Stop Paying for Cold Data: 80% Cost Reduction with On-Demand Hot–Cold Data Loading in Milvus Tiered Storage

Introduces and details Milvus 2.6's Tiered Storage architecture, which enables on-demand loading of hot and cold data. Explains the inefficiencies of the previous full-load model and how Tiered Storage addresses them through lazy loading, partial loading (tenant-aware vector index loading, column-level scalar field loading), and LRU cache eviction. Provides performance and cost-saving evaluations.