CNR
• Rende, Italy
• talia_at_si.deis.unical.it
• San Feliu de Guixols, Spain, June 1999

2
OUTLINE

• Computational Simulation
• Cellular Automata
• Parallel Computing and Cellular Automata
• Parallel CA Systems
• CAMEL and CARPET
• Performance
• Conclusions

3
Computational Simulation

• Computational simulation is based on the use of
  computers for modeling and simulation of complex
  phenomena and systems in science and engineering.
• A computer equipped with problem-solving software
  tools may represent a virtual laboratory where it
  is possible to build a model for a given problem
  and run it under different conditions.
• A significant amount of work has been carried out
  in the area of PSEs for problems specified by
  differential equations.

4
Computational Simulation

• Our work has a different emphasis from much of
  the work in PSEs by focusing on cellular automata
  models, tools and systems for problem solving.
• Parallel computers represent a class of machines
  that can effectively support the computational
  simulation approach.
• This talk discusses the combined use of parallel
  computing systems and cellular automata for
  high-performance computational simulation.

5
Cellular Automata

• A cellular automaton is a discrete dynamical
  system composed of a large number of cells in a
  (1-D, 2-D, or 3-D) lattice.
• Each cell interacts only with its neighbor cells
  and there exists an evolution rule of the
  cellular automaton that changes the state of all
  cells simultaneously ( in parallel ) at discrete
  time steps.

6
Cellular Automata
The global behavior of the system is determined
by the evolution of the states of all cells as a
result of multiple interactions.
7
Cellular Automata
Two-dimensional lattices and neighborhoods
8
Cellular Automata

• Cellular automata (CA) are discrete dynamical
  systems that are often described as a counterpart
  to partial differential equations, which have the
  capability to describe continuous dynamical
  systems.
• CA offer us a perfect model for exploring systems
  with no central control and parallel execution.
• Explore, using a decentralized control, systems
  where simple local interaction occur for a large
  population of cells over a period of time.
• The basic idea is to try to describe a complex
  system simulating this system by the interaction
  of a massive number of cells following simple
  rules, not to describe its global behavior using
  difficult equations.

9
Parallel Computing and Cellular Automata

• The use of high-performance parallel computers
  supports the CA model to be of practical use for
  solving real world problems.
• Cellular Automata model represents a parallel
  computing model that can be implemented on
  parallel architectures because of its inherent
  parallelism and locality.
• CA exploit data parallelism by partitioning cells
  on the processing elements of a parallel
  computer.
• In a distributed-memory MIMD computer the
  parallelism of CA can be exploited using the SPMD
  (Single Program Multiple Data) approach.

10
Parallel CA Systems

• Several cellular environments have been
  implemented on sequential computers (PCs,
  workstations)
• In the past years, significant parallel cellular
  environments have been designed
• CAPE ? PECANS
• CAMEL ? P-CAM
• StarLogo ? NEMO
• DEVS ? CELLULAR
• CAM-8 (hardware)
• Application areas physics, biology, genetics,
  chemistry, cryptography, image processing,
  environment modeling, town planning, and finance.

11
Features of Parallel CA Systems

• The main features of parallel cellular software
  environments are
• a high-level programming layer for designing
  software models of complex phenomena and systems
  which is independent from the underlying parallel
  architecture.
• a graphical user interface (GUI) that allows the
  visualization of the complete evolution of a
  simulation of dynamic systems and the display of
  numerical values associate to a simulation.
• control and tuning facilities that allow for
  on-line computational steering of application
  evolution.
• scalable high-performance that allows these
  environments to support the efficient execution
  of very complex real world phenomena that cannot
  be simulated on sequential computers.

12
P-CAM

• P-CAM (Parallel Cellular Automata Modeling) is a
  simulation environment based on a computational
  framework of interconnected cells (virtual
  particles).
• Cells are arranged in graphs that define the cell
  interconnections and interactions.
• An initial decomposition of the graph on a
  parallel machine is generated and a load
  balancing strategy is used also to migrate cells.
• P-CAM has been implemented at University of
  Amsterdam using the C language with the MPI
  library.
• P-CAM has been used for modeling computational
  fluid dynamics, a lattice Boltzmann solver in
  combination with a simple aggregation process,
  and a parallel finite element simulation of an
  underwater explosion.

13
StarLogo

• StarLogo is a CA-based extension of the Logo
  programming language which has been designed by
  Mitchel Resnick at MIT.
• Starlogo provides simple constructs to define the
  evolution rules of the cells that compose a
  cellular automaton. A Starlogo environment has
  been implemented on the Connection Machine 1.
• Starlogo has been used to simulate
• the spread of a fire through a forest
• disease diffusion
• freeway traffic flow
• ant colonies
• gas molecules diffusion.

14
A Forest Fire Simulation Model

• The fire starts on specific point of the forest,
  and spreads to neighboring trees.
• The fire spreads in four directions north, east,
  south, and west.
• The fire's chance of diffusing towards all the
  forest depends critically on the density of trees
  in the forest.

15
NEMO

• NEMO (Neighborhood Modelling) is a CA based
  environment designed by the PARADIGM group at
  Carleton University.
• NEMO has been implemented on a MIMD distributed
  memory computer (Alex AVX Series II) providing,
  like the other CA systems mentioned here,
  transparent parallelism.
• The NEMO parallel system has been used for
  developing computational applications such as
• earthquake modeling,
• ice tracking,
• forest fire modeling, and
• crystal growth processes.
• This system is currently used for the development
  of a parallel tool for processing, storing and
  visualizing of spatial data in application areas
  such as GIS, remote sensing, medical computing,
  and robotics.

16
DEVS-C

• DEVS-C is a high-performance environment based
  on the DEVS (Discrete Event System Specification)
  formalism that supports the analysis, design and
  simulation of discrete event dynamic systems.
• The DEVS formalism, based on cellular automata
  theory and discrete event simulation, provides a
  means of specifying a mathematical object called
  a system defined as a lattice of cells.
• DEVS-C DEVS parallel implementation, uses the
  C language for programming simulation and
  models designed by the DEVS formalism.
• The DEVS-C system has been implemented at
  University of Arizona on parallel computers such
  as the Thinking Machines CM-5 and the IBM SP2.
• Applications watershed hydrology, ecosystem
  modeling.

17
CAMEL

• CAMEL is a CA environment designed for
  message-passing parallel computers.
• CAMEL can be used as a virtual laboratory for
  scientific applications it hides parallelism
  issues from a user.
• In CAMEL a user specifies only the transition
  function of a single cell of the system he wants
  to simulate by CARPET, a high-level cellular
  language defined as an extention of the C
  language.
• CAMELot is a portable implementation (Meiko CS-2,
  Linux PCs clusters, SGI) developed in the ESPRIT
  project COLOMBO by ISI-CNR, EPCC and ENEA.

18
CAMEL

• CAMEL implements parallel cellular automata using
  the SPMD model.
• The run-time system is composed of a set of
  macrocell processes. Each macrocell implements a
  partition of cells on a single processor of a
  parallel computer ? the programmer does
  not specify data distribution!
• All the macrocell processes execute in parallel
  to update the state of the cells that compose an
  automaton.
• A graphical user interface (GUI) allows dynamic
  visualization and on-line steering facilities.

19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
CARPET

• CARPET (CellulAR Programming EnvironmenT) is a
  language for parallel cellular processing.
• In a CARPET program a user can define
• the basic rules of the system to be simulated
• the macroscopic quantities that constitute the
  global parameters of a model
• but nothing needs to be specified about the
  parallel execution.
• The main features of CARPET are
• the definition of the cell state as a set of
  typed substates
• the simple definition of complex neighborhoods,
  that can be also time dependent, in a
  d-dimensional discrete Cartesian space
• addressing and updating operations that respect
  the CA semantics.

23
CARPET

• cadef
• / A simple forest fire model /
• dimension 2
• radius 1
• state (short ground)
• neighbor Moore8(0,-1N,-1,-NW,-1,0W,
• -1,1SW,0,1S,1,1SE,1,0E, 1,-1NE)
• if((cell_ground tree)
• (N_groundfire S_groundfire
  E_groundfire
• W_groundfire N_groundfire
  SW_groundfire
• SE_ground fire N_ground fire))
• update(cell_ground, fire)
• else
• if (cell_ground fire)
• update(cell_ground, dead)

24
CARPET
/ The Jacoby finite difference method in CARPET
/ cadef dimension 2 radius 1 state
(floal value) neighbor Neum4(0,-1N,-1,0W
, 0,1S,1,0E) if (step 0) if (GetY
256) cell_value 1.0 else cell_value
0.0 update(cell_value,(N_value W_value
S_value E_value)/4)
25
CARPET CAMEL

• CAMEL and CARPET have been used to implement
  complex algorithms for several real life
  applications such as
• landslide simulation,
• soil bioremediation models,
• lava flow models,
• freeway traffic simulation,
• gas particles, and
• genetic algorithms.
• COLOMBO (Parallel Computers Improve Clean Up of
  Contaminated Soils) funded by the ESPRIT
  programme.

26
CAMEL Performance Results
27
CAMEL Performance Results
28
Conclusions

• The cellular automata model is an efficient
  paradigm to achieve scalable performance on
  parallel computers.
• Parallel cellular processing provide
• a viable mathematical way to represent problems,
• the scalable performance of MIMD processors, and
• efficient environments supporting for
  interdisciplinary efforts for simulation in
  science and engineering.
• In the past years several environments and tools
  have been developed to support computational
  simulation based on parallel cellular processing.
• These tools provide assistance to users in
  formulating problems, running models on
  high-performance computers, and viewing and
  analyzing the results.

29
Further Info

• http//liinwww.ira.uka.de/ca/
• The Cellular Automata Pages (maintained by
  T. Worsch)

• http//isi-cnr.deis.unical.it1080/talia/CA.html
• CAMEL and CARPET

Cellular Automata Promise and Prospects in
Computational Science Special Issue of FUTURE
GENERATION COMPUTER SYSTEMS, December 1999.