The PRACE industrial offer work in progress Stefan

The PRACE industrial offer – work in progress Stefan Wesner – HLRS, Stephane Requena – GENCI wesner@hlrs. de

Supercomputing Drives Science and Industry through Simulation Ok but we all know that most commercial codes Aerospace cores not do not scale to several thousand thinking of hundred of thousands… Environment So why Weather/ Climatology Pollution / Ozone Hole should PRACE bother about industry? ? Ageing Society Medicine Biology Energy Plasma Physics Fuel Cells Materials/ Inf. Tech Spintronics Nano-science Automotive Finance Virtual power plant Smart Grids 2

LEADING EDGE HPC AND INDUSTRY? 3

Industrial challenges in the Oil & Gas industry: Depth Imaging roadmap Algorithm complexity 15 10 flops 3 -55 Hz 9. 5 PF 1000 Visco elastic FWI petro-elastic inversion 100 10 3 -35 Hz 900 TF elastic FWI visco elastic modeling isotropic/anisotropic FWI elastic modeling/RTM 1 0, 5 isotropic/anisotropic RTM isotropic/anisotropic modeling 0, 1 Paraxial isotropic/anisotropic imaging 3 -18 Hz 56 TF RTM Sustained performance for different frequency content over a 8 day processing duration Asymptotic approximation imaging 1995 2000 2005 2010 2015 2020 Algorithmic complexity Vs. corresponding computing power 4 HPC Power PAU (TF) Courtesy of TOTAL

Industrial challenges in the Aeronautics industry: full plane CFD simulation LES Capacity: # of Overnight Loads cases run Unsteady. R ANS 102 103 104 Available Computational Capacity [Flop/s] 21 1 Zeta (10 ) 18 1 Exa (10 ) RANS Low Speed 15 1 Peta (10 ) RANS High Speed 105 “Smart” use of HPC power: 106 • Algorithms • Data mining • knowledge 1980 HS Design 1990 2000 Data Set CFD-based LOADS & HQ 2010 2020 Aero Optimisation & CFD-CSM Capability achieved during one night batch x 106 12 1 Tera (10 ) 9 1 Giga (10 ) 2030 Full MDO CFD-based noise simulation Real time CFD based in flight simulation Courtesy AIRBUS France

Mechanical and vibratory behaviour of the fuel assemblies inside a nuclear core vessel 2003 2006 2007 Consecutive thermal fatigue event 2010 2015 The whole vessel reactor 9 fuel assemblies No experimental approach up to now Computations enable to better understand the wall thermal loading in an injection. Knowing the root causes of the event define a new design to avoid this problem. Will enable the study of side effects implied by the flow around neighbour fuel assemblies. Computation with an L. E. S. approach for turbulent modelling Refined mesh near the wall. Part of a fuel assembly 3 grid assemblies Better understanding of vibration phenomena and wear-out of the rods. Computations with smaller and smaller scales in larger and larger geometries a better understanding of physical phenomena a more effective help for decision making A better optimisation of the production (margin benefits) 106 cells 3. 1013 operations 107 cells 6. 1014 operations 108 cells 1016 operations 109 cells 3. 1017 operations 1010 cells 5. 1018 operations Fujistu VPP 5000 Cluster, IBM Power 5 IBM Blue Gene/L 1 of 4 vector processors 400 processors 20 Tflops during 1 month 2 month length computation 9 days # 1 Gb of storage # 15 Gb of storage # 200 Gb of storage # 1 Tb of storage # 10 Tb of storage 2 Gb of memory 250 Gb of memory 2, 5 Tb of memory 25 Tb of memory Pre-processing not parallelized … ibid. … Mesh generation … ibid. … IESP/Application Subgroup Scalability / Solver … ibid. … Power of the computer Pre-processing not parallelized 600 Tflops during 1 month 10 Pflops during 1 month Visualisation

PRACE AND INDUSTRY – ACTIVITIES SO FAR

PRACE PP industrial seminars (1/2) • 2 seminars organised by GAUSS and GENCI – In Sept 2008 in Amsterdam with the help of NCF – In Sept 2009 in Toulouse with the help of Airbus, Grand Toulouse and Région Midi-Pyrénées – 1 and 1. 5 day format with • PRACE presentations : strategy, on going activities and key industry WP results (WP 3, WP 4, WP 7, WP 8). • Executive users presentations about industrial needs and expectations (EDF, GM, Schering Plough, Repsol, BMW, ZKB, Free Field Tech and Prospect FS) • Parallel sessions for in depth discussions about – Business models – Network and security – Training/User support & Relations with ISVs 8

PRACE PP industrial seminars (2/2) • Feedback of the attendees – Interest on PRACE upcoming infrastructure and services • Primary : Acces for open or restricted R&D needs • Some interest for commercial activites : studies/production or real business (ISV for Saa. S business models for example) • Need to exchange and to collaborate with academia, numerics and hpc experts • Needs on training, HPC support and technological watch • Network and security issues important (access ok for industrial users, encryptation of data, redundancy and relialibilty of systems, archive, . . . ) – Major open issue in relation to ISVs : (the key point!) • Licences cost and scalability of ISV apps • Availability of commercial apps on Tier 0 systems 9

PRACE INDUSTRIAL OFFER – INITIAL IDEAS

DRAFT PRACE industrial offer – Target Groups • Industrial users at the tera/petascale level – Self-developed code – High level of expertise in using HPC systems • Industrial users relying on ISV codes – Not really interested in HPC as such – Just want shorter time to result ideally in similar (certified environment) as “at home” • ISV Vendors/Tool Vendors – Validation/development of software and supporting infrastructure at large scale (e. g. license server for thousands of nodes) • Hardware Vendors – Aim for expertise for assessment of planned products – Co-Development approaches • Hardware Customers – Consultancy about new and emerging technology and their applicability 11

DRAFT Elements of the PRACE offer • Differentiation Strategy • • Co-Development t n e i c i f f e e g h n i t l b is a t i n e s e ! d l t • Training and Consulting n c e y a k c r s a e m – Lower the barrier to exploit h m t e / t t l t s uavailable resources o y s n s e s r e h It’ t to – Support selection of most f e o appropriate system for the use rter tim Layer integration problem size (tier-x) sho – Unique resources in terms of size and architecture (tier-0 systems) – High level of expertise in emerging technologies / Technology Watch (e. g. WP 9 prototypes) – Consultancy in Porting/Optimisation – Collaboration approaches aiming to shorten time to market in Software, Hardware, Solutions/Architecture – Unification of access models, application procedures across several systems across tier 0/tier-1 – Training of trainers 12

DRAFT Industrial users at tera-/petascale level • USP: PRACE offers at all time competitive HPC systems at very high capabilities. Development can start before commodity phase • Potential type of offer: – Phase 0: Pre-competitive research/Feasibility study (non-profit) • In early stages research consortia that might involve also industrial users are investigating new methods (mathematical, numerical, …) • Similar to purely academic consortia – Phase 1: Product development (profit orientation) • Development of a product is a commercial activity (=computing cannot be free but maybe subsidized if product has social impact) – Phase 2: Product use (profit realisation) • Necessary computing power might be commodity at his stage • PRACE resources offered at market rate 13

DRAFT Industrial users at tera-/petascale level • Caveat: – If there is a real business demand for petascale why should they not buy a system themselves? • PRACE offer is not just “computing cycles”, it is more about offering a complete package (support, training, optimisation, porting and leading edge systems) • Variety of systems is more a problem than a feature in this case – Are the requirements of the large scale industrial user in conflict with the needs of tier-0 research projects? • Service Level Agreements might lead to bad “Quality of Experience” for non-commercial users – Commercial Services require a special infrastructure in terms of • Accounting/billing to be delivered by the hosting site • Legal framework allowing to perform commercial activities 14

DRAFT Industrial users relying on ISV codes • USP: – The variety of different resources offered by PRACE allow selection of best performing architecture for given ISV code – PRACE unified access and use policies across the centres NO allow utilisation of different resources with reduced hassle • Potential type of offer: – Phase 2: Product use (profit realisation) • Necessary computing power might be commodity at his stage • PRACE resources offered at market rate • Caveat – Real need for tier-0 system? Is it just about cycles? – Significant implication on operation (security etc. ) – No product differentiation but only price differentiation 15

DRAFT ISV Vendors / Tool vendors • USP: PRACE offers at all time competitive HPC systems at very high capabilities. Development can start before commodity phase • Potential type of offer: – Phase 0: Pre-competitive research/Feasibility study (non-profit) • In early stages research consortia that might involve also industrial users are investigating new methods (mathematical, numerical, …) – Phase 1: Product development (profit orientation) • Findings of phase 0 are integrated in ISV code/ Tool (could be free access if results are available under favourable conditions later) – Phase 2: Product use (profit realisation) • ISV code / Tools are offered on PRACE resources with dynamic license models (e. g. on demand tokens bought by users directly at the vendor) 16

DRAFT ISV Vendors / Tool vendors • Caveat – Why should an ISV/Tool vendor collaborate? Is there a market for 100. 000 core ISV use? – How does a “reasonable” price for 100. 000 cores jobs impact the major market segment of 128 cores? – “Dynamic licensing” is solved technically but is the “pay-per-use” model a business proposition for ISVs? – Is the market large enough to support “exotic” hardware for an ISV (if not are they willing to allow others to do this for them? ) 17

DRAFT Hardware Vendors Hardware Customers • USP: Early feedback on future products, co-development with experts all over Europe • Existing offer (PRACE prototypes) • USP: Via the PRACE technology watch in-depth overview on current capabilities of large number of products • Potential Offer: – Validate and jointly develop new concepts, solutions for next generation tier-0 systems – Access to tier-0 systems for scalability tests and benchmarking activities – Analysis of requirements and proposal on selection criteria, potential technology candidates – Access to prototypes allowing the execution of tests 18

Conclusions • • PRACE 1 IP has just started to define a potential PRACE industrial offer and a corresponding business model Initial target groups have been identified based on the industrial events from PRACE-PP Initial findings clearly propose that a pure product based business modell (just cycles) is inappropriate and that a more productservice system (PSS) based approach is needed Major open issues – Incentives for ISV/Tool/Hardware vendor to collaborate? – Differentiation strategy of PRACE compared to in-house or other resource providers? – Potential barrier of the legal form of the PRACE AIBSL under belgian law and “commercial” activities – How to establish a trust relationship to an “abstract” entity such as PRACE AIBSL? Relationships to industrial users are fragile! 19

Dr. Stefan Wesner Deputy Director, HLRS PRACE Task Leader on the PRACE industrial offer wesner@hlrs. de THANK YOU FOR YOUR ATTENTION 20

PRACE 1 IP WP 5 : Industrial User Relations New workpackage / PRACE PP 10 partners from 8 countries involved WP 5 goals and objectives ü Establish continuous relations with industrial users of HPC ü Develop a proposal for a PRACE offer for industry WP 5 deliverables • Task 5. 1 Code Title Date D 5. 1. 1 First Industrial Seminar M 10 D 5. 1. 2 Requirements of industrial users M 18 D 5. 1. 3 Second Industrial Seminar M 18 Code Title Date D 5. 2. 1 Business models for industrial access to infrastructures M 10 D 5. 2. 2 PRACE industrial offer assessment M 23 21 • Task 5. 2

Task 5. 1 - Relations with industrial users of HPC Leader : BSC, co leader : GENCI • Establish a permanent link with industrial users – Organisation of 2 industrial seminars during the project • One in end of March 2011 organised by KTH/SNIC in Stockholm (Sweden) • The other around dec 2011 organised by BSC and CINECA in Italy – Participation in the setup of the PRACE user forum – Inputs for the PRACE training portal • Finalise the collect of industry needs and expectations – Industrial seminars & dedicated face to face meetings with industrial users – Selection of industrial pilots for using the PRACE RI facilities – Participate to the refresh of the Scientific Case 22

Task 5. 2 - Definition of the PRACE industrial offer Leader : HLRS, co leader : CINECA • Analysis on existing European infrastructures or existing Tier 1 and in USA, Japan and China of HPC usage by industry – Business models (commercial activities and/or free of charge precompetitive and collaborative R&D) and integrated services proposed – Impact on security, administration, peer review procedures, … • Elaboration of a proposal for a PRACE industrial offer – Based on previous analysis – Will include results from task 5. 1 and feedback of industrial pilots – Will need to address legal aspects wrt PRACE RI statuses 23

PRACE PP industrial seminars (add-on) • These 2 seminars attracted – More than 157 executive attendees, – Representing 74 industrial companies from more than 10 differents scientific domains, – Coming from 18 different countries. 24