European distributed supercomputing infrastructure is being born
In one of the most important moves to bring together national supercomputing infrastructures to advance science and technology in Europe, several leading European HPC centres devised an innovative strategy to build a terascale supercomputing facility with continental scope, called Distributed European Infrastructure for Supercomputing Applications (DEISA).
Led by IDRIS-CNRS (France) the DEISA project started its activities in May 2004 with eight partners: FZJ and RZG in Germany, CINECA in Italy, EPCC and ECMWF in the UK, CSC in Finland and SARA in the Netherlands. The project is partially funded by the European Commission under the 6ththFramework Program, as part of a vigorous initiative aimed at deploying grid enabled, production quality research infrastructures in Europe.
The main objective of the DEISA project is to enable scientific discovery across a broad spectrum of science and technology, by the deployment and operation of a world class, persistent, production quality, distributed supercomputing environment. This becomes possible through a deep integration of existing national high-end platforms, tightly coupled by a dedicated network and supported by innovative system and grid software. Strategies of coordinated operation have been identified and agreed, which will make the integrated infrastructure superior to the sum of its parts.
This integrated supercomputing power is intended to bring a boost in competitiveness for Europe in scientific areas where extreme performance is needed. The provision of high performance computing resources to researchers has traditionally been the objective and mission of the national HPC centres in Europe. However, the increasing global competition between Europe, USA, and Japan is inducing growing demands for computational resources at the highest performance levels, as well as a need of fast innovation. To stay competitive, major investments are needed every two years – an innovation cycle that is difficult to follow even for the most prosperous countries.
The architecture of the DEISA supercomputing environment has been designed following a number of strategic requirements: the necessity of operating in a transparent and non-disruptive way as a layer on top on the existing national services, the desire to hide complex grid technologies from the end scientific users, and the need to guarantee persistence and portability of scientific applications, since they are an essential part of the corporate wealth of research organizations.
The DEISA supercomputing infrastructure is constituted of two layers. At the innermost layer, similar computing platforms (same architecture and operating system) are glued together to create a “distributed virtual supercomputer”. The resulting platform is a super-cluster of computing nodes which are located in a few places in different countries, but which appears to end users as a single unified system. In the first phase of the project, four IBM supercomputers in Germany (FZJ and RZG), France (IDRIS) and Italy (CINECA) are being integrated in this way. The resulting system will consist of more than 4000 processors, a huge memory space and an aggregate computing power of over 22 teraflops. In the second phase, other IBM systems (in particular, the Finnish system) will be added to the super-cluster.
The key integration technology for the distributed super-cluster (in addition to the network itself) is the capability to efficiently share data across a wide area network, provided by a global file system (in this case, IBM’s Global Parallel File System, GPFS). The high added value of this integrated platform arises from the possibility of redistributing the computational workload by migrating jobs across national borders, in order to free huge resources for one specific application in one site.
At the next layer of the DEISA infrastructure, this IBM super-cluster is federated with other selected leading computing platforms to constitute a heterogeneous supercomputing grid, which will include vector platforms and Linux clusters. The first platform to be integrated in the grid is SARA’s SGI ALTIX supercomputer, providing 416 Itanium processors. The DEISA heterogeneous grid will provide a number of relevant services to the scientific community: workflow management based on UNICORE middleware (complex applications that visit several platforms to perform a job), high performance global data management in the whole grid (sharing data between different applications, applications accessing distributed data), grid applications that run on several platforms simultaneously, and, last but not least, portals and Web interfaces to hide complex environments from end users.
The DEISA infrastructure fully exploits the network bandwidth provided by the European research network GEANT and the national research networks (today DFN in Germany, RENATER in France, and GARR in Italy). It also depends critically on the aggressive evolution planned for these and other European organizations, to enhance the network performance. “The DEISA concept is based on the educated guess that network bandwidth will become, by the ends of this decade, a commodity very much like raw computing power became a commodity in the early 90’s”, said Prof. Victor Alessandrini from IDRIS-CNRS, director of the project. “A tightly integrated European supercomputing environment is mandatory to share the extreme computational resources that are needed for extreme efficiency and performance. This is the road that is being paved by DEISA”.
DEISA can expand horizontally by adding new systems, new architectures, and new partners thus increasing the capabilities and attractiveness of the infrastructure in a non-disruptive way. Recently, agreement has been reached with three leading organizations in Europe (HLRS and LRZ in Germany, and BSC, the new supercomputing centre in Barcelona, Spain) to join the DEISA Consortium, and detailed negotiations are under way. With this enlargement, practically all the leading computing platforms in Europe would be in the DEISA grid. Moreover, DEISA will be open to collaboration with other Europe HPC centres and related initiatives world-wide, like TeraGrid in the USA or the other major European research infrastructure project, EGEE, led by CERN.
In summary, DEISA is focused on the advancement of science in Europe. DEISA collaborates with leading European research groups initially from several scientific and industrial disciplines (Material Sciences, Cosmology, Fusion Research, Life Sciences, Computational Fluid Dynamics, and Environmental Sciences). The researchers aim to demonstrate that DEISA will enable new research results and enhanced scientific output in several different ways. As the infrastructure consolidates, DEISA will be launching aggressive scientific initiatives to enable new, ground breaking research activities that could not come to life otherwise.
Jari Jarvinen | alfa