5 Million Profiles Reads/Second - Caching at LinkedIn Scale

LinkedIn scaled their profile service to handle 5 million reads per second by implementing an integrated caching architecture using Couchbase, achieving a 99% cache hit rate. The solution combines Couchbase cache with Espresso (their NoSQL database) and Brooklin (data streaming platform) to handle both read and write paths efficiently. Key design decisions include resilience mechanisms for cache failures, high availability through data center replication, and data consistency using System Change Numbers (SCN) with Compare-And-Swap operations. This resulted in 90% reduction in storage nodes, 60% lower tail latency, and 10% cost savings.