Abstract. This contribution presents algorithms, methods, and software for a Grid resource manager, responsible for resource brokering and scheduling in early production Grids. The broker selects computing resources based on actual job requirements and a number of criteria identifying the available resources, with the aim to minimize the total time to delivery for the individual application. The total time to delivery includes the time for program execution, batch queue waiting, input/output data transfer, and executable staging. Main features of the resource manager include advance reservations, resource selection based on computer benchmark results and network performance predictions, and a basic adaptation facility.

The increasingly common practice of using multiple distributed storage systems as a distributed data store within which large datasets may be replicated has led to the problem of how to access replicated data efficiently. Multiple-source parallel transfers can improve data throughput time by transferring data from several replicas in parallel. However, we then face the problem of deciding how to distribute the data load among different storage resources. We propose a Tuned Conservative scheduling technique that uses predicted mean and variance network information to make data distribution decisions. This stochastic scheduling technique uses a tuning factor to adjust the amount of the data assigned to a link in accordance with the variability of the network performance. We apply our technique to the GridFTP implementation in the Globus Toolkit and demonstrate that the technique can produce data transfer times that are significantly faster and less variable than those of other techniques. 1.

Abstract. This paper describes the architecture built into the Imperial College e-Science Infrastructure (ICENI) for handling performance meta-data. The architecture provides a means to gathering performance information, processing this information to populate the performance store, and to use this performance information to aid in the selection of resources and component implementations. The Performance Framework is developed in a “pluggable ” manner allowing alternate implementations of the three main features to be used. Performance Stores may be either data stores or based on analytical models. 1

In this paper we discuss the difficulties involved in the scheduling of applications on computational grids. We highlight two main sources of difficulties: firstly, the size of the grid rules out the possibility of using a cen-tralized scheduler

playsakeyroleinGridresourcemanagementasitenables thesystemtomeetuserexpectationswithrespecttotime requirements and temporal dependence of applications, increases predictability of the system and enables coallocation of resources. Despite these attractive features, adoptionofadvancereservationsislimitedmainlydueto thefactthatrelatedalgorithmsaretypicallycomplexand fail to scale to large and loaded systems. In this work weconsidertwoaspectsofadvancereservations.First,we investigatetheimpactofheterogeneityonGridresource management when advance reservations are supported. Second,weemploytechniquesfromcomputationalgeometrytodevelopanefficientheterogeneity-awarescheduling algorithm.OurmainfindingisthatGridsmaybenefitfrom highlevelsofresourceheterogeneity,independentlyofthe totalsystemcapacity.Ourresultsshowthatouralgorithm performswellacrossseveraluserandsystemperformance andovercomethelackofscalabilityandadaptabilityof existingmechanisms. I.

Abstract. While the Grid is becoming a common word in the context of distributed computing, users are still experiencing long phases of adaptability and increased complexity when using the system. Although users have access to multiple resources, selecting the optimal resource for their application and appropriately launching the job is a tedious process that not only proves difficult for the naïve user, but also leads to ineffective usage of the resources. A generalpurpose resource broker that performs application specific resource selection on behalf of the user through a web interface is required. This paper describes the design and prototyping of such a resource broker that not only selects a matching resource based on user specified criteria but also uses the application performance characteristics on the resources enabling the user to execute applications transparently and efficiently thereby providing true virtualization. 1

Abstract. Parallelism is one of the main sources for performance improvement in modern computing environment, but the efficient exploitation of the available parallelism depends on a number of parameters. Determining the optimum number of threads for a given data parallel loop, for example, is a difficult problem and dependent on the specific parallel platform. This paper presents a learning-based approach to parallel workload allocation in a costaware manner. This approach uses static program features to classify programs, before deciding the best workload allocation scheme based on its prior experience with similar programs. Experimental results on 12 Java benchmarks (76 test cases with different workloads in total) show that it can efficiently allocate the parallel workload among Java threads and achieve an efficiency of 86 % on average.

Grid computing enables sharing, selection and aggregation of large collections of geographically and organizationally distributed heterogeneous resources to increase computational, and storage power, resource accessibility and utilization for solving large-scale data intensive problems in science, engineering and commerce. One of the distinct characteristics of grid system is resource heterogeneity. The effective use of a Grid requires the definition of an approach to manage the heterogeneity of the involved resources that can include computers, data, network etc. In order to develop an efficient resource management and scheduling strategies for grid environments, how heterogeneity affects performance of systems and applications is to be well understood. In this study, the potential impact that heterogeneity of grid resources has on the performance and reliability achieved by grid applications and system is analyzed. We have made an attempt to categorize and quantitatively characterize heterogeneity based on various resource characteristics. We quantify the heterogeneity impact of varying computational power, communication bandwidth,

Abstract—Scheduling algorithm for batch-mode dataintensive jobs is a key issue in data-intensive Grid applications. It focuses on how to minimize the overhead of transferring the required data set to the executing grid site. Existing approaches pay attention to the access cost of a data-intensive job at each executing grid site for replicating the required data set. However, they neglect the influence from potential behaviors of jobs in the waiting queue at each grid site when the access cost is evaluated. In this paper, we consider the influence of potential behaviors on the access cost, and propose a data-intensive job scheduling algorithm with potential behaviors. Furthermore, the causation of potential behaviors is analyzed. The simulation result in OptorSim shows that it has better performance in mean job time of all jobs, total number of replications, total number of local files accesses and effective network usage than the scheduling algorithm based on access cost. Index Terms—distributed computing, grid computing, data grid, job scheduling, access cost, replica replacement I.

Service-based Systems (SBS) are being adopted by many distributed systems. Applications in SBS can often be viewed as the composition of various computing services following specific workflows. These workflows often need to satisfy various timing and resource constraints. In this paper, a software cybernetics approach to deploying and scheduling workflows with timing and resource constraints in SBS is presented. In our approach, a logic-based technique for modeling and solving timing and resource constraints for workflows in SBS is developed to generate the initial resource assignments, schedules and deployment plans of agents for workflows. The principles and concepts in software cybernetics are applied to guide the synthesis of software controllers for monitoring and adapting system behavior.