Анализ трассировок кеша и выбор политики вытеснения

data

sim-analyze is a Claude Code skill that reads a cache trace file, computes access-pattern statistics, and recommends the cache eviction policies most likely to perform well for that workload. It auto-detects the trace format, counts total accesses and distinct keys, then characterizes frequency distribution (Zipfian, uniform, or bimodal), recency patterns, scan bursts, and temporal shifts in the popular set. Based on this profile it classifies the workload — frequency-biased (LFU/TinyLFU), recency-biased (LRU/LIRS), mixed (W-TinyLFU, ARC), or scan-heavy (S3-FIFO, 2Q) — and runs a quick validation via `simulator:run` against Caffeine, Clairvoyant, and LRU at a suggested cache size. The final report covers a trace summary, policy recommendations with reasoning, validation results, and a ready-to-use `/sim-compare` invocation for deeper multi-size analysis. It is aimed at engineers doing cache performance tuning and eviction algorithm selection.