Dynamic Objective and Advance Scheduling in Federated Grids

Abstract

Abstract. In this paper we present a dynamic mapping strategy for scheduling independent tasks in Federated Grids. This strategy is performed in two steps: first we calculate a new objective, and then we apply advance scheduling to meet the new objective. The results obtained by simulation show that the combination of these two steps reduces the makespan and increases the throughput. Thus, the mapping strategy proposed meets two of the most common objective functions of tasks scheduling problems: makespan and performance of the resources. The presented algorithm is easy to implement, unlike Genetic Algorithms is fast enough to be used in a realistic scheduling, and is efficient. In addition, the information the strategy needs can be provided by any Grid Information Service, and its does not require the deployment of complex prediction services or service level agreement: it can work in any Grid.

Preparation and Catalysis of Carbon‐Supported Iron Catalysts for Fischer–Tropsch Synthesis

“…In our previous work 3 we proposed a generic decentralized model for scheduling independent tasks in a Federated Grid. The scheduling algorithms considered for this model were four: SO (Static Objective), DO (Dynamic Objective), SO‐AS (Static Objective and Advance Scheduling), and DO‐AS (Dynamic Objective and Advance Scheduling) 4. These algorithms are based on a performance model 5 that allows to parameterize and compare different grids.…”

“…Simulation results demonstrated that DO‐AS provides the best job distribution and makespan. However, DO‐AS only considers one internal resource, and one external resource to calculate the performance of the whole Federated Grid (for further details, refer to 4). Table I summarizes the results obtained by DO‐AS.…”

“…Based on Federated Grid requirements, in our previous work 3 we proposed a generic decentralized model for scheduling independent tasks in Federated Grids. Various mapping strategies 4 were presented. These algorithms are mainly based on a performance model 5 that allows to parameterize and compare the different grids forming a Federated Grid.…”

A decentralized model for scheduling independent tasks in Federated Grids

“…As depicted in Figure 5 we have used the same scenario than in our previous work [31], [30]. There are two grid resources in this test scenario: the DSA (Distributed System Architecture) and the LCG (LHC Computing Grid).…”

“…Ë The Dynamic Objective and Advance Scheduling (DO-AS) version combines the two mechanisms that have demonstrated to be an enhancement: DO and AS. Thus, DO-AS compared with GridWay's current scheduling policy, clearly satisfies the makespan and performance objectives of the resources [30], which are two of the most common objective functions of tasks scheduling problems [12].…”