Reinforcement Learning Approach for Intelligent Page Replacement

Abstract

The research focuses on the concept of reinforcement learning approach for intelligent page replacement. The improvement in “Reinforcement learning (RL)” can lead to enhanced intelligent page replacement with the help of enabling systems to improve learning of optimal page eviction policies by experience instead of relying on static algorithms such as FIFO or LRU. The objectives of the research include, to develop a reinforcement learning agent for enhanced page selection, comparison of performance with Optimal Page Replacement benchmark algorithm, adaptability under the varying workload conditions along with long-term learning efficiency in lowering page faults. “Explanatory research design”, “Positivism research philosophy” and “Deductive research approach” has been discussed. This study uses the secondary data approach to analyse reinforcement learning based page replacement strategies. Through thematic analysis based on peer-reviewed literature, key dimensions such as adaptability, performance optimisation, computational complexity and implementation feasibility were reviewed. The research concludes that reinforcement learning models can be used for improving page hit ratios and reducing page faults by dynamically learning from memory accessibility patterns. However, there are still significant challenges related to issues such as scalability, state-space representation and runtime overhead. Overall, reinforcement learning shows a lot of promise as an adaptive alternative to the traditional heuristic based page replacement techniques used in modern operating systems.