Autor

María Arsuaga-Ríos, Miguel A.Vega-Rodríguez, Francisco Prieto Castrillo.

Evento

ECCS'10 European Conference on Complex Systems
Lugar: Lisboa (Portugal)
Fecha: 13-17 Septiembre 2010
Tipo de publicación: Póster

Grid computing is a distributed computing environment which can be thought as a complex network. However, the current implementation makes difficult the job scheduling by using the time and cost involved in real productions. Currently, schedulers usually take into account only requirements of the experiment itself, such as operating system or storage capacity like the WMS (Workload Management System) of glite. However, the new approaches of schedulers consider QoS(Quality Of Service) objectives, like execution time or costs for an experiment. In this sense, DBC(Deadline Budget Constraint)[1] is a scheduler used by Nimrod-G that assembles these values using a greedy algorithm.

This scheduler presumes to have a good performance, but sometimes it fails in finding the whole set of selections of the experiment. In this paper, a scheduler decision support is presented, offering to the end user a range of alternatives to achieveits outlook in terms of time and money. This collection of solutions shows the suitable resource for each job, thatcompose the experiment and their order of execution. A new optimization algorithm is implemented, namely, MOABC(Multi-objective Artificial Bee Colony), to solve the problem. This algorithm is based on the recent ABC (ArtificialBee Colony) algorithm [2] adding multi-objective[3] properties to the preceding versions.The scheduler is implemented in the well-known grid simulator, GridSim[4]. It has been used and modified torecreate the performance of network topologies, processing elements, workflows, etc. This allows the evaluation andcomparison with others schedulers like DBC.