University of Extremadura University of Extremadura Department of - - PowerPoint PPT Presentation

▶
university of extremadura university of extremadura
SMART_READER_LITE
LIVE PREVIEW

University of Extremadura University of Extremadura Department of - - PowerPoint PPT Presentation

Aug 19, 2023 705 likes •1.03k views

University of Extremadura University of Extremadura Department of Electronics Department of Electronics CAPI Research Group CAPI Research Group Public Resource Computing (PRC) Public Resource Computing (PRC) projects at Extremadura projects

slide-1
SLIDE 1

Public Resource Computing (PRC) Public Resource Computing (PRC) projects at Extremadura projects at Extremadura

  • Dr. Carlos J. García Orellana
  • Dr. Carlos J. García Orellana

Associate Professor Associate Professor Research Associate at CERN Research Associate at CERN

University of Extremadura University of Extremadura

Department of Electronics Department of Electronics CAPI Research Group CAPI Research Group

carlos@capi.unex.es CERN, September 2006

slide-2
SLIDE 2

CERN, September 2006 CERN, September 2006

2

Summary Summary

Summary. Summary.

► Context.

Context.

  • Global data.

Global data.

  • IT Regional Government policy.

IT Regional Government policy.

  • CETA-CIEMAT.

CETA-CIEMAT.

► CERN – Extremadura Collaboration Agreement.

CERN – Extremadura Collaboration Agreement.

► Ongoing PRC projects at Extremadura.

Ongoing PRC projects at Extremadura.

► Conclusions.

Conclusions.

slide-3
SLIDE 3

CERN, September 2006 CERN, September 2006

3

Extremadura context - General Extremadura context - General

► Geographical data.

Geographical data.

  • South-west of Spain.

South-west of Spain.

  • Autonomous community (similar to 'canton').

Autonomous community (similar to 'canton').

  • Area: 41.634 km

Area: 41.634 km2

2 (as Switzerland).

(as Switzerland).

  • Population: 1.084.600 (1/7 of Switzerland)

Population: 1.084.600 (1/7 of Switzerland). .

  • Density: 26 hab./km

Density: 26 hab./km2

2.

.

  • Two provinces: Badajoz and Cáceres.

Two provinces: Badajoz and Cáceres.

  • Main

Main cities cities: :

► Badajoz (145.000 inhab.).

Badajoz (145.000 inhab.).

► Cáceres (90.000 inhab.).

Cáceres (90.000 inhab.).

  • It is one of the economically most modest regions of Spain.

It is one of the economically most modest regions of Spain.

► Remarks.

Remarks.

  • Large territory, population small and dispersed.

Large territory, population small and dispersed.

  • To provide population with the access to services, a distributed

To provide population with the access to services, a distributed approach, in all aspects, is necessary. approach, in all aspects, is necessary.

slide-4
SLIDE 4

CERN, September 2006 CERN, September 2006

4

Extremadura context – University Extremadura context – University

► University of Extremadura (UEX).

University of Extremadura (UEX).

  • Founded in 1973.

Founded in 1973.

  • One University, four campuses.

One University, four campuses.

  • Main campuses at Badajoz and Cáceres.

Main campuses at Badajoz and Cáceres. Others at Mérida and Plasencia. Others at Mérida and Plasencia.

  • 26.000 students, 75 studies.

26.000 students, 75 studies.

  • Around 150 research groups.

Around 150 research groups.

slide-5
SLIDE 5

CERN, September 2006 CERN, September 2006

5

Extremadura context – IT Regional Government Policy Extremadura context – IT Regional Government Policy

► Objectives

Objectives

  • Not to lose the progress in IT.

Not to lose the progress in IT.

  • To attract people related to those technologies to

To attract people related to those technologies to Extremadura (IBM and INDRA software factories). Extremadura (IBM and INDRA software factories).

  • To increase the investments in R&D.

To increase the investments in R&D.

  • Emphasis on educational sector.

Emphasis on educational sector.

  • To use open standards.

To use open standards.

  • Technological neutrality.

Technological neutrality.

► Main programs:

Main programs:

  • GNU/Linex.

GNU/Linex.

  • Educational network.

Educational network.

  • Technological literacy.

Technological literacy.

  • Scientific network.

Scientific network.

  • Others (Vivernet, ...)

Others (Vivernet, ...) Problems:

  • Population dispersion.
  • Extremadura is not, in fact, the

richest region of Spain.

slide-6
SLIDE 6

CERN, September 2006 CERN, September 2006

6

Extremadura context – IT Main Programs (I) Extremadura context – IT Main Programs (I)

► Linex.

Linex.

  • Started in 2002: the first Linux distribution promoted by a

Started in 2002: the first Linux distribution promoted by a public administration. public administration.

  • A Debian/Linux flavour (

A Debian/Linux flavour (http://www.linex.org http://www.linex.org). ).

  • Linux distribution thought out for Extremadura.

Linux distribution thought out for Extremadura.

  • Versions for primary school, secondary school, and public

Versions for primary school, secondary school, and public administration. administration.

  • New 2006 version: many new features.

New 2006 version: many new features.

slide-7
SLIDE 7

CERN, September 2006 CERN, September 2006

7

Extremadura context – IT Main Programs (II) Extremadura context – IT Main Programs (II)

► Educational and institutional network.

Educational and institutional network.

  • Interconnects more than 1.700 buildings.

Interconnects more than 1.700 buildings.

  • Includes all secondary schools.

Includes all secondary schools.

  • Almost 70,000 computers are running GNU/Linex.

Almost 70,000 computers are running GNU/Linex.

  • Bandwidth: 2 Mbps (more where needed).

Bandwidth: 2 Mbps (more where needed).

► Technological literacy.

Technological literacy.

  • 33 centres distributed in Extremadura.

33 centres distributed in Extremadura.

  • 600.000 people trained in basic internet access.

600.000 people trained in basic internet access.

  • Actuations in order to allow every people to get access to ADSL

Actuations in order to allow every people to get access to ADSL services. services.

  • All villages of Extremadura have ADSL service since 2005.

All villages of Extremadura have ADSL service since 2005.

slide-8
SLIDE 8

CERN, September 2006 CERN, September 2006

8

Extremadura context – IT Main Programs (III) Extremadura context – IT Main Programs (III)

► Scientific Network.

Scientific Network.

  • To interconnect the main cities of Extremadura.

To interconnect the main cities of Extremadura.

  • During 2006 the main campuses of the University will be connected.

During 2006 the main campuses of the University will be connected.

  • To be finished in 2007.

To be finished in 2007.

  • Bandwidth: 10 Gbps.

Bandwidth: 10 Gbps.

  • Connected to RedIris (Spanish scientific network) and Portugal.

Connected to RedIris (Spanish scientific network) and Portugal.

► Recent actions.

Recent actions.

  • All computers in Extremadura public administration will run Linex in

All computers in Extremadura public administration will run Linex in

  • ne year.
  • ne year.
  • Documents interchange will be done using open standards.

Documents interchange will be done using open standards.

  • The Regional Government was invited to talk about its experiences

The Regional Government was invited to talk about its experiences in in “Unitar Symposium on ICT Policy Issues for Development” “Unitar Symposium on ICT Policy Issues for Development”, held in , held in United Nations (New York) on August 2006. United Nations (New York) on August 2006.

slide-9
SLIDE 9

CERN, September 2006 CERN, September 2006

9

Extremadura context – CETA-CIEMAT Extremadura context – CETA-CIEMAT

► CETA-CIEMAT is an excellence research centre that has been created

CETA-CIEMAT is an excellence research centre that has been created at Trujillo (Extremadura) in a former abbey. at Trujillo (Extremadura) in a former abbey.

► Objectives.

Objectives.

  • Its current lines of research are Grid focused and mainly oriented to

Its current lines of research are Grid focused and mainly oriented to EELA (EGEE spin-off) and BOINC. EELA (EGEE spin-off) and BOINC.

  • To offer an infrastructure for computation, storage and support to

To offer an infrastructure for computation, storage and support to various BOINC projects; as well as production and development of various BOINC projects; as well as production and development of applications for this platform. applications for this platform.

slide-10
SLIDE 10

CERN, September 2006 CERN, September 2006

10

Extremadura context – CETA-CIEMAT (II) Extremadura context – CETA-CIEMAT (II) At the end of 2005 CIEMAT purchased for CETA the following At the end of 2005 CIEMAT purchased for CETA the following computing resources: computing resources:

► Capacities: CPU.

Capacities: CPU.

  • 97 Blades Dell PowerEdge 1855 Dual Xeon 3.2GHz, 4 Gb RAM.

97 Blades Dell PowerEdge 1855 Dual Xeon 3.2GHz, 4 Gb RAM.

  • 56 HP Proliant BL25p, dual Opteron (2.0 GHz) dual core, 4 Gb RAM.

56 HP Proliant BL25p, dual Opteron (2.0 GHz) dual core, 4 Gb RAM.

  • Totals: 418 CPU cores, 639,462 SpecInt2K.

Totals: 418 CPU cores, 639,462 SpecInt2K.

► Capacities: Storage.

Capacities: Storage.

  • In hard disk: 11 “logical bricks” with 2 servers plus 1 DS4100 plus 1

In hard disk: 11 “logical bricks” with 2 servers plus 1 DS4100 plus 1

  • EXP100. Total: 120 TB.
  • EXP100. Total: 120 TB.
  • In tape: IBM LTO 3584, 4 drives and 400 tapes. Total: 160 TB.

In tape: IBM LTO 3584, 4 drives and 400 tapes. Total: 160 TB.

In 2007, the CETA will be connected to scientific network at 10 GBps. In 2007, the CETA will be connected to scientific network at 10 GBps.

slide-11
SLIDE 11

CERN, September 2006 CERN, September 2006

11

Summary Summary

Summary. Summary.

► Context.

Context.

► CERN – Extremadura Collaboration Agreement.

CERN – Extremadura Collaboration Agreement.

  • Description.

Description.

  • BOINC.

BOINC.

  • Garfield.

Garfield.

► Ongoing PRC projects at Extremadura.

Ongoing PRC projects at Extremadura.

► Conclusions.

Conclusions.

slide-12
SLIDE 12

CERN, September 2006 CERN, September 2006

12

CERN-Extremadura Collaboration Agreement CERN-Extremadura Collaboration Agreement

► General.

General.

  • Signed in February 2005 at Mérida.

Signed in February 2005 at Mérida.

  • General framework for specific projects.

General framework for specific projects.

  • Focused in “Open Source” technologies.

Focused in “Open Source” technologies.

► Fields of collaboration.

Fields of collaboration.

  • Healthcare.

Healthcare.

  • E-Learning.

E-Learning.

  • IT.

IT. First project: To develop Mammogrid project technologies in Extremadura hospitals. First project: Web application to teach the Physics of Light to secondary students. First project: BOINC - Use secondary schools' resources for Desktop Grid Computing (in contrast to Volunteer Computing).

slide-13
SLIDE 13

CERN, September 2006 CERN, September 2006

13

CERN-Extremadura Collaboration Agreement - BOINC CERN-Extremadura Collaboration Agreement - BOINC

► Berkeley Open Infrastructure for Network Computing (BOINC).

Berkeley Open Infrastructure for Network Computing (BOINC).

  • PRC middleware resulted from the evolution of

PRC middleware resulted from the evolution of SETI@home SETI@home project. project.

  • One server can have thousands of clients.

One server can have thousands of clients.

  • Easy to start a project with moderate knowledge on computing.

Easy to start a project with moderate knowledge on computing.

  • Not all applications are suitable (high compute/IO ratio needed).

Not all applications are suitable (high compute/IO ratio needed).

  • Web interface, support for large amount of data, ...

Web interface, support for large amount of data, ...

  • Details:

Details: http://boinc.berkeley.edu http://boinc.berkeley.edu. .

► Goals of collaboration.

Goals of collaboration.

  • To train Extremadura staff in BOINC by participating in CERN projects.

To train Extremadura staff in BOINC by participating in CERN projects.

  • To evaluate the computing power that we can obtain in practice.

To evaluate the computing power that we can obtain in practice.

  • To develop and implement administration tools.

To develop and implement administration tools.

  • To carry out projects of interest.

To carry out projects of interest. I'm here for this reason.

slide-14
SLIDE 14

CERN, September 2006 CERN, September 2006

14

CERN-Extremadura Collaboration Agreement - Garfield CERN-Extremadura Collaboration Agreement - Garfield

► About Garfield.

About Garfield.

  • Is a computer program for the detailed simulation of two- and three-

Is a computer program for the detailed simulation of two- and three- dimensional drift chambers. dimensional drift chambers.

  • Details at

Details at http://www.cern.ch/garfield http://www.cern.ch/garfield. .

  • Used by LHC community.

Used by LHC community.

► Garfield on BOINC.

Garfield on BOINC.

  • New application to complement SixTrack (

New application to complement SixTrack (L LHC@home HC@home). ).

  • Application ported to BOINC and server installed when I arrived to

Application ported to BOINC and server installed when I arrived to CERN. CERN.

  • First collaboration with ATLAS.

First collaboration with ATLAS.

  • My initial work was centered on job submission and assimilation systems

My initial work was centered on job submission and assimilation systems (supervised by Ignacio Reguero). (supervised by Ignacio Reguero).

  • More on BOINC activities at CERN in CNL-2006-02.

More on BOINC activities at CERN in CNL-2006-02.

slide-15
SLIDE 15

CERN, September 2006 CERN, September 2006

15

CERN-Extremadura Collaboration Agreement – Garfield (II) CERN-Extremadura Collaboration Agreement – Garfield (II)

► Server.

Server.

  • Installation of a new server

Installation of a new server boinc-server.cern.ch boinc-server.cern.ch (thanks to P. Defert). (thanks to P. Defert).

  • Jobs submission system using AFS filesystem (based on

Jobs submission system using AFS filesystem (based on LHC@home LHC@home work work). ).

  • Assimilation system based in AFS, too.

Assimilation system based in AFS, too.

  • Web page (

Web page (http://boinc-server.cern.ch/garfield/cern_stats2.php http://boinc-server.cern.ch/garfield/cern_stats2.php) and ) and OpenOffice database to get statistics. OpenOffice database to get statistics. After validation (BOINC uses redundancy), the completed jobs have to be returned to the user through BOINC API. We return the results and standard

  • utput files.
slide-16
SLIDE 16

CERN, September 2006 CERN, September 2006

16

CERN-Extremadura Collaboration Agreement – GarField (III) CERN-Extremadura Collaboration Agreement – GarField (III)

► Clients.

Clients.

  • We have used 'lxboinc' cluster at CERN (using recycled CPUs).

We have used 'lxboinc' cluster at CERN (using recycled CPUs).

  • 30 machines from participant groups at Extremadura (University and

30 machines from participant groups at Extremadura (University and CETA-CIEMAT). CETA-CIEMAT).

  • 45 test machines from two schools in Extremadura.

45 test machines from two schools in Extremadura.

► Results.

Results.

  • Delivered jobs: 26,597.

Delivered jobs: 26,597.

  • CPU days (as Pentium IV 2.8 GHz): 2,579.

CPU days (as Pentium IV 2.8 GHz): 2,579.

  • BOINC credits: 314,167.

BOINC credits: 314,167.

  • All the works submitted have been delivered without problems.

All the works submitted have been delivered without problems.

slide-17
SLIDE 17

CERN, September 2006 CERN, September 2006

17

CERN-Extremadura Collaboration Agreement – More work CERN-Extremadura Collaboration Agreement – More work

► Garfield port to Windows.

Garfield port to Windows.

  • Important to use Garfield in

Important to use Garfield in LHC@home LHC@home. .

  • Done with cygwin (but using Windows API).

Done with cygwin (but using Windows API).

  • Numerical differences:

Numerical differences: ✗ Between Linux and Windows versions. Between Linux and Windows versions. ✗ Small if same compiler version is used. Small if same compiler version is used. ✗ New validation daemon has been needed. New validation daemon has been needed.

► Work in progress.

Work in progress.

  • Deploy Garfield in

Deploy Garfield in LHC@home LHC@home. .

  • Prepare Garfield to run in the Grid.

Prepare Garfield to run in the Grid.

  • Test Garfield with several thousand of machines in Extremadura.

Test Garfield with several thousand of machines in Extremadura.

Problems:

  • A lot of linking problems.
  • Client-Application IPC.
  • ...
slide-18
SLIDE 18

CERN, September 2006 CERN, September 2006

18

Summary Summary

Summary. Summary.

► Context.

Context.

► CERN – Extremadura Collaboration Agreement.

CERN – Extremadura Collaboration Agreement.

► Ongoing PRC projects at Extremadura.

Ongoing PRC projects at Extremadura.

  • Participants.

Participants.

  • CAPI group project.

CAPI group project.

► Conclusions.

Conclusions.

slide-19
SLIDE 19

CERN, September 2006 CERN, September 2006

19

PRC at Extremadura – Participants PRC at Extremadura – Participants

► University (UEX) and CETA have started to work on BOINC projects.

University (UEX) and CETA have started to work on BOINC projects.

  • At the University, around 1,500 computers will be used as BOINC

At the University, around 1,500 computers will be used as BOINC clients. clients.

► Initial groups.

Initial groups.

  • ARCO (Arquitectura de Computadores y diseño lógico) (UEX) -

ARCO (Arquitectura de Computadores y diseño lógico) (UEX) - http://arco.unex.es http://arco.unex.es. .

  • GEA (Grupo de Evolución Artificial) (UEX) -

GEA (Grupo de Evolución Artificial) (UEX) - http://gea.unex.es http://gea.unex.es. .

  • CETA (Centro Extremeño de Tecnologías Avanzadas) (CIEMAT) -

CETA (Centro Extremeño de Tecnologías Avanzadas) (CIEMAT) - http://www.ciemat.es http://www.ciemat.es. .

  • CAPI (Clasificación de Patrones y Análisis de Imágenes) (UEX) -

CAPI (Clasificación de Patrones y Análisis de Imágenes) (UEX) - http://capi.unex.es http://capi.unex.es. .

  • To be extended soon by means of training courses.

To be extended soon by means of training courses.

slide-20
SLIDE 20

CERN, September 2006 CERN, September 2006

20

PRC at Extremadura – ARCO Group PRC at Extremadura – ARCO Group

RADIO NETWORK DESIGN ( RADIO NETWORK DESIGN (RND@home RND@home) )

► Objective.

Objective.

  • Determine a set of locations for radio antennas in a geographical area in

Determine a set of locations for radio antennas in a geographical area in

  • rder to offer high radio coverage using the smallest antennas number.
  • rder to offer high radio coverage using the smallest antennas number.

► Features.

Features.

  • Important problem in mobile

Important problem in mobile telecommunications (such as telecommunications (such as mobile telephony), sensor mobile telephony), sensor networks, etc. networks, etc.

  • It is an NP-hard combinatorial

It is an NP-hard combinatorial problem. problem.

► Current state:

Current state: ACTIVE in ACTIVE in http://arcoboinc.unex.es/rnd http://arcoboinc.unex.es/rnd. .

► Needed computing power:

Needed computing power:

  • Around 1,500 days in a PC.

Around 1,500 days in a PC.

slide-21
SLIDE 21

CERN, September 2006 CERN, September 2006

21

PRC at Extremadura – GEA Group PRC at Extremadura – GEA Group

LASER DYNAMICS DESKTOP GRID (LDDG) LASER DYNAMICS DESKTOP GRID (LDDG)

► Objective:

Objective: Simulation of the dynamics of laser devices using a cellular automata Simulation of the dynamics of laser devices using a cellular automata individual-based model. individual-based model.

► Main goals:

Main goals:

Understand the emergence of macroscopic behaviours in lasers (paradigmatic complex Understand the emergence of macroscopic behaviours in lasers (paradigmatic complex systems) arising from the interaction of simple microscopic components. systems) arising from the interaction of simple microscopic components.

Simulate specific optoelectronic devices of arbitrary shape. Simulate specific optoelectronic devices of arbitrary shape.

► Parametric sweeps: independent parallelism, perfect for BOINC deployment.

Parametric sweeps: independent parallelism, perfect for BOINC deployment.

► Current state: Porting 2D model to BOINC.

Current state: Porting 2D model to BOINC.

► Compute time for a complete parametric sweep, for a standalone standard PC:

Compute time for a complete parametric sweep, for a standalone standard PC:

2D model: 30 days. 2D model: 30 days.

3D model: 8 years. 3D model: 8 years.

slide-22
SLIDE 22

CERN, September 2006 CERN, September 2006

22

PRC at Extremadura – CETA-CIEMAT PRC at Extremadura – CETA-CIEMAT

► Some of the current activities of BOINC at CETA-CIEMAT.

Some of the current activities of BOINC at CETA-CIEMAT.

  • Genetic algorithms for solving the placement & routing problem of

Genetic algorithms for solving the placement & routing problem of reconfigurable Hardware (FPGA). reconfigurable Hardware (FPGA).

  • Implementation of API for the development of BOINC scientific

Implementation of API for the development of BOINC scientific applications based on Distributed Genetic Algorithms. applications based on Distributed Genetic Algorithms.

  • Management and storage of scientific results in a database.

Management and storage of scientific results in a database.

  • Infrastructure (servers, storage, backup) to lodge BOINC projects.

Infrastructure (servers, storage, backup) to lodge BOINC projects.

slide-23
SLIDE 23

CERN, September 2006 CERN, September 2006

23

PRC at Extremadura – CAPI Group PRC at Extremadura – CAPI Group

► About us.

About us.

  • I'm leading a five people group (four

I'm leading a five people group (four with PhD). with PhD).

  • Research around applications of

Research around applications of Pattern Recognition and Optimization. Pattern Recognition and Optimization.

  • Mainly with Neural Networks and

Mainly with Neural Networks and Genetic Algorithms (GA). Genetic Algorithms (GA).

  • Feature extraction with PCA and ICA.

Feature extraction with PCA and ICA.

  • Special focus on imaging problems.

Special focus on imaging problems.

slide-24
SLIDE 24

CERN, September 2006 CERN, September 2006

24

PRC at Extremadura – CAPI Group works PRC at Extremadura – CAPI Group works

► Cloud cover classification.

Cloud cover classification.

  • Source of data: Meteosat images.

Source of data: Meteosat images.

  • Best results: ICA feature selection using Genetic Algorithm in a Beowulf

Best results: ICA feature selection using Genetic Algorithm in a Beowulf cluster. cluster.

CM (Middle cloud) CL (Low cloud) L (Land) S (Sea) CH (High cloud) CV (Vertical cloud) F (Fog)

Input images Output

slide-25
SLIDE 25

CERN, September 2006 CERN, September 2006

25

PRC at Extremadura – CAPI Group works PRC at Extremadura – CAPI Group works

► Alpha spectra fitting.

Alpha spectra fitting.

  • Two steps: GA initial guessing and then Levenberg-Marquardt final fit.

Two steps: GA initial guessing and then Levenberg-Marquardt final fit.

  • Collaboration with Nuclear Physics Group of UEX.

Collaboration with Nuclear Physics Group of UEX.

Energy (MeV)

4.96 5.00 5.04 5.08 5.12 5.16

Counts

1 10 100 1000 10000 4.96 5.00 5.04 5.08 5.12 5.16

  • 4

4 ______ 240Pu

_______ 239Pu

slide-26
SLIDE 26

CERN, September 2006 CERN, September 2006

26

PRC at Extremadura – CAPI Group works PRC at Extremadura – CAPI Group works

► Bovine Livestock contour extraction.

Bovine Livestock contour extraction.

  • Contour description by means of Point Distribution Models (PDM).

Contour description by means of Point Distribution Models (PDM).

  • Potential image built using a edge neural classifier

Potential image built using a edge neural classifier. .

  • Search done with repeated GA and ASM (

Search done with repeated GA and ASM (Active Shape Models Active Shape Models). ).

  • Results at

Results at http://nernet.unex.es/cows http://nernet.unex.es/cows. .

slide-27
SLIDE 27

CERN, September 2006 CERN, September 2006

27

PRC at Extremadura – CAPI Group works PRC at Extremadura – CAPI Group works

► Diagnostic of Microcalcification clusters in Mammographies.

Diagnostic of Microcalcification clusters in Mammographies.

  • Digital Database for Screening Mammography (DDSM) is been used for

Digital Database for Screening Mammography (DDSM) is been used for training and testing training and testing. .

  • More than 4,000 classifiers trained. Analysing previous results.

More than 4,000 classifiers trained. Analysing previous results.

  • We want to improve results by optimizing the classifier with GA, using

We want to improve results by optimizing the classifier with GA, using BOINC. BOINC.

slide-28
SLIDE 28

CERN, September 2006 CERN, September 2006

28

PRC at Extremadura – CAPI Group BOINC project PRC at Extremadura – CAPI Group BOINC project

► Problem description.

Problem description.

  • We want to optimize classifier design with GA.

We want to optimize classifier design with GA.

  • Objective GA function: training of a neural classifier (

Objective GA function: training of a neural classifier (~ ~ 2 hours). 2 hours).

  • Very high computational cost (

Very high computational cost (~ 3 years with our Beowulf cluster) ~ 3 years with our Beowulf cluster). .

► Parameters to optimize.

Parameters to optimize.

  • Number of components.

Number of components.

  • Number and the selected components.

Number and the selected components.

  • Size of neural classifier.

Size of neural classifier.

Neural classifier Preprocessor

ICA Feature extraction Feature selection

Optimization with Genetic Algorithms

Input Output

slide-29
SLIDE 29

CERN, September 2006 CERN, September 2006

29

PRC at Extremadura – CAPI Group BOINC project PRC at Extremadura – CAPI Group BOINC project

► Project description.

Project description.

  • Design of a BOINC GA simulator for problems with a time-consuming

Design of a BOINC GA simulator for problems with a time-consuming

  • bjective function.
  • bjective function.
  • In a first step we'll use only BOINC.

In a first step we'll use only BOINC.

  • Problem: GA needs synchronization between generations (iterations)

Problem: GA needs synchronization between generations (iterations) and usually, a few BOINC jobs take more time in finishing. and usually, a few BOINC jobs take more time in finishing.

  • Solution: to use a cluster to compute those few jobs instead of

Solution: to use a cluster to compute those few jobs instead of BOINC (as Kenyon/Cheliotis and Sottrup/Pedersen proposed). BOINC (as Kenyon/Cheliotis and Sottrup/Pedersen proposed).

► Work status.

Work status.

  • Neural Networks application ported to BOINC.

Neural Networks application ported to BOINC.

  • Test server installed with validation system.

Test server installed with validation system.

  • Now, interfacing GA simulator with BOINC server (submission and

Now, interfacing GA simulator with BOINC server (submission and assimilation). assimilation).

  • We have applied for a grant to Junta de Extremadura.

We have applied for a grant to Junta de Extremadura.

slide-30
SLIDE 30

CERN, September 2006 CERN, September 2006

30

Summary Summary

Summary. Summary.

► Context.

Context.

► CERN – Extremadura Collaboration Agreement.

CERN – Extremadura Collaboration Agreement.

► Ongoing PRC projects at Extremadura.

Ongoing PRC projects at Extremadura.

► Conclusions.

Conclusions.

slide-31
SLIDE 31

CERN, September 2006 CERN, September 2006

31

PRC at Extremadura – Conclusions PRC at Extremadura – Conclusions

► Conclusions.

Conclusions.

  • Desktop Grid Computing with BOINC could be a good

Desktop Grid Computing with BOINC could be a good alternative for regions with modest resources. alternative for regions with modest resources.

  • Due to the Agreement between CERN and Junta de

Due to the Agreement between CERN and Junta de Extremadura, desktop grid computing is now possible in our Extremadura, desktop grid computing is now possible in our region. region.

  • At least there are four projects with BOINC to be developed

At least there are four projects with BOINC to be developed by Extremadura research community. by Extremadura research community.

slide-32
SLIDE 32

► Thanks for your attention.

Thanks for your attention.

► Questions?.

Questions?.