160 likes | 310 Views
Enhancing SegHidro/BRAMS experience through EELA. José Flávio M. V. Júnior Paulo Ricardo M. Gomes. Introduction to SegHidro. Seg urança Hídr ica Aims at improving water management Brazilian Northeast – Semi-Arid Region Provides a Grid Portal Helps researchers and decision-makers
E N D
Enhancing SegHidro/BRAMS experience through EELA José Flávio M. V. Júnior Paulo Ricardo M. Gomes
Introduction to SegHidro • Segurança Hídrica • Aims at improving water management • Brazilian Northeast – Semi-Arid Region • Provides a Grid Portal • Helps researchers and decision-makers • Better management of water resources • Coupling of hydro-meteorological models
SegHidro Portal • Web-based framework to run simulations • Coupling of computational models • Need for computational power • OurGrid middleware • Open, Free-to-Join, Cooperative • Supports Bag-of-Tasks (BoT) applications • Easy interface to use the grid
SegHidro Interface • Products/Applications
Introduction to BRAMS • Brazilian Regional Atmospheric Modeling System • Numerical prediction model • Simulates atmospheric circulations on limited areas • Based on RAMS • Targets PC Clusters under Linux • Open-Source
BRAMS – Compilation and Setup • Based on C and Fortran 90 • Compiling tests : Intel and GNU Compilers • Best option: Intel Compilers (free for non-commercial use) • Alternative: Portland Group Compilers • Necessary libraries: HDF and MPI(mpich)
BRAMS – Execution • Steps: • 1. Conversion into a format that BRAMS understands (GRIB2DP) • 2. BRAMS MAKEVFILE MODE • 3. BRAMS INITIAL MODE • MPI mode is evoked • 4. Post-Processing (RAMSPOST50) • 5. Image generation (GrADS)
BRAMS – Execution • Execution Diagram
SegHidro/BRAMS – Daily Production • BRAMS Daily forecasting datasets
SegHidro/BRAMS – Daily Production • 2 HP Clusters – 10 Nodes • Double Intel Xeon 2.4Ghz processors each node • 6 Day Forecasting ~ 9 Hours • 15 Day Forecasting ~ 23 Hours • Output data is published on a server and in a data catalog that are used by the portal
SegHidro/BRAMS – Meteorologic Portlet • Validation • Parameters • Data Availability • Data automatically downloaded, everyday, and stored on a server shared via NFS by special grid nodes • Simulation is sent to OurGrid via API • Job Description • Selects appropriate machines • Decides which mode will BRAMS run: Parallel(MPI) or Sequential(one processor)
SegHidro/EELA • Current grid infrastructure: OurGrid • Resource availability • Great for independent tasks • No direct support to tightly-coupled applications(MPI) • Meteorologic Portlet’s perfomance will depend on grid’s cluster availability • Grid community formed mainly by desktop computers, clusters generally busy running daily production • MPI support from EELA Grid Infrastructure
SegHidro/EELA • Porting steps: • Get acquainted with EELA’s gLite middleware • Explore gLite’s components, in order to optimize the run and make the new system compatible with the Portal architecture • Finally, make SegHidro’s and gLite’s API communicate
Expectations • Enhance SegHidro Portal performance • SegHidro Team will acquire useful Grid knowledge • Relationship narrowing between EELA and LSD, UFCG
Thank you! Merci! Danke! Grazie! Gracias! Obrigado!