Design and implementation of distributed mobile computing platform using hadoop

This thesis is aimed at design and evaluation of a distributed cloud computing platform using mobile nodes connected via cellular network. Such a "Mobile Cloud" capability is motivated by situations in which the cellular access network has limited backhaul bandwidth to the Internet as may be the case in developing regions or disaster recovery scenarios. In view of the growing importance of the OpenStack and Hadoop/MapReduce framework, we first evaluate the performance of Hadoop in the mobile cloud environment and then consider modifications necessary to improve performance. The study was conducted using an experimental methodology based on the GENI open WiMAX base station at WINLAB, which was used to set up a cluster of 4G/cellular mobile clients running Hadoop. The results show that as expected, performance of Hadoop cloud applications is severely degraded due to mobile device channel quality variations in the baseline case, but that significant improvements (~5x in run-time) can be realized with relatively modest cross-layer (i.e. base-station and signal-to-noise ratio) aware strategies. The results also show that, there are certain types of applications that are not suitable for Distributed Mobile Computing, e.g. applications that generate large outputs, while applications with small sized output benefit greatly from Mobile Cloud.