LSM-tree Compaction Acceleration Using In-storage Processing
Minje Lim, Jeeyoon Jung, Dongkun Shin · 2021
Log structured merge tree (LSM-tree) is widely used to implement key-value stores. To maintain the structure of the LSM-tree, it is necessary to perform compaction to remove duplicated entries. Compaction can delay the user’s write operation and affects performance. If In-storage processing is applied to compaction, data traffic between Host-Storage can be reduced and tasks can be processed quickly to improve system performance. If compaction is efficiently processed by applying in-storage processing, overall DB performance can be improved. In this paper, we designed an IP that can accelerate the compaction and integrated it into the Cosmos+ OpenSSD.