Data rearrange based on mining block access sequence in Cloud Storage

Abstract

In Cloud Storage system, storage systems have to service for large numbers of scattered data access nodes, and data I/O almost are in random access form, in addition, the distribution storage of data result in data transmission between the Cloud nodes increases greatly, then the performance of the cloud storage system was be restricted remarkably. In this paper, we put forward a system for improving access performance in Cloud Storage system. Through the access trace, the process which doing disk I/O should be detected. The processes that are executed contemporaneous should be regarded as a group. Then, the block access sequences belong to same process groups could be mined based on the Frequent Sequence Mining. And the block relation could be obtained. Ultimately, the related blocks could be rearranged to the near location. It could reduce the head moving of disk during access and realize the mapping from random access to sequence access partly. Furthermore, the data could be migrated between the storage nodes according to the block relation and reducing the data transfer through network between nodes. We have also evaluated the benefits of the system. Our result using real system workloads show that with the data rearrange and data transfer, about 10%-20% I/O response time could be reduced.