Presentation of GRID-TLSE Managing Services Comments on Data - PowerPoint PPT Presentation

Presentation of GRID-TLSE General Overview Software Architecture Main Resources Managing Scenarios Presentation of GRID-TLSE Managing Services Comments on Data management in GRID TLSE (prospective) Comments on data management http://www.enseeiht.fr/lima/tlse ACI GDS Meeting, May 20th, 2005

Outline Presentation of GRID-TLSE General Overview Software Architecture General Overview Main Resources Managing Scenarios Software Architecture Managing Services Comments on Data Main Resources management in GRID TLSE (prospective) Comments on data Managing Scenarios management Managing Services Comments on Data management in GRID TLSE (prospective) Comments on data management

GRID-TLSE Project Presentation of GRID-TLSE T ests for L arge S ystems of E quations General Overview Software Architecture Main Resources Main purpose: Sparse linear algebra Web expert site. Managing Scenarios Funding: ACI GRID, 01/03 – 01/06. Managing Services Comments on Data management in GRID Partners: TLSE (prospective) ◮ Academic partners: CERFACS, ENSEEIHT-IRIT, Comments on data management LaBRI, LIP-ENSL; ◮ Industrial partners: CNES, CEA, EADS, EDF, IFP; ◮ International links: LBNL-Berkeley, Parallab-Bergen, Univ. of Florida, RAL, Old Dominion Univ., Univ. of Minnesota, Univ. of Tennessee, Univ. of San Diego, Indiana Univ., Tel-Aviv Univ.

Expertise for Sparse Matrices: Motivations Presentation of GRID-TLSE General Overview Goal: Provide a friendly test environment for expert and Software Architecture non-expert users of sparse linear algebra software. Main Resources Managing Scenarios Managing Services Easy access to: Comments on Data ◮ Software and tools: public... as well as commercial, management in GRID TLSE (prospective) sequential... as well as parallel; Comments on data management ◮ A wide range of computer architectures; ◮ Matrix collections. Goal (bis): Provide a testbed for sparse linear algebra software developers. Scope of TLSE: focus on direct methods for sparse matrices

Why Using a Grid ? Presentation of GRID-TLSE General Overview Software Architecture ◮ Sparse linear algebra software makes use of Main Resources Managing Scenarios sophisticated algorithms for (pre-/post-) Managing Services processing/solving a sparse system Ax = b . Comments on Data ◮ Multiple parameters interfere for efficient execution management in GRID TLSE (prospective) of a sparse solver: Comments on data ◮ Ordering; management ◮ Amount of memory; ◮ Architecture of computer; ◮ Libraries available. ◮ Determining the best combination of parameter values is a multi-parametric problem.

Why Using a Grid ? Presentation of GRID-TLSE General Overview Software Architecture ◮ Sparse linear algebra software makes use of Main Resources Managing Scenarios sophisticated algorithms for (pre-/post-) Managing Services processing/solving a sparse system Ax = b . Comments on Data ◮ Multiple parameters interfere for efficient execution management in GRID TLSE (prospective) of a sparse solver: Comments on data ◮ Ordering; management ◮ Amount of memory; ◮ Architecture of computer; ◮ Libraries available. ◮ Determining the best combination of parameter values is a multi-parametric problem. ◮ Well-suited for execution over a Grid.

Main Components of the Site Presentation of GRID-TLSE General Overview Software Architecture Main Resources ◮ Sparse matrix software: direct solvers. Managing Scenarios Managing Services ◮ Database: matrices, scenarios, bibliography, Comments on Data experimental results. management in GRID TLSE (prospective) ◮ High-level administrator interface for the definition, Comments on data the deployment, and the exploitation of services over management a Grid: Weaver. ◮ Interactive Web interface with the Grid: WebSolve. ◮ Use of tools developed within GRID-ASP project (LIP-ReMAP, LORIA-R´ es´ edas, LIFC-SDRP): DIET.

Examples of Requests (scenarios) Presentation of GRID-TLSE General Overview Software Architecture ◮ Memory required to factor a matrix, with which Main Resources algorithm/solver/input parameters ? Managing Scenarios Managing Services Comments on Data ◮ Error analysis as a function of the threshold pivoting management in GRID TLSE (prospective) value. Comments on data management ◮ Minimum time on a given computer to factor a given unsymmetric matrix. (naive or more elaborated scenario) ◮ Which ordering heuristic is the best one for solving a given problem?

Scenario examples: Ordering sensitivity Presentation of GRID-TLSE General Overview Software Architecture ◮ Phase 1: Get orderings (permutations): Main Resources Managing Scenarios - one solver: get all of its internal orderings. Managing Services - more than one solver: get all possible orderings from Comments on Data all solvers. management in GRID TLSE (prospective) ◮ Phase 2: Obtain value of required metrics for each Comments on data ordering: management - for metrics of type estimation, the analysis is performed for each required solver. - for metrics of type effective, the factorization is also performed. ◮ Phase 3: Report metrics for all combinations of solvers/orderings

Scenario examples: Minimum time Presentation of GRID-TLSE General Overview Software Architecture Main Resources Managing Scenarios ◮ Phase 1: Get orderings from all solvers. Managing Services ◮ Phase 2: For each ordering and requested solver Comments on Data management in GRID TLSE (prospective) - perform Flops estimation - keep best ordering per solver. Comments on data management ◮ Phase 3: For each solver: - factorize with BOTH selected ordering and internal default ordering - report statistics with minimum time.

Software Architecture Presentation of GRID-TLSE General Overview Software Architecture Client Main Resources Managing Scenarios Consult Expert Site : Grid−TLSE Managing Services Modify Connection Comments on Data management in GRID WebSolve Expert TLSE (prospective) Database Expertise Synthetic Comments on data Request Results management Consult/Modify Scenarios Matrix History Bibliography Collect. Services Logfiles Weaver Static Dynamic Partial ( RAL−BOEING / Parasol ) Solver Results Runs t s a S t MIDDLEWARE : FAST + DIET Matrix provided by the client Results Services Grid Solvers

Expertise Run Presentation of GRID-TLSE General Overview GRID TLSE User Software Architecture Main Resources WEBSOLVE Managing Scenarios Managing Services Expertise Request WEAVER Comments on Data management in GRID TLSE (prospective) Scenarii Comments on data Services management DIET ✄✁✄✁✄ ☎✁☎✁☎ ✠✁✠✁✠✁✠ ✡✁✡✁✡ ✆✁✆✁✆ ✝✁✝✁✝ ✞✁✞✁✞ ✟✁✟✁✟ SOLVERS ✂✁✂✁✂ �✁�✁� ☛✁☛✁☛ ☞✁☞✁☞

Expertise Run Presentation of GRID-TLSE General Overview GRID TLSE User Software Architecture Main Resources WEBSOLVE Managing Scenarios Managing Services Expertise Request WEAVER Comments on Data management in GRID TLSE (prospective) Scenarii Scenarii Step Comments on data Services management DIET ☎✁☎✁☎ ✄✁✄✁✄ ✠✁✠✁✠✁✠ ✡✁✡✁✡ ✆✁✆✁✆ ✝✁✝✁✝ ✟✁✟✁✟ ✞✁✞✁✞ SOLVERS ✂✁✂✁✂ �✁�✁� ☛✁☛✁☛ ☞✁☞✁☞

Expertise Run Presentation of GRID-TLSE General Overview GRID TLSE User Software Architecture Main Resources WEBSOLVE Managing Scenarios Managing Services Expertise Request WEAVER Comments on Data management in GRID TLSE (prospective) Scenarii Scenarii Step Comments on data Services management Solver Run DIET ✄✁✄✁✄ ☎✁☎✁☎ ✠✁✠✁✠✁✠ ✡✁✡✁✡ ✆✁✆✁✆ ✝✁✝✁✝ ✟✁✟✁✟ ✞✁✞✁✞ SOLVERS ✂✁✂✁✂ �✁�✁� ☛✁☛✁☛ ☞✁☞✁☞

Expertise Run Presentation of GRID-TLSE General Overview GRID TLSE User Software Architecture Main Resources WEBSOLVE Managing Scenarios Managing Services Expertise Request WEAVER Comments on Data management in GRID TLSE (prospective) Scenarii Scenarii Step Comments on data Services management Solver Run DIET ✄✁✄✁✄ ☎✁☎✁☎ ✠✁✠✁✠✁✠ ✡✁✡✁✡ ✆✁✆✁✆ ✝✁✝✁✝ ✟✁✟✁✟ ✞✁✞✁✞ SOLVERS ✂✁✂✁✂ �✁�✁� ☛✁☛✁☛ ☞✁☞✁☞

Expertise Run Presentation of GRID-TLSE General Overview GRID TLSE User Software Architecture Main Resources WEBSOLVE Managing Scenarios Managing Services Expertise Request WEAVER Comments on Data management in GRID TLSE (prospective) Scenarii Scenarii Step Comments on data Services management Solver Run DIET ✄✁✄✁✄ ☎✁☎✁☎ ✠✁✠✁✠✁✠ ✡✁✡✁✡ ✆✁✆✁✆ ✝✁✝✁✝ ✟✁✟✁✟ ✞✁✞✁✞ SOLVERS ✂✁✂✁✂ �✁�✁� ☛✁☛✁☛ ☞✁☞✁☞