Evaluation on Applicability of Measurement-Based Join Cost Calculation Method Using Different Generation CPUs

Cloud computing is gaining increasing attention, and users are increasingly demanding higher performance. In particular, the database-as-a-service (DaaS) market is rapidly expanding. Attempts to improve database performance have led to the use of nonvolatile memory as a durable database medium instead of existing storage devices. However, the CPU architecture of different physical servers may be of different generations. For database systems running on a cloud infrastructure, the cost of storing databases in nonvolatile memory decreases and the cost of CPU operation increases relative to choosing the most suitable execution plan for a database query. In our previous study, we proposed a measurement-based join cost calculation method to select the best join method. However, when the cost calculation method is not portable to database systems with a CPU with architecture differing from that of the CPU used to measure the statistical information, re-measuring the statistical information, and recreating the cost calculation model becomes necessary. The current study is aimed at developing a method for updating the measurement-based cost calculation formula to support a CPU with architecture from a different generation without the need to re-measure the statistical information of the CPU. Our approach focuses on reflecting architectural changes, such as cache size and associativity, memory latency, and branch misprediction penalty, in the components of the cost calculation formulas. The updated cost evaluation formulas estimated the cost of joining different generation-based CPUs accurately in 66% of the test cases. Therefore, an in-memory database system using the measurement-based cost calculation method can be applied to a database system with CPUs from different generations.