Amazon ElastiCache provides a managed key-value cache for reducing inference-server load when some queries are much more popular than others. It supports Redis and Memcached and is suited to runtime inference rather than batch inference. section 15.5.2
A practical caching setup should:
- Use a simple key-value cache with a low time to live, choosing the exact duration through data analysis. section 15.5.2
- Let batch inference update the cache when a key changes, rather than using the cache as part of batch inference itself. section 15.5.2
- Monitor CPU and RAM usage, request volume, and active instances while ensuring the serving system is accessible only to the application. section 15.5.2

