Chapter 7. Log Structured Storage

Brief

LSM Storages are focused on write-heavy workloads given that, very briefly, it's all all about appending values and merging then some point in the future. Opposite to B-Tree where we need to locate the node, either from disk or cache, and update in place. We only append values, making it much more faster. Read workloads are much slower in LSM dues to the needs of aggregating values.

On a hardware POV, B-Tree as costly because due to N sorts of SMO's (Structure Modification Operations) on writes/reads, alot of the IO is random. LSM are sequential by design.