Load Balancing to Achieve High-Performance Parallel and Distributed Systems
Description
- Project Title:
- Load Balancing to Achieve High-Performance Parallel and Distributed Systems
- Acronym:
- LYDIA
- Number:
- 8144
- Work Area:
- Parallel computing and distributed systems
- Coordinator:
- Foundation for Research and Technology - Hellas
FORTH
Matzapetaki - P.O. Box 1527
GR- 71110 HERAKLION
- Coordinator Country:
- GR
- Partners
- University of Zürich CH
Siemens AG D
S.N.I. D
Universität Dortmund D
EHEI GR
- Contact Point:
- Dr. Christos Nikolaou
- Telephone:
- +30/81 229324
- Fax:
- +30/81 229342
- E-Mail:
- nikolaow@csi.forth.gr.
- Keywords:
- load balancing, distributed operating systems, high performance transaction processing, parallel computing, future architectures, interface definition languages
- Start Date:
- to be announced
- Duration:
- months
- Status:
- starting
- Abstract:
- The purpose of this project is to develop efficient load-balancing algorithms for future distributed system architectures, motivated by the major applications which they will be supporting. Applications which are specifically targeted include transaction processing, intelligent office systems, and general purpose parallel computing. The system configurations of particular interest to LYDIA include clusters of workstations, distributed memory parallel machines and virtual shared memory parallel machines. An Interface Definition Language (IDL), specific for load balancing will be developed, to facilitate the flow of load balancing information and associated decision making.
AIMS
The purpose of LYDIA is to develop efficient load Balancing algorithms and adaptive resource management techniques for parallel and distributed systems.
APPROACH AND METHODS
Managing the performance of distributed and parallel computing systems is a challenge, because of their inherent complexity and the diversity of new applications. LYDIA proposes to contribute to the effort of adaptive resource management and load balancing of complex parallel and distributed systems by:
- enhancing the understanding of the basic issues in load balancing of parallel and distributed systems: trade-off between balancing the processor load versus locality of reference, the need for load-balancing at various abstraction levels of a parallel system; the necessity to take into account both long-term (eg data access patterns) and short-term (eg bursts of random unit of work arrivals) workload characteristics.
- devising, analysing, experimenting and evaluating a wide range of load balancing algorithms that draw on expertise from various fields: parallel and distributed systems design expertise, adaptive/stochastic control, decision theory, neural networks, scheduling theory and performance analysis.
- designing and demonstrating the use of an Interface Definition Language (IDL) to model hardware and software interfaces that provide the necessary environment information for load balancing decisions.
POTENTIAL
We expect that the synergy of the various backgrounds of the partners (performance analysts, decision/stochastic control theorists, computer systems engineers and experimentalists) will produce several new, effective algorithms for load balancing. We also expect to produce useful tracing and profiling tools for distributed and parallel applications, which will be made available to the scientific and industrial community. Finally, we anticipate that the introduction of IDL will generate lively discussions, and many useful suggestions for improvement that can lead to the potential industry-wide adaption of an Interface Definition Language for load balancing. Such a language would greatly facilitate the performance management of complex distributed systems of the future and should contribute towards the eventual realisation of self managed (self tuned) computing systems.
Sven Müßig, last update 07-nov-1995. Your feedback is welcome.