PRACE - Partnership for Advanced Computing in Europe Key Performance Indicators. Philippe Segers GENCI (on behalf of PRACE aisbl)

Transcription

1 PRACE - Partnership for Advanced Computing in Europe Key Performance Indicators Philippe Segers GENCI (on behalf of PRACE aisbl) GSF-OECD workshop 3 rd November 2015, Paris

2 Partnership for Advanced Computing in Europe PRACE is an international not-for-profit association under Belgian law, with its seat in Brussels, born in 2010 PRACE counts 25 members and 2 observers The PRACE Hosting Members are France, Germany, Italy and Spain PRACE is governed by the PRACE Council in which each member has a seat. The daily management of the association is delegated to the Board of Directors PRACE is funded by its members as well as through a series of implementation projects supported by the European Commission 03/11/2015 GSF-OECD workshop: PRACE KPIs 2

3 4 Hosting Members offering core hours on 6 world-class machines MareNostrum: IBM BSC, Barcelona, Spain Universe Sciences 18% Mathematics and Computer Sciences 6% Fundamental Physics 17% Engineering and Energy 21% Chemical Sciences and Materials 29% Earth System Sciences 9% JUQUEEN: IBM BlueGene/Q GAUSS/FZJ Jülich, Germany Hazel Hen: Cray GAUSS/HLRS, Stuttgart, Germany SuperMUC: IBM GAUSS/LRZ Garching, Germany Access through PRACE Peer Review Criterion: Scientific Excellence CURIE: Bull Bullx GENCI/CEA Bruyères-le-Châtel, France FERMI: IBM BlueGene/Q CINECA, Bologna, Italy GSF-OECD workshop: PRACE KPIs 3

4 First KPIs considered Success ratio of proposals H-index of applicants Scientific Economic Social Environmental Short-term Medium-term Long-term 03/11/2015 GSF-OECD workshop: PRACE KPIs Resource allocation Distribution per job size and duration Training events Publications, PhD theses Project finance structure, co-funding Financial performance of PRACE members PRACE raising awareness European end-users company in HPC area Industry participation in PRACE events HPC related job trend Patents and spin-offs Ecological impact 4

5 Currently implemented KPIs Impact on evolving research Offer and demand Number of projects requested (blue), above technical threshold (red) and awarded (green) International collaboration Ratios of awarded foreign projects (blue), and resources for awarded foreign projects (red) and respective trend-lines 03/11/2015 GSF-OECD workshop: PRACE KPIs 5

6 Currently implemented KPIs Impact on scientific production Publications and thesis Impact on growing know-how Training events Scientific production supported by PRACE: publications (blue), scientific talks (red) and thesis (green) Number of person-days registered at PRACE Training days between 2008 and /11/2015 GSF-OECD workshop: PRACE KPIs 6

7 Currently implemented KPIs Impact on attracting the industrial sector Training Industrial participation in PATCs training days Allocations Industry participation in PRACE allocations: number of projects with industrial participation (blue) and number of industrial participants (red) 03/11/2015 GSF-OECD workshop: PRACE KPIs 7

8 Data collection Database from the peer-review tool: Compiles all information on the calls (resources awarded, machines, project participants, collaborators affiliation, industry partners, etc.) Surveys to projects leaders: Final report at the end of the allocation period Follow-up 2 years after Spreadsheet for training events Media coverage 03/11/2015 GSF-OECD workshop: PRACE KPIs 8

9 Limitations and specificities Time-frame between computation and scientific/economic exploitation: Need to stay informed about publications several years after the allocation is done Long-term impact on competitiveness of industry? Complex infrastructure: Distributed Virtual access Shared by multiple scientific communities and type of users A difficult multi-factors impact assessment: Broad range of actors in the ecosystem: scientists, industrials, vendors, computing centers Qualitative (success stories, etc.) more than quantitative How to assess the overall increase in HPC adoption? Impact of HPC in European competitiveness embedded in the whole value chain Lack of consistent economic and scientific data 03/11/2015 GSF-OECD workshop: PRACE KPIs 9

10 Limitations and specificities Internal limitations: KPIs defined after the infrastructure Historical data not structured to facilitate impact assessment Not always possible to know usage at the time of data collection Limited resources Some manual processing Two levels of impact assessment: At a European level, for the PRACE infrastructure and HPC in general At a national level, for hosting a system or participating in the infrastructure 03/11/2015 GSF-OECD workshop: PRACE KPIs 10

11 Conclusion PRACE: a distributed e-infrastructure devoted to serve its users PRACE Key Performance Indicators: A tool for continuous improvement process Designed to asses the fulfilment of its mission Covering a wide spread of interactions with its stakeholders 03/11/2015 GSF-OECD workshop: PRACE KPIs 11

12 12

13 HPC strategy combining three elements: (a) (b) (c) Computer Science: towards exascale HPC; A special FET initiative focussing on the next generations of exascale computing technology as a key horizontal enabler for advanced modelling, simulation and big-data applications [HPC in FET] achieving excellence in HPC applications; Centres of Excellence for scientific/industrial HPC applications in (new) domains that are most important for Europe [e-infrastructures] providing access to the best supercomputing facilities and services for both industry and academia; PRACE - world-class HPC infrastructure for the best research [e-infrastructures] complemented with training, education and skills development in HPC

14 "Excellent Science" part of H2020 Access to best HPC for industry and academia - PRACE specifications of exascale prototypes technological options for future systems FETHPC: EU development of Exascale technologies Collaboration of HPC Centres and application CoEs provision of HPC capabilities and expertise Excellence in HPC applications (Centres of Excellence) identify applications for codesign of exascale systems Innovative methods and algorithms for extreme parallelism of traditional/emerging applications