A Ceph Heterogeneous Storage Strategy Based on File Access Behavior Awareness

In recent year, the heterogeneity of cluster storage devices has become increasingly prevalent. However, Ceph did not adequately consider heterogeneous storage scenarios during its design. Furthermore, when selecting storage nodes, Ceph solely relies on the computing performance indicators of the hosts where the storage nodes are located, neglecting the characteristics of stored objects, resource loads, and node heterogeneity. This oversight results in an inability to effectively harness the performance attributes of heterogeneous storage devices. Consequently, this paper introduces a Ceph heterogeneous storage strategy grounded in the awareness of file access behaviors. By leveraging a pre-trained classification model to forecast the access characteristics of storage objects and factoring in performance metrics such as CPU load and memory utilization rate of nodes, this approach employs the TOPSIS method to determine OSD (Object Storage Daemon) storage policies. This enables full exploitation of the performance features of heterogeneous storage devices, thereby enhancing the read and write efficiency of the cluster. Empirical studies demonstrate that this methodology can significantly improve the read and write performance of Ceph clusters for larger objects in heterogeneous storage environments.