For write-heavy workloads, which database design is commonly preferred?

For write-heavy workloads, which database design is commonly preferred?

• A. Hash indexes
• B. LSM-backed databases
• C. Geospatial indexes
• D. B-tree indexes

Answer: (B)

For write-heavy workloads, you want a storage design that makes writes cheap and sequential rather than forcing frequent random disk updates. Log-Structured Merge trees do exactly that: writes go into a memory buffer (memtable) and are appended to disk as sorted runs. Periodically, those runs are merged in the background through compaction. This turns random disk writes into sequential I/O, boosting write throughput and reducing write amplification compared with traditional on-disk indexes. Reads in LSM-based systems still work well, aided by bloom filters and caches, but they may need to consult multiple levels and handle compactions. The payoff is clear for writes: high throughput and efficient use of disk bandwidth. Other options don’t align as well with heavy writes. Hash indexes focus on exact lookups and aren’t optimized for range queries or maintaining order; geospatial indexes target spatial queries rather than general write throughput; B-tree indexes provide strong read performance but incur more random I/O on updates, which hurts write-heavy workloads.

For write-heavy workloads, you want a storage design that makes writes cheap and sequential rather than forcing frequent random disk updates. Log-Structured Merge trees do exactly that: writes go into a memory buffer (memtable) and are appended to disk as sorted runs. Periodically, those runs are merged in the background through compaction. This turns random disk writes into sequential I/O, boosting write throughput and reducing write amplification compared with traditional on-disk indexes.

Reads in LSM-based systems still work well, aided by bloom filters and caches, but they may need to consult multiple levels and handle compactions. The payoff is clear for writes: high throughput and efficient use of disk bandwidth.

Other options don’t align as well with heavy writes. Hash indexes focus on exact lookups and aren’t optimized for range queries or maintaining order; geospatial indexes target spatial queries rather than general write throughput; B-tree indexes provide strong read performance but incur more random I/O on updates, which hurts write-heavy workloads.