SmartLTM: Smart Larger-Than-Memory Storage for Hybrid Database Systems
Paulo R. P. Amora, Elvis Marques Teixeira, Francisco D. B. S. Praciano, Javam C. Machado · 2018
Main-memory DBMS can offer hybrid and evolving storage architectures, instead of the traditional row or column storage layouts. Even if RAM is affordable nowadays, it is still a limited resource concerning available storage space in comparison to conventional storage devices. Due to this space restriction, techniques that leverage a trade-off between storage and query performance were developed and should be applied to data that is not frequently accessed or updated. This work proposes SmartLTM, a data eviction mechanism that considers the decisions previously taken by the DBMS in optimizing data storage according to query workload. We discuss how to migrate data, access it and the main differences between our approach and a row-based one. We also analyze the behavior of our solution in different storage media. Experiments show that cold data access with SmartLTM incurs an acceptable 17% of throughput loss, against 26% of the row-based one, while retrieving only half of the data to answer queries.