Hybrid HBase: leveraging flash SSDs to improve cost per throughput of HBase
Anurag Awasthi, Avani Nandini, Arnab Bhattacharya, Priya Sehgal · 2012
Column-oriented data stores, such as BigTable and HBase, have successfully paved the way for managing large key-value datasets with random accesses. At the same time, the declining cost of flash SSDs have enabled their use in several applications including large databases. In this paper, we explore the feasibility of introducing flash SSDs for HBase. Since storing the entire user data is infeasible due to impractically large costs, we perform a qualitative and supporting quantitative assessment of the implications of storing the system components of HBase in flash SSDs. Our proposed Hybrid HBase system performs 1.5-2 times better than a complete disk-based system on the YCSB benchmark workloads. This increase in performance comes at a relatively low cost overhead. Consequently, Hybrid HBase exhibits the best performance in terms of cost per throughput when compared to either a complete HDD-based or a complete flash SSD-based system.