CDA team

CDA team

Evaluation Committee of LAAS-CNRS

December 1-3, 2009

Evaluation 2009 CDA

Outline

1. Elements of understanding 2. Staff 3. Domain of research 4. Major results 5. Visibility 6. Notoriety 7. Data 2005-2009 8. Project

Evaluation 2009 2CDA

1. Elements of understanding

CDA team was founded ex nihilo on October 1, 2007

with researchers from RST (disbanded September 30, 2005) and MAC.

Initial domain of interest: Scientific computing, combinatorial optimization, parallel computing.

Increase Computer Science orientation Move from Mocosy to SINC.

Specific approach multidisciplinary approach (Comp. Sc. and Appl. Maths);

distributed computing, HPC.

no other team works specifically in this domain at LAAS-CNRS

(quite a few teams in France and in the world).

Priority research as defined by CNRS and others.

Evaluation 2009 3CDA

2. Staff

3 permanent researchers

d. el baz (hdr, cnrs, head), m. elkihel (as. prof), j.m. enjalbert (as. prof.)

3 nonpermanent researchers

v. boyer (post doc), m. lalami (doct.) t.t. nguyen (doct.)

Evaluation 2009 4CDA

3. Domain of research: Dist. Comp., HPC

3.1 New trends Evolution of computer science

changes in our vision of computing.

Rise of parallelism parallelism almost everywhere, many cores, GPGPU…

Advent of the Internet Age

Internet of people and things, rise of network technologies.

Convergence of parallel and distributed computing

lack of specialized processors for high performance computing.

Evaluation 2009 5CDA

3.2. Challenges in distributed computing

3.2.1. Specific problems Lack of centre and global clock Unbounded communication delays Faults and topology changes Massive parallelism and heterogeneity

Evaluation 2009 6CDA

Grand challenges

Easiness of programming and using architectures Quest for efficient algorithms

Evaluation 2009 7CDA

4. Major results

Design and analysis of distributed algorithms

general purpose, dedicated architectures application to optimization, boundary value problems, pattern

recognition.

JCAM 2005, IMA JNA 2005, JCAM 2008.

Communication study and management PDP 2007, JPDC 2007.

Design of load balancing techniques JPDC 2005, PDP 2006.

Design of decentralized environment for HPC PDP 2005, Ter@tec 2009, Rempar 2009.

Evaluation 2009 8CDA

4.1 Distributed algorithms

Asynchronous Algorithms with Order Intervals,AAOI convergence study, stopping criteria, implementation

(LCS, N7, Wuppertal University).

Fig. 2a Data exchanges in AAOI Fig 2.b Implementation on IBM SP4up to 128 processors, convection-diffusionproblem

Evaluation 2009 9CDA

4.1 Distributed algorithms (cont.’d)

Design & analysis of distributed algorithms for state acquisition/pattern recognition on a smart surface

mathematical modelling, synchronous and asynchronous algorithms, stopping criteria, convergence study.

ANR Smart Surface (FEMTO-ST, InESS, LAAS-CNRS, LIFC, LIMMS).

Fig. 3a the Smart Surface Fig. 3b communication network Fig. 3c Smart Surface Simulator

Evaluation 2009 10CDA

4.2 Design of decentralized environment

High performance distributed peer to peer computing - self adaptive protocol, direct communications between peers;

- ANR CIP (LAAS-CNRS, LIFC, N7, U. Picardie, EMD);

- LAASnetexp, Nicta platform (Australia).

Fig. 4a Protocol architecture Fig. 4b Results on NICTA platformup to 24 machines, obstacle problem

Evaluation 2009 11CDA

5. Visibility

Cofounder of group KSO of GDR RO, 2007.

M. Hifi, University of Picardie. Cooperation with

- G. Plateau, Paris XIII,

- J.-C. Miellou, LCS Besançon,

- P. Spiteri, IRIT-N7,

- J. Bourgeois LIFC,

- A. Frommer, Wuppertal University, Germany,

- G. Jourjon, NICTA, Eveleigh, Australia.

Evaluation 2009 12CDA

6. Notoriety

Program Committee Chair and Organizing Committee Chair of PDP Toulouse, February 2008.

Program Committee Chair of PDP Weimar, February 2009.

Chair of Special Session IPIA, CEI 39, 2009, Troyes. Program Committee Chair of MSOP2P 2010, Pise. Chair of Special Session MPDPE at Roadef 2010,

Toulouse. Organizer of Tutorial NVIDIA, Toulouse February 2009. Reviews : ANR, Journals: IEEE TPDS, JPDC, Par. Comp., JSA, SIAM JNA, ETNA, Comp. Math.,EJOR,

Comp. Op. Res., EJIE, ….

Evaluation 2009 13CDA

7. Data 2005-2009 and Analysis

8 papers in International Journals. 10 papers in International Conferences. 1 Ph. D. thesis, December 2007. Coordination of ANR Project CIP 2007. ANR project Smart Surface.

Evaluation 2009 14CDA

7. Data 2005-2009 and Analysis

Better organization

- independence, motivation, interactions;

- attend events, buy machines more easily;

- take benefit of opportunities (HPC);

- work on new topics, e.g. GPU computing. Production: many more papers late 2009

- at least 5 papers on October-November 2009.

- 5 papers in preparation. Visibility - CNRS, LAAS, partners, students,

- domain of research, events organization (confs).

Evaluation 2009 CDA 15

8. Project

8.1. Short term

GPU computing

- Speedup = 15 with GTX 260 for dynamic programming applied to

knapsack problems as compared with Intel, Xeon Quadro;

- two papers on this topic at Roadef 2010;

- Iblis machine at CINES;

- 3 permanents (100%, 33%, 33%), 1 post-doc (50%);

- Contacts with NVIDIA, Vega Technologies.

Evaluation 2009 16CDA

8. Project (cont.’d)

8.2. Long TermGrand Challenges in distributed computing and HPC Easiness of programming and using architectures→ P2P computing, decentralized environments,

programming paradigms and models, application deployment;

ANR CIS CIP, ADREAM (reconfiguration);

platforms: GRID 5000.

Quest for efficient algorithms asynchronism:

convergence, stopping criteria …

self organization:

characterization, models, strategies: efficiency → everlastingness.

From D.S. to I.S. (very long term)

Evaluation 2009 17CDA