HBase Read Performance — Optimization Deep Dive

Why it matters

Read latency matters for interactive workloads.

The architecture

Block cache: hot HFile blocks in memory.

Bloom filters: skip HFiles without target row.

How it works end to end

Scan caching: rows returned per RPC. Higher = better throughput, more memory.

Batch: cells per RPC. Prevents OOM on wide rows.

Compaction: fewer HFiles = faster reads.

Short-circuit reads: HDFS local reads faster.