Cora: Data correlations-based storage policies for cloud object storage

Abstract

In a cloud object system, all requests and data are routed through the proxy server, which limits system performance. Exploiting data correlations to prefetch objects can effectively alleviate stress on the proxy server and improve system performance. To achieve this goal, it is necessary to consider the extraction of data correlations, the distribution of correlated objects in the cluster, and the way correlated objects are stored. In this paper, we propose a policy-driven framework, Cora, to support data correlations-based storage policies to efficiently maintain correlations and prefetch correlated objects in the cloud object storage system. Based on the characteristics observed in explicit and implicit data correlations, we design different storage policies with various scheme implementations. Experiments demonstrate that leveraging data correlations can bring significant performance improvements to the cloud object storage system. Throughput and latency are optimized up to 285.24% and 55.39%, respectively.