High Performance Resilient Parallel Computing


QMIPS-RPC - EC-US012

Keywords: parallel processing, HPC


Start Date: 1 August 92 / Status: finished / Duration: 24 months

[ participants / contact]


Objectivies and Approach

The aim of the project is to develop methods and tools for the design of dependable application software which will run correctly on a distributed system in which processing units (PU) may fail. The distributed system considered is composed of autonomuous PUs with large local memory, interconnected via a high-speed networks. the distributed application spftware considered is composed of a set of communicating and interdependent tasks, where each task is assigned to a PU for execution. The approach consists of introducing specific constructs into the distributed software to insure correct execution even when some of the processors fail. The specific constructs are added to some or most of the tasks so that the application would be capable of self-monitoring, fault detection and recovery capabilities. The tasks with monitoring capabilities are called "agents". The adopted approach adds some overhead even when no failures occur but will guarranty correct execution in the presence of failures.

European links: ESPRIT project QMIPS (7269).

Progress and Results

Both the American and European teams have worked on the Algorithms design for the dynamic allocation of tasks, on the evaluation of these algoritms using simulation and on testing them on a workbench of interconnected workstations. Visits between the sites have led to efficient exchange of findings. The simulation evaluation has been completed and has shown very interesting results related to the effect of processors failures and communication delays.


Coordinator

Université René Descartes
rue des Saints-Pères, 45
F - 75006 PARIS, F

EU Partners

Université René Descartes, F
University of Erlangen-Nüremberg, D
INRIA, F
University of Turin, I
CWI, NL
University of Newcastle-upon-Tyne, UK
Imperial College, UK

Non-EU Partners

Courant Institute of Mathematical Science,
New York University, USA

CONTACT POINT

Prof. Erol Gelenbe
Tel: +33 1 428 62 232
Fax: +33 1 428 62 231
E-mail: erol@ehei.ehei.fr


INCO synopses home page INCO cooperation index INCO keyword index
INCO acronym index INCO number index INCO Working Groups index
All
synopses home page all keywords index all acronyms index all numbers index

QMIPS-RPC - EC-US012, May 1997


please address enquiries to the ESPRIT Information Desk

html version of synopsis by Nick Cook