Data Structure Layout Optimization for Multi-Threaded Cache Performance

The patent applies local quality by differentiating the treatment of different data fields within the same structure. Instead of uniformly optimizing for spatial locality, the system calculates code concurrency values for each field pair and applies clustering only to fields with low code concurrency. This selective approach allows spatial locality optimization for suitable fields while preventing false sharing for fields that would conflict, thereby resolving the contradiction between cache performance and false sharing prevention.

The patent changes the parameter of data field grouping by introducing code concurrency as a decision criterion. The clustering algorithm uses calculated code concurrency values to determine which fields should be grouped together in the same cache line. Fields with high code concurrency (likely to be accessed concurrently by multiple threads) are placed in different clusters/cache lines, while fields with low code concurrency are grouped together. This parameter-based approach optimizes memory access efficiency while minimizing cache line invalidation frequency.