Hadoop has become the most used framework for data intensive processes such as mapreduce. Business intelligence and technology forecasting has played a big part in enhancing the popularity of Hadoop framework. In this paper we have proposed an improvement for apache Hadoop yarn (also known as Hadoop 2.0) architecture. Container nodes, application master, resource manager and the pool of resources are arranged and used in a more efficient way in this framework. We are proposing two new algorithms for selection of application master and for selection of container nodes. © 2014 IEEE.