1 / 11

parXXL : A Fine Grained Development Environment on Coarse Grained Architectures

Jens Gustedt - Stéphane Vialle - Amelia De Vivo. GdX - Grid eXplorer. parXXL : A Fine Grained Development Environment on Coarse Grained Architectures. PARA 2006 – UME Ǻ. IO data & device control. Minimize (T dev +∑T exec ). ➔. Run very large simul. ➔. Speedup fine grained

Télécharger la présentation

parXXL : A Fine Grained Development Environment on Coarse Grained Architectures

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Jens Gustedt - Stéphane Vialle - Amelia De Vivo GdX - Grid eXplorer parXXL: A Fine Grained Development Environment on Coarse Grained Architectures PARA 2006 – UMEǺ

  2. IO data & device control Minimize (Tdev+∑Texec) ➔ Run very large simul. ➔ Speedup fine grained simulations. ➔ Software suite parXXL: par::cellnet par::cell par::cntrl par::mem par::bench par::sys par::cpp ParCeL-6 2 previous environments SSCRAP parXXL: A Fine Grained Development Environmenton Coarse Grained Architectures J. Gustedt , S. Vialle, A. De Vivo parXXL: Project overview Distributed server: fine grained simulations on coarse grained architecture Client machine: comfortable fine graineddevelopment environment

  3. 32 1 1 1 1 1 1 1 1 • BSP relaxed • super-steps • Dynamic cell net • evolution • … Cell net computation Cell net creation Cell comm. Cell net evolution parXXL: A Fine Grained Development Environmenton Coarse Grained Architectures J. Gustedt , S. Vialle, A. De Vivo parXXL: Programming model Coarse grained parallelism Fine grained parallelism parXXL

  4. High performances PRO model Handle huge data High portability par::sys par::cntrl par::mem::chunk 2 optimized runtimes Relaxed synchronization Encapsulation of data: memory or file POSIX standard + MPI BSP model Threads parXXL: A Fine Grained Development Environmenton Coarse Grained Architectures J. Gustedt , S. Vialle, A. De Vivo parXXL: BSP level (SSCRAP) SSCRAP: Programming environment

  5. User code //Init function … … … //Term function … … … //Iter function … … … parXXL: A Fine Grained Development Environmenton Coarse Grained Architectures J. Gustedt , S. Vialle, A. De Vivo parXXL: Cellular level (ParCeL-6) Cell Parameters Cell Variables A cell: Cell output channels Cell input channels Cell external IO mechanisms Cyclic running Dynamic cell net evolution 3 kinds of output channels

  6. 1PE parXXL: A Fine Grained Development Environmenton Coarse Grained Architectures J. Gustedt , S. Vialle, A. De Vivo parXXL: Optimized cell nets Basic cell net creations: • Basic user code: • Automatic load balancing • Neighbor cells are not grouped 1 PE 1 PE Main function 1 PE Ex: Cubic cell net Optimized cell nets (par::cellnet) • Optimized user/par::cellnet code: • Explicit load balancing • Neighbor cells are grouped

  7. Cell behavior functions Cell net creation Loop: Cell net computation Cell communications Cell net update Cell net delete User code including par::cell operations User code & optimized par::cellnet code parXXL: A Fine Grained Development Environmenton Coarse Grained Architectures J. Gustedt , S. Vialle, A. De Vivo parXXL: Process deployment parXXL BSP level: par::cntrl par::mem par::bench par::sys par::cpp Worker TCP-Client Worker TCP-Server Worker Worker

  8. Exec time per cell & per cycle (s) Number of cells parXXL: A Fine Grained Development Environmenton Coarse Grained Architectures J. Gustedt , S. Vialle, A. De Vivo Relaxation perfs. on GdX machine 1 – Speed up fine grained simulations using more processors 2 – Extensibility: process larger problem on more processors with similar cost for each cell iteration Scales until 420 106 Cells on 310 PEs

  9. parXXL: A Fine Grained Development Environmenton Coarse Grained Architectures J. Gustedt , S. Vialle, A. De Vivo parXXL: Conclusion & Perspectives • A framework to implement fine grained applications on coarse • grained architectures • January 2006: scales up to 310 PEs (relaxation benchmark) • Optimized runtimes for shm mainframes & for clusters Future: • Implement global IO functions & Hybrid cell communications • Extend the cellular network library (par::cellnet) • Experiment on Grid5000 (French national Grid) • Application to optic & hot plasma simulations (ITER project) Available on http://gforge.inria.fr/ before september 2006

  10. parXXL: A Fine Grained Development Environment on Coarse Grained Architectures Questions?

More Related