Cost-Performance Evaluation of Heterogeneous Tierless Storage Management in a Public Cloud
Reika Kinoshita, Satoshi Imamura, Lukas Vogel, Satoshi Kazama, Eiji Yoshida · 2021
Data analytics, which extracts valuable information from a large amount of data, plays an important role in a company's business decision-making. Data analytical processing is generally I/O-intensive because of the need to retrieve data from storage devices. Public cloud services have recently become a popular choice for data analytical processing because a wide variety of storage volumes is immediately available without preparing real hardware. In this type of public cloud, it is necessary to combine multiple types of storage volumes appropriately to obtain a high I/O throughput at a low cost. In this paper, using Amazon Web Services (AWS), we quantitatively evaluate the advantages of a state-of-the-art heterogeneous tierless storage management (HTSM) technique, that is designed for relational databases, over a traditional storage caching mechanism. Our evaluation with all types of Elastic Block Store (EBS) volumes and TPC-H and TPC-DS benchmarks shows that the HTSM technique outperforms Linux bcache by up to 3.35 times within specified cost constraints. Moreover, we demonstrate that it also mitigates the AWS-specific throughput degradations of storage volumes.