HBase Memory‑Related Performance Tuning Guide

HBase performance tuning is essential and often complex, requiring a solid understanding of its internal mechanisms.

1. Properly configure JVM memory – the default Master and RegionServer each receive 1 GB; adjust them according to cluster resources, for example:

export HBASE_MASTER_OPTS="$HBASE_MASTER_OPTS -Xms8g -Xmx8g"
export HBASE_REGIONSERVER_OPTS="$HBASE_REGIONSERVER_OPTS -Xms32g -Xmx32g"

Ensure at least 10 % of memory remains free for the operating system.

2. Choose an appropriate GC strategy – use ParallelGC + CMS for small heaps (≤4 GB) and G1GC for large heaps (≥32 GB); for midsize heaps (4–32 GB) test both options.

3. Enable MSLAB – the MemStore‑Local Allocation Buffer reduces heap fragmentation and improves Full GC behavior; keep the default settings unless specific adjustments are needed.

4. Consider enabling BucketCache – a high‑performance cache that can operate off‑heap or on SSD; configure parameters such as hbase.bucketcache.ioengine, hbase.bucketcache.size, and enable combined cache mode with hbase.bucketcache.combinedcache.enabled.

5. Adjust read/write cache ratios – tune hfile.block.cache.size and hbase.regionserver.global.memstore.size (both default to 0.4) based on workload characteristics, increasing write cache for write‑heavy scenarios and read cache for read‑heavy scenarios.

6. Summary – Proper JVM sizing, appropriate GC selection, activation of MSLAB and BucketCache, and balanced cache ratios together enhance HBase stability, reduce Full GC occurrences, and improve overall throughput.