Ordonnancement sur plates-formes hétérogènes de tâches partageant des données

Abstract : We study scheduling and load-balancing strategies for distributed heterogeneous platforms. Our problem is to schedule a set of independent tasks in order to reduce the overall execution time. These tasks can use some input data that may be shared: each task can use several data, and each datum can be used by several tasks. Tasks have different execution durations, and data have different sizes. The difficulty is to map on a same processor tasks sharing some data, while keeping a good load-balance across the processors. Our study comprises three parts, progressively generalizing the problem. First, we restrict ourselves to the simple case where there is no data sharing, with homogeneous sizes for tasks and data, and where the platform is a master-slave platform. Data sharing is introduced in the second part, along with heterogeneity for the tasks and data sizes. In the last part, we generalize the platform model to a decentralized set of servers, that are linked through an arbitrary interconnection network. The theoretical complexity of the problem is studied. For simple cases, algorithms to compute an optimal solution are given and validated by experimental results with a real scientific application. For more complicated cases, we propose new heuristics to solve the scheduling problem. These new heuristics, and classic ones like min-min and sufferage are compared through extensive simulations. Thus, we show that our new heuristics perform as efficiently as the classic ones although their algorithmic complexity is an order of magnitude lower.
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00008222
Contributor : Arnaud Giersch <>
Submitted on : Friday, January 21, 2005 - 11:51:30 AM
Last modification on : Thursday, January 11, 2018 - 6:22:09 AM
Long-term archiving on : Friday, April 2, 2010 - 9:13:34 PM

Identifiers

  • HAL Id : tel-00008222, version 1

Collections

Citation

Arnaud Giersch. Ordonnancement sur plates-formes hétérogènes de tâches partageant des données. Réseaux et télécommunications [cs.NI]. Université Louis Pasteur - Strasbourg I, 2004. Français. ⟨tel-00008222⟩

Share

Metrics

Record views

177

Files downloads

477