Optimizing Cloud Economics With Linear Elastic Caching
Google, Thursday, June 25th, 2026
Google Research's linear elastic caching cuts cloud costs by sizing caches dynamically to real-time workloads.
Google researchers introduced linear elastic caching, a dynamic approach that minimizes cloud infrastructure costs by adjusting cache size based on real-time workloads.
Rather than fixed-size caches, the system frames memory allocation as a 'ski rental problem,' using lightweight machine learning to balance memory expense against cache miss penalties.
Testing on production Spanner workloads showed a 5% reduction in total cost of ownership and a 15.5% decrease in memory usage.
The results show elastic strategies significantly outperform traditional peak-load provisioning.