Abstract—For many organizations, one attractive use of cloud resources can be through what is referred to as cloud bursting or the hybrid cloud. These refer to scenarios where an organization acquires and manages in-house resources to meet its base need, but can use additional resources from a cloud provider to maintain an acceptable response time during workload peaks. Cloud bursting has so far been discussed in the context of using additional computing resources from a cloud provider. However, as next generation applications are expected to see orders of magnitude increase in data set sizes, cloud resources can be used to store additional data after local resources are exhausted. In this paper, we consider the challenge of data analysis in a scenario where data is stored across a local cluster and cloud resources. We describe a software framework to enable data-intensive computing with cloud bursting, i.e., using a combi-nation of compute resources from a local cluster and a cloud environment to perform Map-Reduce type processing on a data set that is geographically distributed. Our evaluation with three different applications shows that data-intensive computing with cloud bursting is feasible and scalable. Particularly, as compared to a situation where the data set is stored at one location and processed using resources at that end, the average slowdown of our system (using distributed but the same aggregate number of compute resources), is only 15.55%. Thus, the overheads due to global reduction, remote data retrieval, and potential load imbalance are quite manageable. Our system scales with an average speedup of 81 % when the number of compute resources is doubled. I.

Asynchronous algorithms have been demonstrated to improve scalability of a variety of applications in parallel environments. Their distributed adaptations have received relatively less attention, particularly in the context of conventional execution environments and associated overheads. One such framework,MapReduce,hasemergedasacommonlyusedprogramming framework for large-scale distributed environments. While the MapReduce programming model has proved to be effective for data-parallel applications, significant questions relating to its performance and application scope remain unresolved.Thestrictsynchronizationbetweenmapandreduce phases limits expression of asynchrony and hence, does not readily support asynchronous algorithms. This paper investigates the notion of partial synchronizations in iterative MapReduce applications to overcome global synchronization overheads. The proposed approach applies a locality-enhancing partition on the computation. Map tasks execute local computations with (relatively) frequent local synchronizations, with less frequent global synchronizations. This approach yields significant performance gains in distributed environments, even though their serial operation counts are higher. We demonstrate these performance gains on asynchronous algorithms for diverse applications, including PageRank, Shortest Path, and K-Means. We make the following specific contributions in the paper — (i) we motivate the need to extend MapReduce with constructs for asynchrony, (ii) we propose an API to facilitate partial synchronizations combined with eager scheduling and locality enhancing techniques, and (iii) demonstrate performance improvements from our proposed extensions through a variety of applications from different domains. I.