E.R. Rodrigues, A.J. Preto, S. Stephany, H.F. de Campos Velho (2006): A Simple Job Scheduler for a Grid-based Climatology Application, Seventh International Meeting on High Performance Computing for Computational Science (VECPAR'06), July 09-12, Rio de Janeiro (RJ), Brazil.

Abstract: Grid computing is becoming very common as a solution to implement high-throughput and large-dataset applications such as climate change studies. The Globus Toolkit is becoming an emerging standard for grid applications, but it is still lacking scheduling capabilities which must be supplied independently. This work presents new job scheduler that was proposed and developed in a grid implemented through the Globus Toolkit. This grid is intended for a climatology application that employs a mesoscale meteorological model, the BRAMS (Brazilian Regional Atmospheric Modeling System). An usual scheme to study long term temporal evolution of the climate is to execute climatology ensembles. Each ensemble is composed of a meteorological simulation encompassing several years for different initial conditions. Each realization is given by the execution of BRAMS, and corresponds to a specific job. The amount of jobs in climatology studies is typically many hundreds. The new job scheduler accesses a job database that stores job requests. A resource list is used by the scheduler to search for available clusters and assign them jobs in a round-robin scheme that assures a good load balancing, even considering the heterogeneity of the clusters that compose the grid. Performance results are presented and discussed.