作者: Hobin Yoon , Juncheng Yang , Sveinn Fannar Kristjansson , Steinn E. Sigurdarson , Ymir Vigfusson
关键词: Merge (version control) 、 Computer science 、 Mutant 、 Cloud computing 、 Operating system 、 Cloud database 、 Storage type 、 Latency (engineering) 、 Data migration 、 Workload
摘要: Today's cloud database systems are not designed for seamless cost-performance trade-offs changing SLOs. Database engineers have a limited number of due to the storage types offered by vendors, and switching different type requires time-consuming data migration new database. We propose Mutant, layer log-structured merge tree (LSM-tree) stores that dynamically balances cost performance organizing SSTables (files store subset records) into based on SSTable access frequencies. implemented Mutant extending RocksDB found in our evaluation delivers with YCSB workload real-world trace. Moreover, through additional optimizations, lowers user-perceived latency significantly compared unmodified