Ensembles of distributed, heterogeneous resources, also known as Computational Grids, have emerged as critical platforms for high-performance and resource-intensive applications. Such platforms provide the potential for applications to aggregate enormous bandwidth, computational power, memory, secondary storage, and other resources during a single execution. However, achieving this performance potential in dynamic, heterogeneous environments is challenging. Recent experience with distributed applications indicates that adaptivity is fundamental to achieving application performance in dynamic grid environments. The AppLeS (Application Level Scheduling) project provides a methodology, application software, and software environments for adaptively scheduling and deploying applications in heterogeneous, multiuser grid environments. In this article, we discuss the AppLeS project and outline our findings.

In this paper, we consider the problem of scheduling distributed biological sequence comparison applications. This problem lies in the divisible load framework with negligible communication costs. Thus far, very few results have been proposed in this model. We discuss and select relevant metrics for this framework: namely max-stretch and sum-stretch. We explain the relationship between our model and the preemptive uni-processor case, and we show how to extend algorithms that have been proposed in the literature for the uni-processor model to the divisible multi-processor problem domain. We recall known results on closely related problems, we show how to minimize the max-stretch on unrelated machines either in the divisible load model or with preemption, we derive new lower bounds on the competitive ratio of any on-line algorithm, we present new competitiveness results for existing algorithms, and we develop several new on-line heuristics. We also address the Pareto optimization of max-stretch. Then, we extensively study the performance of these algorithms and heuristics in realistic scenarios. Our study shows that all previously proposed guaranteed heuristics for max-stretch for the uni-processor model prove to be inefficient in practice. In contrast, we show our on-line algorithms based on linear programming to be near-optimal solutions for maxstretch. Our study also clearly suggests heuristics that are efficient for both metrics, although a combined optimization is in theory not possible in the general case.Keywords: Bioinformatics, heterogeneous computing, scheduling, divisible load, linear programming, stretch RésuméDans ce rapport, nous nous intéressonsà l'ordonnancement d'applications comparant de manière distribuée des séquences biologiques. Ce problème se situe dans le domaine des tâches divisibles avec coûts de communications négli-geables. Jusqu'à présent, très peu de résultats ontété publiés pour ce modèle. Nous discutons et sélectionnons des métriques appropriées pour notre cadre de travail,à savoir le max-stretch et le sum-stretch. Nous expliquons les relations entre notre modèle et le cadre mono-processeur avec préemption, et nous montrons commentétendre au cadre des tâches divisibles sur multi-processeur les algorithmes proposés dans la littérature pour le cas mono-processeur avec pré-emption. Nous rappelons les résultats connus pour des problématiques proches, nous montrons comment minimiser le max-stretch sur des machines non corrélées (que les tâches soient divisibles ou simplement préemptibles), nous obtenons de nouvelles bornes inférieures de compétitivité pour tout algorithme on-line, nous présentons de nouveaux résultats de compétitivité pour des algorithms de la littérature, et nous proposons de nouvelles heuristiques on-line. Nous nous intéressonségalement au problème de la minimisation Pareto du max-stretch. Ensuite, nousétudions, de manière extensive, les performances de tous ces algorithmes et de toutes ces heuristiques, et ce dans un cadre réa-liste. Notreétude montre que les ...

No abstract

Computational Grids, composed of distributed and often heterogeneous computing resources, have become the platform of choice for many performance-challenged applications. Proof-of-concept implementations have demonstrated that both Grids and clustered environments have the potential to provide great performance benefits to distributed resource-intensive applications. However, at the present time, careful staging, scheduling, and/or reservation of resources is essential in order for applications to achieve performance in Grid environments. If Computational Grids and shared computational clusters are to achieve their full potential, it must be possible for users to achieve application performance at any given time, and when other users are present in the system. In this paper, we describe the initial development of an AppLeS (application-level scheduler) for the resource selection portion of the Synthetic Aperture Radar Atlas (SARA) application, developed at the Jet Propulsion Laboratory (JPL) and the San Diego Supercomputer Center (SDSC). We demonstrate the effectiveness of application scheduling for distributed data applications such as SARA by providing a performance-efficient strategy for retrieving SARA data files in everyday, multiple-user Grid environments.

AbstractÐThe Computational Grid [12] has been proposed for the implementation of high-performance applications using widely dispersed computational resources. The goal of a Computational Grid is to aggregate ensembles of shared, heterogeneous, and distributed resources (potentially controlled by separate organizations) to provide computational ªpowerº to an application program. In this paper, we provide a toolkit for the development of globally deployable Grid applications. The toolkit, called EveryWare, enables an application to draw computational power transparently from the Grid. It consists of a portable set of processes and libraries that can be incorporated into an application so that a wide variety of dynamically changing distributed infrastructures and resources can be used together to achieve supercomputer-like performance. We provide our experiences gained while building the EveryWare toolkit prototype and an explanation of its use in implementing a large-scale Grid application.