Foundations of General-Purpose Parallel Computing
Description
- Project Title:
- Foundations of General-Purpose Parallel Computing
- Acronym:
- GEPPCOM
- Number:
- 9072
- Work Area:
- Basic Aspects of Multiple Computing Systems
- Coordinator:
- Consorzio PADOVA Ricerche
Galleria degli Scrovegni, 7
I- 35121 PADOVA
- Coordinator Country:
- I
- Partners
- Universidad Politecnica de Valencia E
Computer Technology Institute GR
Istituto di Elaborazione dell'Informazione-CNR I
University College Cork IRL
Oxford University UK
- Contact Point:
- Prof. G. Bilardi
- Telephone:
- +39/49 828 7653
- Fax:
- +39/49 828 7699
- E-Mail:
- bilardi@art.dei.unipd.it
- Keywords:
- parallel computing, area-universal computing, bulk synchronous computing, parallel numerical algorithms, parallel combinatorial algorithms
- Start Date:
- to be announced
- Duration:
- months
- Status:
- starting
- Abstract:
- GEPPCOM aims at a better understanding of general-purpose parallel computing, necessary to realise cost-effective and usable parallel computing systems. The integration of the architectural, algorithmic, and programming levels will be investigated with focus on Area-Universal Networks and on Bulk-Synchronous Parallelism.
AIMS
The project aims at a better understanding of general-purpose parallel computing. The focus is on area-universal architectures and Bulk-Synchronous Parallelism (BSP). Area-Universal Networks (AUNs) are probably (almost) as efficient as any other computer of about the same cost (semiconductor area) and hence are excellent candidates for general purpose computing (indeed, AUNs are already adopted in some commercial machines). BSP provides a promising platform for a portable and efficient programming model.
APPROACH AND METHODS
The proposed research will cover various architectural, algorithmic, and programming aspects, and their interrelations. Specific objectives include:
- A characterisation of the class of AUNs.
- The design of fault-tolerant AUNs.
- The relation between area-universal computation and the following paradigms for parallel computation: Shared-Memory, Bounded-Degree Networks, and BSP.
- Numerical algorithms for AUNs (specifically, for various vector and matric operations, QR decomposition, eigenvalues, singular values and data structures for sparse matrices).
- Combinatorial algorithms for AUNs (specifically, for sorting and routing, pattern searching and matching, and image processing).
- Bulk-Synchronous Programming Languages (GL and extensions of Fortran).
POTENTIAL
The result of this project will strengthen the foundations for both the design of general-purpose architectures and the development of efficient and usable software for such architectures. Such results should be quite relevant to the realisation of powerful general-purpose parallel computers in the near future.
Sven Müßig, last update 07-nov-1995. Your feedback is welcome.