Similarity-Based Garbage Collection for Compressing Live Data Segments

The patent segments the garbage collection process into distinct phases: data gathering phase where similar data blocks are identified and grouped using similarity hashes, and compression phase where grouped data is compressed. This segmentation allows complex similarity-based compression to be performed only on grouped segments rather than all data, improving storage efficiency while managing complexity through phased processing.

The patent performs preliminary action by pre-grouping similar data blocks using similarity hashes before the actual compression operation. Data segments are identified and grouped in advance based on their similarity characteristics, so that when compression occurs, only pre-identified similar segments are compressed together. This preliminary grouping improves compression efficiency without requiring complex real-time analysis during compression.

The patent decouples data gathering and compression operations from logical element operations by segmenting the processing into independent phases. The data gathering phase identifies similar blocks using similarity hashes independent of logical element boundaries, and the compression phase operates on grouped data separately from logical element management. This segmentation reduces computational overhead by avoiding repeated processing of the same data across multiple logical operations.

The patent uses similarity hashes as copies or representations of data segments to identify similar data without requiring full data comparison. Instead of comparing actual data content, the system compares hash representations, which are computationally cheaper copies. This copying approach reduces computational costs while maintaining the ability to identify similar data segments for efficient compression.

The patent uses similarity hashes as simplified copies or representations of data segments to detect similarity. Instead of performing complex direct comparison of actual data blocks, the system computes and compares hash representations that capture essential similarity characteristics. This copying approach transforms the difficult problem of direct data comparison into a simpler hash comparison task, improving data reduction efficiency while reducing detection complexity.

The patent replaces the mechanical approach of direct data block comparison with a computational hash-based system. Instead of mechanically comparing data byte-by-byte or block-by-block, the system substitutes this with computational hash functions that quickly generate similarity identifiers. This substitution transforms a computationally intensive mechanical comparison process into an efficient computational operation, enabling effective similarity detection at scale.