Partage d'expériences industrielles : calcul intensif, SaaS, Cloud, etc. - Maximilien Landrain

Gérer vos applications et vos infrastructures « as-a-Service »
Présentation de Maximilien Landrain (Senior Applications Engineer, NOESIS) lors des Rencontres SaaS, Cloud & innovation organisées par SysFera le 23 mai à Clamart.

Transcript

1. Partage d'expériences industrielles : calcul intensif, SaaS, Cloud, etc. Maximilien Landrain Senior Applications Engineer Rencontres SaaS, Cloud & innovation 23 mai 2012 - Clamart
2. Agenda 1 Introduction: Optimus – Noesis Solutions references HPC industrial experience 2 Research & Innovation experience 3 Conclusion: Experiences and Challenges 4
3. Agenda 1 Introduction: Optimus – Noesis Solutions references HPC industrial experience 2 Research & Innovation experience 3 Conclusion: Experiences and Challenges 4
4. Capture, Parameterize, Federate, Automate Capture any engineering process, integrate heterogeneous simulations, automate design improvement execution Open and extendable environment for PIDO Create plugins, Python scripting, interface with everything generate workflows dynamically Enabling real-life multi-disciplinary optimization Multi-level, multi-disciplinary, user intuitive extendable interface Our product: Optimus Optimus Process Integration Design Optimization Optimize Capture Deploy
5. Distributed and heterogeneous computations Virtualization and cloud computing Capture, structure and re-use Engineering Knowledge Modular SW components to ‘compose’ design flow Interface with everything: be open and extendable Solve multi-disciplinar, multi-physic engineering problems Provide formalized interoperability solutions Implement scalable automation and execution Our Research & Innovation priorities Solve Complexity Manage Heterogeneity Cloud Computing
6. Optimus Industry References BMW- VW- Audi  RSM Modeling: Chassis Pre-Design  Engine combustion, Nvh, Crash, Durability, Die Casting, Powertrain …etc  Robustness Pedestrian Crash Simulation  Car Body Sensibility Analysis Safran Group  Fuel Systems, Engine Blade Design, Engine pylon  Turbo pump Balancing, cryogenic cooling  Landing Gear braking steering and landing  Power Transmission, Electronic control units JP Electronics (Nikon, Canon, Olympus, Alpine)  Optical Optimization  Multi-Body Mechanic Robustness  RSM Modeling of mechanical parts  Electronic and Electric Optimization
7. Agenda 1 Introduction: Optimus – Noesis Solutions references HPC industrial experience 2 Research & Innovation experience 3 Conclusion: Experiences and Challenges 4
8. Main Solution Elements in Optimus #1. Simulation Model • Accurate and Realistic • Physical & Functional Simulation V cycle #2: Optimus Process  Scenario Capturing Process  Link to Data (PLM & SLM)  Drive Designs Changes in Parametric models  Store Design Space Exploration #3: Result Data Mining  Drill into exploration results  Refine models along V cycle  Balance Objectives and decide design changes  Sync with multiple disciplines
9. Optimus HPC Strategies • Workflow level • Experiment level • Workflow + Experiment level time EH1 Exp 2 EH2 EH3 Exp 1 time Exp 1 on EH1 Exp 2 on EH2 time EH1 EH2 EH3 EH4  The cluster load graphs on the right refer to the execution of a 2 experiments Optimus Method for a workflow of 3 independent parallel analyses (below) Execution time of Analysis1 Execution time of Analysis2 Execution time of Analysis3
10. Aero-Meca-Acoustic Optimization Aero-Thermo-Mechanic Optimization Aero-Mechanic Optimization Aerodynamic constrained Optimization Objective and Constraints  Reduce specific consumption  Increase life time of component  Reduce engine mass  Decrease risk during exploitation Industry Case: Safran Group Optimus Industrial Development  20 Design Variables  80 Results  Up to 1024 CPUs
11. Industry Case: Audi AG Optimus Industrial Development  Without Parallelization:  TOTAL Process Simulation Time: 69 hours  Optimization requires 500 simulation runs  1,400 DAYS !!!!  With Workflow Level Parallelization:  Longest job: 22 hours on 8 CPUs  Optimization requires 500 simulation runs  With Workflow and Optimization Level Parallelization: -> 6% Weight Gain!  Per optimization iteration: 3 days 8 hours  16.6 days for complete optimization  Max gap observed : 4 hours  96 Design Variables  2 HPC clusters  800 CPUs  15 days Computation
12. Agenda 1 Introduction: Optimus – Noesis Solutions references HPC industrial experience 2 Research & Innovation experience 3 Conclusion: Experiences and Challenges 4
13. Research Case: Bridge International Cooperation on Grid Technologies IST Call 6 Distributed Workflow for Multi-objective Optimization Based on SOA and Grid
14. Research Case: SmarLM EC FWP 7 – GRID Friendly Licensing Management Rencontres SaaS, Cloud & innovation 23 mai 2012 - Clamart New Licensing Mechanism for GRIDs  Licenses as distributed services on the network  Smarter management for large distributed computing networks Build innovative licensing management and business models to exploit GRID environments New business models for GRIDs  Creation of new industrial opportunities exploiting GRID computing  Extension of Service Level Agreements  Secure commercial revenues while boosting usage on distributed networks Provide a new generic licensing virtualization framework and integrate in major Grid middleware solutions SmartLM will provide a generic and flexible licensing virtualization technology for new service-oriented business models across organization boundaries.
15. Agenda 1 Introduction: Optimus – Noesis Solutions references HPC industrial experience 2 Research & Innovation experience 3 Conclusion: Experiences and Challenges 4
16. HPC Experience and Challenges Optimus Platform  Provide Optimus with robust configuration with HPC RMS systems  Adapt Optimus and Simulation Software for Efficient execution  Secure HPC Distribution with private/unknown HPC/Grid/Cloud environment  Manage Software Vendor Simulation Licenses and billing from the grid/cloud  Engage new Software Vendors business models  Combine execution and storage of large datasets in SLM environment HPC RMS Platform  Provide Easy-to-Deploy HPC capability for both SME and Large Account  Secure HPC Distribution with private/unknown HPC/Grid/Cloud environment  Extend Monitoring and Alerts systems to warn end user of anomalies  Manage commercial software licenses between large users  Optimization of available resources through smart engines
17. Explore Sysfera- Noesis Solutions Joined Collaboration to address new challenges CAE Workflows Design Exploration Technologies Manage HPC Distribution & Storage
18. Thank you Questions Rencontres SaaS, Cloud & innovation 23 mai 2012 - Clamart www.noesissolutions.com