The main players in building up the European HPC Ecosystem are:

• the ESFRI roadmap and 35 new research infrastructures
• PRACE establishing Petaflop computing centres
• EU-supported infrastructure projects such as DEISA, EGEE, GEANT2
• Euroepan Grid Initiative, eGI
• Policy groups such as HET, ESFRI and e-IRG
• Regional activities
• National infrastructures
• International centres such as CERN, EBI and ECMWF
• User communities with HPC requirements, such as fusion and climate

PRACE has already taken the first steps and realized some fine achievements. The project is responsible for the production of the HPC part of the ESFRI roadmap, for the creation of a vision involving 15 European countries, for

bringing the scientists together, and for the creation of the Scientific Case.

As Dr. Bachem announced at ISC'08, the MoU has been signed and the project proposal submitted and approved. PRACE kicked off mid 2007.

The HPC Ecosystem has to support the upper layers of the pyramid, consisting of the HPC centres, services and European projects addressing HPC, Grid, and networking. Activities which enable an efficient usage of the upper layers are the inclusion of national HPC infrastructures, software development and the solving of scalability issues.

In terms of policy and strategy, work will be done in projects such as HET - HPC in Europe Taskforce, e-IRG, and

ESFRI. ESFRI is a strategy forum with a consulting role to the European Union. It has a wide representation of scientists in various disciplines. ESFRI has elaborated a roadmap process for major new European research infrastructures. This roadmap was published in 2006. There have been set up preparatory projects for each project. The ESFRI-list update is in process and will be released in autumn 2008.

The topics to discuss in the HPC Ecosystem involve the questions of how to collaborate between the different projects and how to involve all relevant stakeholders. The DEISA project, for example, is setting up an HPC infrastructure and services and will have a strong impact on science and research. There is already a follow-up project, DEISA 2. More information is available at www.deisa.eu

The building of a virtual European HPC centre involves the national HPC centres. The goals are to enhance the existing distributed European HPC environment to a turnkey operational infrastructure, to advance the computational sciences in Europe by supporting user communities and extreme computing projects, and to enhance the service provision by offering a complete variety of options of interaction with computational resources.

As far as the Tier0/Tier1 top layers of the HPC Ecosystem are concerned, PRACE is designing an infrastructure that will enable the operation of shared petascale European systems and enhancing performance in selected sites.

The main difference between T0 and T1 centres consists in the policy and usage models. T1 centres can evolve to T0 centres for strategic and political reasons. The T0 machines automatically degrade to the T1-level by aging.

T0 centres are leadership-class European systems that have to step into competition with the leading systems worldwide. They will be cyclically renewed.

DEISA is about 1 Petaflop/s aggregated peak performance. The project delivers the most powerful European supercomputers for the most challenging projects and aims to deliver top-level Europe-wide application enabling power. The Grand Challenge projects are being performed on a regular basis.

The Infrastructure and Services are dedicated 10 GB/s GEANT2 network interconnecting systems with transparent high speed data access. They provide unified and seamless access to European Supercomputing resources. Portals for transparent access to complex supercomputing environments have to be set up for the deployment and operation of a continental HPC eInfrastructure.

The resources should be offering an aggregated peak performance of about 1 Petaflop/s and provide the best suited supercomputing architectures.

A dedicated high speed network involves common AAA single sign on and accounting and budgetting. The global data management involves high performance remote I/O and data sharing with global file systems, high performance transfers of large data sets, a User Operational Infrastructure, a distributed common production Environment, a job management service, and a common user support and help desk.

The system operational infrastructure will include common monitoring and information systems and common system operation.

As for the global application support there will be set up supported services for work flow management and an alternative access method, called Grid middleware UNICOR DESHL. The project team also foresees science gateways to supercomputing resource, and portals, as well as job rerouting between identical architectures. There must be multiple ways to access the software layers and the work flow management.

The 2008 DEISA partners' resources will be ready at the DEISA 2 project start.

Rob Pennington from the National Center for Supercomputing at the University of Illinois at Urbana-Champaign stated that the current funding programmes are not focused on preparing applications but on funding for systems, basic HW and SW, and on funding for integrating multiple systems into a Grid. The scientific funds are separately allocated. Education and training needs to be internationally organized with a long term strategy for the human infrastructure.

The US National Science Foundation (NSF) Cyberinfrastructure strategy is concentrated on HPC, data, data analysis and visualization, virtual organisations for distribute communities, and on the learning and workforce development.

The TeraGrid timeline is outrolled as follows: first, the set-up of a distributed Terascale facility, then the development of an extended Terascale facility, followed by the Terascale expansion programme. Rob Pennington also mentioned the FY05-09 operations and science outreach and the HPC operations.

Other issues are the networking, operations and security, the software integration, the community engagement, the science gateways, and the user support co-ordination.

The Petascale time systems require Petascale science applications, Petascale engineering applications, and

other Petascale applications.

The exploitation of hardware and the applications challenges need our attention too.

The Blue Waters sustained Petascale system includes Blue Waters System Characteristics, based on IBM PERCS

with about 1 petaflops sustained performance on real applications, 200.000 cores using multicore POWER7, more than 32 gigabutes of main memory per SMP, more than 10 petabytes of user disk storage, 100 Gbps external connectivity, and based on Fortran and Co-Array Fortran.

There is a new generation of scientists and engineers. We should not forget the undergraduate education. There are materials on HPC computing applications and technologies needed to augment existing courses on computer issues.

Rob Pennington concluded that collaboration opportunities have to be considered in the areas of:

• hardware: multi-core, accelerators, storage
• software: applications tools/libraries, development environments, visualization/remote access, simulations
• applications: scalability and performance
• education and training: undergraduate, graduate and post-graduate
• scale out and grid: Grid middleware and policies, applications relevant to Grid architectures, interactions/interchanges
• facilities: design and efficiency, instrumentation
• industrial engagement: ISVs, workforce development