HotSpot & Cache: an Optimization Method for Small Objects Storage in SWIFT of OpenStack Cloud

In the big-data era, cloud platforms such as OpenStack aims at the optimization of storage performance for large objects. However small files severely hurdle its performance. In this paper a framework, namely HotSpot & Cache, is proposed to optimize the storage performance for small objects in SWIFT of OpenStack cloud. A double-proxy architecture is adopted for HotSpot & Cache architecture, including an external proxy node, an internal proxy node, and a dedicated cache server. Storage performance optimization for small objects is accomplished in the unit of partitions in storage nodes, hot-spot partitions are predicted and cached in the dedicated cache server. A series of facilities are implemented for small objects storage performance optimization, e.g. Sampling & Statistics, Hot-Spot Prediction, and a Request Redirection. The access frequency and the size of a partition is periodically sampled and statistically accumulated to predict the hot-spot of a partition for the next interval. Experimental results demonstrate the effectiveness of the proposed HotSpot & Cache architecture.