Modelo de Escalonamento Aplicativo para Bag of Tasks em Ambientes de Nuvem Computacional.

Abstract

Cloud computing is a reality in research, development and business environments. Thus, new chalenges related to scheduling of available resources by the managers of cloud service providers. Considering cloud computing as an infrastructure for execution of tasks with high computational cost, the development context of this work is part of a reality where applications with large processing demands are subjected to distributed environments. It was decided to use Bag of Tasks applications (BoT) to run in a cloud infrastructure, supported by an open-source platform. Since BoT are one of the most demanding type of applications, an adequate scheduling strategy is necessary for a better usage of hardware resources. The scheduling strategy model proposed in this study deals is task consolidation and takes into account only the attributes used to describe BoT applications, like arrival time of the job, duration, number of tasks, amount of processing required and number of CPUs available. In this strategy, it stands out the scheduling algorithm in application level, responsible to consolidate BoT application tasks, the main contribution of this work. For the evaluation of the proposed scheduling model, it was developed different study cases with submission of processing tasks, from BoT applications, executed in a real cloud infrastructure with OpenStack. The experiments showed that the distribution of load generated by BoT taks in different number of virtual machines have an impact in total execution time.