by Marco Vanneschi
PQE 2000 is an Italian national project, of high strategic technological and commercial value, in the area of massively parallel general purpose supercomputing. It is one of the most significant technology transfer activities of CNR in the framework of a cooperation with Alenia Spazio - Finmeccanica. The other major Italian research agencies, ENEA and INFN, and Italian Universities also participate in the initiative.
CNR Project for General Purpose Parallel Architectures
The Subproject on Parallel Architectures (part of the CNR Project on Informatics and Parallel Processing, 1989-1994) was focused on general purpose massively parallel systems (see Ercim News, No.19, p.25).The results included design methodologies, prototyping and evaluation of new physical and abstract parallel architectures, of efficient and machine independent programming tools for several languages and paradigms, and of highly parallel operating systems and run-time supports.
The main goal has been to give a concrete, advanced contribution to the national industry and user community, and to stimulate the transfer of results for the development of next generation parallel computer systems.There have been significant activities for technology and know-how transfer in the area of parallel computer systems and their applications. Of special importance is the PQE 2000 Project.
Objectives of PQE 2000
The general framework of PQE 2000 calls for a coordination of interests and objectives throughout Europe, not just in Italy, as the project is characterised by a strong industrial target "upfront".
The overall goal is to realize, at the industrial level, a massively parallel system for general-purpose high-performance computing, characterized by innovative solutions at both hardware and software levels, and technologically competitive in the market of supercomputing systems and applications. The project also includes a wide range of significant applications in the various fields of science and technology, finance, business and public administration, as well as in high performance embedded and real time systems.
To achieve such ambitious goals, a strong commitment and investment from high tech industry is crucial. This condition is satisified by the leading role played by Alenia Spazio and Finmeccanica in the project.
Basic technologies for PQE 2000
The technologies on which PQE 2000 is based, and which will be integrated to create a new technology for the TeraFlops and PetaFlops supercomputing market, are mainly:
A collaboration between PiNa4 (IRSIP - CNR, Naples, and theDepartment of Computer Science, University of Pisa) and Meiko resulted in a proposal for a new hardware-software architecture for general purpose massively parallel machines, that will be now integrated with the Quadrics technology to give rise to the PQE 2000 Project. PQE stands for Pina Quadrics Europe.
Starting from these basic technologies, efforts will be concentrated on developing new technologies for PQE 2000, ranging from evolutionary VLSI components for processing units and I/O, to very advanced software tools for the optimisation, portability and restructuring of parallel programs.
A strategic goal is to establish collaborations with selected partners in order to develop application libraries, application software and benchmarks. The European industrial community and supercomputer users should not only greatly benefit from the availability of the PQE 2000 platform but, hopefully, will be willing to support the Project in an active way.
Overview of the PQE 2000 characteristics
Massive parallelism is considered the key for achieving the required performance values: TeraFlops for the beginning of the next century, while the PetaFlops generation is foreseen for 2010-2015. The PQE 2000 solution to the "generality + performance" problem consists in the integration of general purpose, medium/coarse grain MIMD machines with fine grain SIMD/PIM machines. In the short/medium term, this will be achieved through the suitable integration of the Meiko/PiNa4 technology with the Quadrics technology. In this way the most innovative approach possible is adopted, compatibly with the requirement for a short time-to-market strategy and with a competitive, continuous progress from the TeraFlops to the PetaFlops generation. The PQE 2000 layer architecture is shown in the figure.
The other key issue in the project is the development of an advanced software technology. In PQE 2000, as in any other massively parallel computer of the next generation, the development of parallel programs should be fully machine-independent, and the task of optimising the usage of the machine components should be delegated to powerful compiling tools. Portability and reusability are fundamental requirements.
Recently, models that "restrict"the forms in which computations may be expressed have been recognised as the most promising for the massively parallel systems of the next generation. High Performance Fortran (HPF) and other models with data parallel extensions belong to this approach. More advanced"restricted" models belong to the structured parallel programming class, also called template-based programming. P3L (Pisa Parallel Programming Language) is an example of this approach, which has been experimented successfully in a significant number of different applications within the CNR Subproject on Parallel Architectures.
In PQE 2000: