Datacompaction

Datacompaction refers to techniques that reduce the amount of storage occupied by data and improve access performance by reorganizing and rewriting data. It combines restructuring of layout with data-preserving transformations such as compression, deduplication, and the removal of obsolete or redundant records. Datacompaction is used in databases, file systems, data lakes, and streaming systems to optimize storage efficiency and query or read performance.

In log-structured storage systems and LSM-tree databases, compaction is a controlled process that merges multiple data files, discards tombstones or expired entries, and rewrites data to more contiguous blocks. This reduces read amplification and speeds lookups but can be CPU- and I/O-intensive during the operation. In traditional relational databases, vacuuming or garbage collection reclaims space from deleted rows and reorganizes pages. In big data and cloud storage, compaction often means combining numerous small files into larger ones to reduce metadata overhead and improve throughput; it may also apply columnar compression and encoding schemes to further reduce size.

A number of design considerations influence when and how aggressively to compact: workload patterns, latency requirements, storage media, and consistency guarantees. Aggressive compaction can reclaim space and improve performance but may temporarily degrade write latency or cause bursts of I/O. Effective strategies balance data retention, throughput, and resource consumption, and often run periodically or adaptively in response to workload.

Common contexts include Cassandra and RocksDB use of compaction, HBase with major/minor compactions, HDFS or cloud data lakes with file-level compaction, and PostgreSQL vacuuming. Datacompaction is an ongoing area of storage system design, with tradeoffs evolving as hardware and workloads change.