An adaptive multi-level caching strategy for Distributed Database System

Abstract

The Internet of Things (IoT) has become a promising technology for addressing societal and industrial challenges by connecting geo-distributed and diverse devices to create smart systems worldwide. Nowadays, Distributed Database Systems (DDBS) are widely used in Internet of Things (IoT) frameworks to improve performance of massive data access. However, most recent caching techniques focus on collaborative caching and neglect the unbalanced workloads on hot data nodes, which may potentially become a bottleneck of latency-sensitive IoT services. To tackle this issue, we propose an adaptive multi-level caching strategy to solve the performance bottleneck caused by hot data access and therefore improve data access performance in DDBS. The essential idea of our strategy is to dynamically adjust cache resource allocation and cache size in different data nodes according to realtime access rate. In this way, more resources could be allocated to hot data node to speed up queries and eliminate the bottleneck. In addition, instead of the traditional Least Recently Used (LRU) algorithm, an efficient cache replacement algorithm is also proposed and implemented in our DDBS. The testbed experimental results indicate that the performance of DDBS with our adaptive multi-level caching strategy can be significantly improved by 20% compared with traditional strategies.