Data-centered computing and implementation in web service

摘要

Parallel computer architectures bring synchronization problem which bog down the system. And the speed gap between CPU and memory keeps growing, making data operation more expensive. In some data-intensive applications, there are similar or coincident child operations in parallel threads, which take unnecessary resources. To effectively exploit parallelization, reduce the price of data operation, the concept of data-centered computing is proposed. By considering the similarity and relations between data and merging them into unions as threads' input, data-centered computing can eliminate data operation redundancy of data-intensive applications. An implementation in web service proves that data-centered computing can effectively improve performance.