An New Data Parallelism Approach with High Performace Clouds

This paper presents a new high performance cloud framework that support simplified data parallel applications. and describes the design and implementation of it. This cloud implements stream processing paradigm over clusters of computers, either within single data center or across wide area. Unlike MapReduce, this Cloudpsilas stream processing paradigm is simpler, more straightforward, and reaches better performance in certain cases. The experimental results show this cloud is more straightforward, simpler to use, and it is about twice as fast as Hadoop.