Algorithmic tools for efficient scheduling in parallel computing environments

Lines connect the coordinator with its partners. CORDIS does not always give exact coordinates for projects before 2014. These points are placed at city or country level.

Project objective

The objective of this project is to develop a set of efficient techniques and algorithmic tools for scheduling in parallel systems, towards a general purpose use of parallel computing. An integrated approach is proposed that incorporate three crucial factors affecting the key problem of parallel scheduling: the nature of applications, the features of parallel machines and the performance criteria of a scheduling system. In this context it is proposed to develop and analyse efficient scheduling tools supporting a wide range of applications in a machine-independent way and having the potential of becoming building blocks of today and feature parallel systems. The project will be carried out in three consecutive phases: First, it is proposed to set up a framework of accurate qualitative and quantitative descriptions of the aspects involved in the problem. The outcome of this part will be a generalized formal statement of the problem preserving, as far as possible, flexibility and universality. This will enable us to classify the problem's variants and to define reasonable objective functions. The design, in the above defined context, of new scheduling algorithms/policies is the second and main part of this research. Given the hardness of the problem, heuristic techniques may be the only available approach. Several techniques that can support fast heuristic solutions will be exploited, including graph and decision theory, process to processor assignment, approximation theory etc. All kinds of algorithms will be of our interest: deterministic, dynamic (on-line) and randomised Methods coming from artificial intelligence and the development of parallel algorithms are two promising research directions proposed. Finally, the algorithms coming from such an integrated approach to parallel scheduling will be tested for several major applications (scientific-engineering, real-time, etc.) that are to be supported and several systems' configurations/topologies (grids, hypercubes, meshes, etc.). Since the analysis of the algorithms in this area is usually difficult we expect that we can gain insight with simulation and experimentation. This project concerns a return fellowship and will allow the applicant to continue the work done during his HC&M grants and to transfer his future carrier in a less favored country (Greece). The project will be carried out in a international known institute, the National Technical University of Athens (NTUA), that offers an very good environment for the accomplishment of the proposed work.

Original text from CORDIS.

Participants

• NATIONAL TECHNICAL UNIVERSITY OF ATHENS · ATHENSCoordinatorGreece
• Not availableCity levelGreece

Links

• View on CORDIS