Authors:
Henryk Krawczyk
and
Jerzy Proficz
Affiliation:
Gdansk University of Technology, Poland
Keyword(s):
Task assignment, Computation cluster, Multimedia processing.
Related
Ontology
Subjects/Areas/Topics:
Distributed and Mobile Software Systems
;
Grid, Peer-To-Peer, and Cluster Computing
;
Software Engineering
Abstract:
The paper describes a computational model of the KASKADA platform. It consists of two main elements: a computational cluster, and a task graph. The cluster is represented by a finite set of the nodes with the specific maximum loads. The graph contains nodes representing tasks to be executed, and the edges representing continuous data flow between the tasks. The tasks are executed concurrently and the data flows between them are directed and acyclic. For such a model, the problem of task-to-nodes assignment is analysed, and two optimisation goals are defined: low cluster fragmentation, and minimum processing latency. For both problems the heuristic algorithms are described. The simulation results of the described algorithms are provided, and their evaluation is performed. Finally, the future algorithm improvements are suggested.