An Adaptive Multi-level Caching Strategy For Disributed Database Systems

In distributed database system, data has been stored in different data nodes. The loadof each node may be unbalanced, because of the accessing frequency of each data nodecan be different and variable. So the overall performance is poor. Therefore, efficient anddynamic organization and management of cache to improve the overall performance of adistributed database is a hot research issues currently.Adaptive multi-level caching mechanism uses real-time dynamic cache resourceallocation techniques in distributed database systems to improve query response speedwhile solve the bottlenecks problem caused by hot data node. A two-tier system ofconcurrent execution strategy has been proposed based on access of data nodes and SQLdata tables, SQLs have been divided into multiple queues that can be executedconcurrently to improve SQL concurrent execution speed; A hierarchical cache partitionmanagement strategy has been proposed to improve the speed of cache matching bydividing the cache area into multi levels; Taking the virtual cache technology into cachemanagement, the node with poor performance in distributed database can get more cacheresource by adjusting the mapping between physical and virtual cache dynamically, andsolve the bottlenecks problem caused by hot data node; an adaptive semantic cachereplacement algorithm RFS (Replacement For Semantic) has been proposed to insteadthe conventional LRU (Least Recently Used) methods, it takes query execution time,history hit time interval, type and interval of hitting and the size of the cache items as analternative measure of weight, and improve the cache hit rate.The analysis of experimental results showed that: in a distributed database systemenvironment, the overall performance of adaptive multi-level caching mechanismimproved by15%or more compared to the single-stage static caching mechanism, andimproved by about30%compared to non-caching mechanism. In addition, theperformance of RFS improved by about30%compared to traditional LRU cachereplacement algorithm.