An Adaptive Hybrid Storage Mechanism for Specialization-Fusion Dynamic Balance: Architecture Design and Simulation Validation

Database systems face a fundamental tension between specialization and fusion: specialized engines deliver extreme performance for specific workloads but create data silos, while multi-model databases unify data management yet often compromise specialized performance. Building on our prior work submitted to WCSE 2026 [1], this paper presents the architecture design and simulation validation of an adaptive hybrid storage mechanism for achieving specialization-fusion dynamic balance. The mechanism comprises three core components: (1) Quantitative Evaluation: four metrics—Write Throughput Index (\({\mathrm{WTI}}\)), Compression Efficiency Ratio (\({\mathrm{CER}}\)), Cross-Model Query Latency (\({\mathrm{CMQL}}\)), and Balance Coefficient (\(\beta\))—to systematically measure a system's position on the specialization-fusion spectrum; (2) Comparative Validation: real-world experiments with the “KingbaseES mode” and the “CrateDB mode” reveal workload-dependent optimal balance; justifying the need for adaptivity; (3) Adaptive Realization: the AdaptStore hybrid storage engine dynamically adjusts storage formats via fine-grained transitions, learning-based optimization, and predictive adaptation. We implement a simulation prototype of AdaptStore and validate its effectiveness through extended simulation experiments (including comparisons with TimescaleDB and MongoDB) and parameter sensitivity analysis. Simulation results show that AdaptStore achieves write throughput up to 96% of a specialized database (with a gap of less than 5% for the hottest 20% of data) and reduces cross-model query latency by 30–50% compared to static hybrid systems, providing a reproducible design reference for adaptive database systems.