Miguel FERREIRA

1
A web application for distributed computing

• Miguel FERREIRA
• Department of Computer Science
• Miguel CASQUILHO
• CPQ Centre for Chemical Processes
• Department of Chemical Engineering
• IST Instituto Superior Técnico
• UTL Technical University of Lisbon
• http//web.ist.utl.pt/mcasquilho/

Lisboa (Portugal), 1922 June 2013
2
A web application for distributed computing

• Abstract  ?  A web application of scientific
  computing is presented, using MPI parallel
  programming, for a typical problem in packaging.
  As no analytical solution seems viable, a Monte
  Carlo approach was used. The following points
  are addressed, starting with the original
  (serial) solution
• Packaging problem and local serial computing
• Remote parallel computing
• Conclusions
• From the authors experience, the web environment
  is clearly recommended for scientific computing,
  for academic and industrial purposes.
• Scientific computing on the Internet in a
  site, the user inserts his data, executes a
  program, and receives his results (no
  installations).

3
A web application for distributed computing

1. Packaging problem and local serial computing
2. Remote parallel computing
3. Conclusions

Alfredo Keil (18501907) Caminho da floresta
4
1. 2. 3. Packaging problem and local serial
computing

• Packaging problem
• A container to be filled with objects of random
  size (weight, thickness, etc.)
• Content constrained by lower, L, and upper, U,
  specification limits
• How to get 2 kg of oranges ?
• Minimize the fractions out of spec (below L and
  above U)
• Random behaviour considered
• Truncated Gaussian
• Truncation ? from calibration (fruits), moulding
  (biscuits)

A pack of biscuits Schematically
5
1. 2. 3. Packaging problem and local serial
computing

• Local solution using serial computing
• Web interface to receive user input
• Local call of the computing application
• Native Fortran program that runs the Monte Carlo
  simulation
• Treatment of output and graphic
• Formatting and presentation of results on the Web
• Original Fortran 90 program was previously
  developed for stand-alone local use.
• The new, parallel MPI version of the program was
  ported to C.

6
1. 2. 3. Packaging problem and local serial
computing

• The packaging problem was studied with
• (a) Simple, sequential filling (usual)
• (b)Tube-like (FIFO) filling
• Strategy (b), as expected, revealed better
  results

7
A web application for distributed computing

1. Packaging problem and local serial computing
2. Remote parallel computing
3. Conclusions

Carlos Botelho (1899-1982) Lisboa, tapeçaria de
Portalegre
8
1. 2. 3. Remote parallel computing

• Remote solution using parallel computing
• The original Fortran program was ported to C, and
  extended with MPI
• The number of Monte Carlo trials (e.g., 109) was
  evenly distributed by the available nodes
• Web interface unchanged from the serial version
• A new communication layer developed between the
  web interface and the MPI program
• Remote call of the computing application
• Transfer of input and output end-to-end
• The remote side to call the MPI programa locally

(As imagens são hyperlinks)
9
1. 2. 3. Remote parallel computing

• System topology

10
1. 2. 3. Remote parallel computing

• Results of an MPI run, 109 trials, finer (better)
  than serial 106

11
A web application for distributed computing

1. Packaging problem and local serial computing
2. Remote parallel computing
3. Conclusions

Henrique Medina (19011988)Tomando chá
12
1. 2. 3. Conclusions

• A packaging problem frequently found in Industry
  was addressed.
• The problem was previously solved with a Fortran
  serial application, executable on the Web.
• The original problem was ported to C and then
  parallelized with MPI, also accessible through
  the Web.
• Making applications accessible over the Internet
  avoids platform incompatibilities (Windows, Mac,
  Linux, etc.), and surpasses the limitations of
  the users computer.
• The use of the Internet for scientific computing
  dispenses with any required software
  installation.
• It is, thus, recommended for academic or
  industrial purposes.

13
Acknowledgements

• DEI Department of Computer Science and
  Engineering, Instituto Superior Técnico (IST),
  Technical University of Lisbon (UTL)
• CPQ Centro de Processos Químicos (Centre for
  Chemical Processes), IST, UTL
• DEQ Department of Chemical Engineering, IST, UTL
• CIIST Centro de Informática do IST (Centre for
  Informatics), IST, UTL, Dr. Jorge Amador
• Milipeia, Lab. de Computação Avançada (Lab. of
  Advanced Computing), University of Coimbra
• Prof.ª Fátima C. Rosa (DEQ, IST, UTL)
• Keywords
• Scientific computing, Internet, parallel
  computing, MPI

14
Bibliography

• Casquilho, M. 2012 Scientific computing on the
  Internet, a little-trodden field, Universidade
  Aberta, Porto Salvo.
• Casquilho, M. 2013 Computação científica,
  Internet, Indústria (Scientific computing,
  Internet, Industry), 1.st Portuguese Meeting on
  Mathematics for Industry, FCUP, Universidade do
  Porto, Porto, 0608 June.
• Ferreira, M., M. Casquilho 2013 Scientific
  computing over the Internet an example in
  Geometry, WorldCIST'13, World Conference on
  Information Systems and Technologies, Olhão
  (Portugal)
• Franco, B., M. Casquilho 2011 A Web
  application for scientific computing combining
  several tools and languages to solve a
  statistical problem, CISTI'2011, 6.ª Conferência
  Ibérica de Sistemas e Tecnologias de Informação,
  Chaves (Portugal)
• Hughes, Alan 2006 University-industry linkages
  and UK science and innovation policy, Working
  Paper No. 326, Centre for Business Research,
  University of Cambridge, Cambridge (UK)

15
A web application for distributed computing

• Miguel FERREIRA
• Department of Computer Science
• Miguel CASQUILHO
• CPQ Centre for Chemical Processes
• Department of Chemical Engineering
• IST Instituto Superior Técnico
• UTL Technical University of Lisbon
• http//web.ist.utl.pt/mcasquilho/

Lisboa (Portugal), 1922 June 2013