# Original `LSM`
+ Each component is structured as a `B+-Tree`
+ When **`Ci`** is full, a rolling merge process is triggered
+ When merge happens, a range of leaf pages are merged from **`Ci`** into **`Ci+1`**

![`smart`](LSMRollingMerge_aa_001_20221020.png)

------------------------------------------------------------------------------------

Because random disk reads and writes are 3 orders of magnitude slower than sequential reads and writes, `LSM` tries to convert disk reads and writes to sequential reads and writes. So why is `LSM` doing this? Actually, this is related to the fact that there are no optimization methods for disk writing, and there is room for optimization in both hardware and software for disk read. Read performance is easier to achieve than write performance in general.

Diagram referenced from `O’Neil, P., et al.: The log-structured merge-tree (LSM-tree). Acta Inf. 33(4), 351–385 (1996)`.