L1 L2 Cache Eviction Policy Based on Read Write Access Patterns

The patent applies local quality by differentiating data based on access patterns (read-intensive vs. write-intensive) and directing different types of data to different cache locations. The system assigns a characteristic value to each data element that reflects its access pattern, and uses this to determine optimal placement in either L1 or L2 cache, thereby optimizing access time for each data type while utilizing available cache capacity.

The patent inverts the conventional approach by not moving all evicted data to L2 cache, but rather selectively keeping certain data in L1 cache based on access patterns. Instead of the traditional policy of moving all data out of L1 to make room, the system identifies read-intensive data and preserves it in L1 despite eviction pressure, thereby maintaining high access speeds for frequently read data while still utilizing L2 cache for write-intensive data.

The patent implements self-service by enabling data elements to effectively 'choose' their own cache location based on their access patterns. Each data element is assigned a characteristic value that reflects whether it is read-intensive or write-intensive, and the system automatically directs read-intensive data to L1 cache while directing write-intensive data to L2 cache, thereby optimizing performance for each data type without manual intervention.

The patent applies parameter changes by introducing a characteristic value parameter that changes based on data access patterns. This parameter is used to dynamically adjust the cache eviction decision, transforming the cache management policy from a static, size-based approach to a dynamic, pattern-based approach. The characteristic value serves as an additional dimension for making eviction decisions, enabling optimization for different access patterns without significantly increasing system complexity.