Virtualized computing infrastructures are often used to create clusters of resources tailored to solve tasks taking into account particular requirements of these tasks. An important objective is to evaluate such requirements and request optimal amount of resources which becomes challenging for parallel tasks with intercommunication. In previous works we investigated how light-weight container-based virtualization can be used for creating virtual clusters running MPI applications. Such cluster is configured according to the requirements of particular application and allocates only necessary amount of resources from the physical infrastructure leaving space for co-allocated clusters running without conflicts or resource races. In this paper we investigate similar concepts for MapReduce applications based on Hadoop framework that use Cloudply virtualization tool to create and manage light-weight virtual Hadoop clusters on Amazon cloud resources. We investigate performance of several Hadoop benchmarks in differen