Optimizing cloud economics with linear elastic caching

Linear Elastic Caching addresses the problem of minimizing total cache cost by accounting for both cache misses and memory footprint over time. The approach connects this problem to ski rental algorithms and proposes a practical algorithm combining online caching with ski rental policies, including a machine learning-based variant optimized for production workloads. Evaluations show this elastic caching approach significantly reduces total cache cost compared to traditional fixed-size cache policies in Google Spanner and other production systems.

Source: Google Research | 2026-06-25