Author(s)

Yang Pi, Zhanglong Wang

Corresponding Author

Yang Pi

ABSTRACT

RADOS block device (RBD) is the storage service component in Ceph, which provides an important block storage service for OpenStack and other cloud computing platforms. In response to the challenges faced by enterprises in terms of higher resource utilization and read/write rates on block storage services, Ceph provides Cache Tiering to improve cluster performance in heterogeneous storage environments. However, the least recently used (LRU) algorithm in Cache Tiering would evict more valuable data due to cache pollution which causes higher latency for some requests; Meanwhile, when the data is allocated on the storage node, the Controlled Replication Under Scalable Hashing (CRUSH) algorithm only considers the storage node capacity, which makes Ceph unable to dynamically balance the node's I/O load. To solve these problems, a storage selection strategy based on the prediction model is proposed to increase the hit ratio of object access in the cache pool and improve the overall I/O performance of the cluster; Besides, the cache pool I/O load balancing strategy is also proposed. Compared with the native mechanism, the proposed block storage optimization mechanism can achieve higher I/O throughput and more balanced I/O load.

KEYWORDS

Ceph, Block storage, Cache Tiering, Storage selection, Load balancing