Research And Implementation Of Ceph Tiered Storage Optimization Strategy

The distributed storage system Ceph not only has the characteristics of high reliability,high performance,scalability and automatic data balancing,but also combines the advanced design ideas of decentralization and non-metadata addressing.These features make Ceph a popular choice in the storage space.In the storage tiered architecture,Ceph takes the SSD storage device pool as the cache layer of the HDD storage device pool in order to further improve the performance of the system.But under the tiered storage architecture,on the one hand,Ceph has a problem of cache pollution,which creates unnecessary I/O in the system.At the same time,the unbalanced load of storage nodes in the cache pool will affect the stability of system performance.On the other hand,the time delay of the client accessing the system needs to be improved.On the above issues,this thesis analyzes the related mechanisms of Ceph's tiered storage architecture to propose an optimization strategy and has completed the following tasks:1.A cache-based storage optimization strategy is proposed for the problem of cache pollution and node load balancing,under the tiered distributed storage architecture.This strategy performs long-term granular analysis on objects in the cluster to more accurately determine the heat of the object,and selects the storage pool according to the heat of the storage object,thereby reducing the overhead caused by cold data entering the cache pool.On this basis,the load balancing strategy of the cache pool node is further proposed.The strategy performs load migration by considering the capacity,I/O performance,and I/O load of the nodes in the cache pool,so that the overall performance of the cache pool is more balanced and stable.Finally,the optimization strategy proposed in this thesis is designed and implemented,and compared with the original strategy of the system through experiments,which proves that the optimization strategy proposed in this thesis can improve the performance of system I/O performance.2.A cache prefetch strategy based on the cost model is proposed to solve the delay problem caused by the client accessing the storage cluster.This strategy first selects appropriate objects for prefetching according to the relevance of objects in the same object set,and then establishes the prefetching cost model of objects by obtaining the object heat and the network transmission delay,so as to further select prefetching objectsreasonably.The experimental comparison between the cache prefetch strategy based on the cost model and the native system shows that the proposed strategy can effectively reduce the time delay of the client accessing the storage object.The research shows that under the tiered storage architecture of Ceph,the cache pool storage optimization strategy and the cache prefetch strategy based on the cost model have important significance for improving the access performance of the Ceph distributed storage system.