How HBase performs updates at disk transfer rate instead of disk seek rate?

Question

I am reading the book HBase: The Definitive Guide and it is mentioned that while traditional relational databases perform updates/deletes at seek rate (B-trees), HBase performs updates/deletes at transfer rate (LSM trees).

I am aware how LSM trees work, though I fail to understand why the author claims that updates happen at transfer rate. When a client issues a specific update operation then a corresponding entry will be written to the in-memory memstore and there will also be an entry that will be appended to the WAL. Periodically the memstore will be merged with the HFiles stored on disk.

How that translates to HBase performing updates/deletes at transfer rate?

Answer 1

HBase performs updates and deletes when it merges multiple LSM trees into one (memstore compaction) or when it does files compaction (merge of multiple HFiles into one) until than it just stores multiple versions of the same record.

Transfer rate database paradigm in this case means that, all updates are applied only during transferring the date to the new place.