Adaptive Cache Replacement in Efficiently Querying Semantic Big Data

Abstract

In this paper, we address the problem of accelerating the query answering in semantic big data, such as Linked Data. Most importantly, performance factor is the cache replacement policy to scale the semantic big data. Due to the limited space in a cache, the less frequently accessed data should be removed to allow more space to the hot triples (frequently accessed). Existing cache approaches fail to perform dynamically on the triple stores as the fundamental structure of the triple stores is different from the relational data management system (RDBMS) store. The performance bottleneck of the triple store makes its real-world application cumbersome. To achieve a closer performance similar to RDBMS, we propose an Adaptive Cache Replacement (ACR) a policy that predicts the hot triples (frequently accessed) from the query log. Our proposed algorithm effectively replaces the cache with the frequently accessed triples. We apply the exponential smoothing forecasting method to collect most frequently accessed triples. The result of our evaluation shows that the proposed scheme outperforms the existing caching replacement schemes, such as LRU (least recently used) and LFU (least frequently used), in term of higher hit rates and less time overhead.