Access-Load-Aware Dynamic Data Balancing For Cloud Storage Service

Haopeng Chen,Zhenhua Wang,Yunmeng Ban
DOI: https://doi.org/10.1007/978-3-642-41428-2_25
2013-01-01
Abstract:Cloud storage is the typical way for storing massive data in Big Data Era. Dynamic data balancing is important for cloud storage since it aims to improve the utilization of computing resource and the performance of data process. However, storage-load-aware data balancing, adopted by almost all existing cloud storage services and systems, is far less effective than access-load-aware one for typical cloud applications with hotspots of data. This paper focuses on the latter and puts forward a mechanism of dynamic data balancing for optimization of resource utilization. The mechanism detects the overloaded and underloaded physical nodes and virtual nodes by monitoring their utilization of resource. Then, it dynamically balances the access load among the nodes by pair, merge, mark, scale up and scale down operations. This mechanism is useful for the applications with hotspots in data. So it is a complementation of storage-load-aware data balancing. The results of experiments on Swift demonstrated the effectiveness of this mechanism.
What problem does this paper attempt to address?